\ No newline at end of file
diff --git a/objects.inv b/objects.inv
index 836b896d1..6dcc88f9d 100644
Binary files a/objects.inv and b/objects.inv differ
diff --git a/search/search_index.json b/search/search_index.json
index 268949d28..31167e36c 100644
--- a/search/search_index.json
+++ b/search/search_index.json
@@ -1 +1 @@
-{"config":{"lang":["en"],"separator":"[\\s\\-]+","pipeline":["stopWordFilter"]},"docs":[{"location":"","title":"Home","text":"
Ingestion and analysis of genetic and functional genomic data for the identification and prioritisation of drug targets.
This project is still in experimental phase. Please refer to the roadmap section for more information.
For information on how to configure the development environment, run the code, or contribute changes, see the contributing section. For known technical issues and solutions to them, see the troubleshooting section.
"},{"location":"contributing/","title":"Environment configuration and contributing changes","text":""},{"location":"contributing/#one-time-configuration","title":"One-time configuration","text":"
The steps in this section only ever need to be done once on any particular system.
Google Cloud configuration: 1. Install Google Cloud SDK: https://cloud.google.com/sdk/docs/install. 1. Log in to your work Google Account: run gcloud auth login and follow instructions. 1. Obtain Google application credentials: run gcloud auth application-default login and follow instructions.
Check that you have the make utility installed, and if not (which is unlikely), install it using your system package manager.
Run make setup-dev to install/update the necessary packages and activate the development environment. You need to do this every time you open a new shell.
It is recommended to use VS Code as an IDE for development.
"},{"location":"contributing/#how-to-run-the-code","title":"How to run the code","text":"
All pipelines in this repository are intended to be run in Google Dataproc. Running them locally is not currently supported.
In order to run the code:
Manually edit your local workflow/dag.yaml file and comment out the steps you do not want to run.
Manually edit your local pyproject.toml file and modify the version of the code.
This must be different from the version used by any other people working on the repository to avoid any deployment conflicts, so it's a good idea to use your name, for example: 1.2.3+jdoe.
You can also add a brief branch description, for example: 1.2.3+jdoe.myfeature.
Note that the version must comply with PEP440 conventions, otherwise Poetry will not allow it to be deployed.
Do not use underscores or hyphens in your version name. When building the WHL file, they will be automatically converted to dots, which means the file name will no longer match the version and the build will fail. Use dots instead.
Run make build.
This will create a bundle containing the neccessary code, configuration and dependencies to run the ETL pipeline, and then upload this bundle to Google Cloud.
A version specific subpath is used, so uploading the code will not affect any branches but your own.
If there was already a code bundle uploaded with the same version number, it will be replaced.
Submit the Dataproc job with poetry run python workflow/workflow_template.py
You will need to specify additional parameters, some are mandatory and some are optional. Run with --help to see usage.
The script will provision the cluster and submit the job.
The cluster will take a few minutes to get provisioned and running, during which the script will not output anything, this is normal.
Once submitted, you can monitor the progress of your job on this page: https://console.cloud.google.com/dataproc/jobs?project=open-targets-genetics-dev.
On completion (whether successful or a failure), the cluster will be automatically removed, so you don't have to worry about shutting it down to avoid incurring charges.
"},{"location":"contributing/#how-to-generate-a-local-copy-of-the-documentation","title":"How to generate a local copy of the documentation","text":"
Run poetry run mkdocs serve. This will generate the local copy of the documentation and will start a local server to browse it (URL will be printed, usually http://127.0.0.1:8000/).
"},{"location":"contributing/#how-to-run-the-tests","title":"How to run the tests","text":"
When making changes, and especially when implementing a new module or feature, it's essential to ensure that all relevant sections of the code base are modified.
If during development you had a question which wasn't covered in the documentation, and someone explained it to you, add it to the documentation. The same applies if you encountered any instructions in the documentation which were obsolete or incorrect.
Documentation autogeneration expressions start with :::. They will automatically generate sections of the documentation based on class and method docstrings. Be sure to update them for:
Dataset definitions in docs/reference/dataset (example: docs/reference/dataset/study_index/study_index_finngen.md)
Step definitions in docs/reference/step (example: docs/reference/step/finngen.md)
If you see errors related to BLAS/LAPACK libraries, see this StackOverflow post for guidance.
"},{"location":"troubleshooting/#pyenv-and-poetry","title":"Pyenv and Poetry","text":"
If you see various errors thrown by Pyenv or Poetry, they can be hard to specifically diagnose and resolve. In this case, it often helps to remove those tools from the system completely. Follow these steps:
Close your currently activated environment, if any: exit
Officially, PySpark requires Java version 8 (a.k.a. 1.8) or above to work. However, if you have a very recent version of Java, you may experience issues, as it may introduce breaking changes that PySpark hasn't had time to integrate. For example, as of May 2023, PySpark did not work with Java 20.
If you are encountering problems with initialising a Spark session, try using Java 11.
If you see an error message thrown by pre-commit, which looks like this (SyntaxError: Unexpected token '?'), followed by a JavaScript traceback, the issue is likely with your system NodeJS version.
One solution which can help in this case is to upgrade your system NodeJS version. However, this may not always be possible. For example, Ubuntu repository is several major versions behind the latest version as of July 2023.
Another solution which helps is to remove Node, NodeJS, and npm from your system entirely. In this case, pre-commit will not try to rely on a system version of NodeJS and will install its own, suitable one.
On Ubuntu, this can be done using sudo apt remove node nodejs npm, followed by sudo apt autoremove. But in some cases, depending on your existing installation, you may need to also manually remove some files. See this StackOverflow answer for guidance.
After running these commands, you are advised to open a fresh shell, and then also reinstall Pyenv and Poetry to make sure they pick up the changes (see relevant section above).
Dataset is a wrapper around a Spark DataFrame with a predefined schema. Schemas for each child dataset are described in the json.schemas module.
Source code in src/otg/dataset/dataset.py
@dataclass\nclass Dataset:\n\"\"\"Open Targets Genetics Dataset.\n\n `Dataset` is a wrapper around a Spark DataFrame with a predefined schema. Schemas for each child dataset are described in the `json.schemas` module.\n \"\"\"\n\n _df: DataFrame\n _schema: StructType\n\n def __post_init__(self: Dataset) -> None:\n\"\"\"Post init.\"\"\"\n self.validate_schema()\n\n @property\n def df(self: Dataset) -> DataFrame:\n\"\"\"Dataframe included in the Dataset.\"\"\"\n return self._df\n\n @df.setter\n def df(self: Dataset, new_df: DataFrame) -> None: # noqa: CCE001\n\"\"\"Dataframe setter.\"\"\"\n self._df = new_df\n self.validate_schema()\n\n @property\n def schema(self: Dataset) -> StructType:\n\"\"\"Dataframe expected schema.\"\"\"\n return self._schema\n\n @classmethod\n def from_parquet(\n cls: type[Dataset], session: Session, path: str, schema: StructType\n ) -> Dataset:\n\"\"\"Reads a parquet file into a Dataset with a given schema.\n\n Args:\n session (Session): ETL session\n path (str): Path to parquet file\n schema (StructType): Schema to use\n\n Returns:\n Dataset: Dataset with given schema\n \"\"\"\n df = session.read_parquet(path=path, schema=schema)\n return cls(_df=df, _schema=schema)\n\n def validate_schema(self: Dataset) -> None: # sourcery skip: invert-any-all\n\"\"\"Validate DataFrame schema against expected class schema.\n\n Raises:\n ValueError: DataFrame schema is not valid\n \"\"\"\n expected_schema = self._schema\n expected_fields = flatten_schema(expected_schema)\n observed_schema = self._df.schema\n observed_fields = flatten_schema(observed_schema)\n\n # Unexpected fields in dataset\n if unexpected_struct_fields := [\n x for x in observed_fields if x not in expected_fields\n ]:\n raise ValueError(\n f\"The {unexpected_struct_fields} fields are not included in DataFrame schema: {expected_fields}\"\n )\n\n # Required fields not in dataset\n required_fields = [x.name for x in expected_schema if not x.nullable]\n if missing_required_fields := [\n req\n for req in required_fields\n if not any(field.name == req for field in observed_fields)\n ]:\n raise ValueError(\n f\"The {missing_required_fields} fields are required but missing: {required_fields}\"\n )\n\n # Fields with duplicated names\n if duplicated_fields := [\n x for x in set(observed_fields) if observed_fields.count(x) > 1\n ]:\n raise ValueError(\n f\"The following fields are duplicated in DataFrame schema: {duplicated_fields}\"\n )\n\n # Fields with different datatype\n if fields_with_different_observed_datatype := [\n field\n for field in set(observed_fields)\n if observed_fields.count(field) != expected_fields.count(field)\n ]:\n raise ValueError(\n f\"The following fields present differences in their datatypes: {fields_with_different_observed_datatype}.\"\n )\n
Validate DataFrame schema against expected class schema.
Raises:
Type Description ValueError
DataFrame schema is not valid
Source code in src/otg/dataset/dataset.py
def validate_schema(self: Dataset) -> None: # sourcery skip: invert-any-all\n\"\"\"Validate DataFrame schema against expected class schema.\n\n Raises:\n ValueError: DataFrame schema is not valid\n \"\"\"\n expected_schema = self._schema\n expected_fields = flatten_schema(expected_schema)\n observed_schema = self._df.schema\n observed_fields = flatten_schema(observed_schema)\n\n # Unexpected fields in dataset\n if unexpected_struct_fields := [\n x for x in observed_fields if x not in expected_fields\n ]:\n raise ValueError(\n f\"The {unexpected_struct_fields} fields are not included in DataFrame schema: {expected_fields}\"\n )\n\n # Required fields not in dataset\n required_fields = [x.name for x in expected_schema if not x.nullable]\n if missing_required_fields := [\n req\n for req in required_fields\n if not any(field.name == req for field in observed_fields)\n ]:\n raise ValueError(\n f\"The {missing_required_fields} fields are required but missing: {required_fields}\"\n )\n\n # Fields with duplicated names\n if duplicated_fields := [\n x for x in set(observed_fields) if observed_fields.count(x) > 1\n ]:\n raise ValueError(\n f\"The following fields are duplicated in DataFrame schema: {duplicated_fields}\"\n )\n\n # Fields with different datatype\n if fields_with_different_observed_datatype := [\n field\n for field in set(observed_fields)\n if observed_fields.count(field) != expected_fields.count(field)\n ]:\n raise ValueError(\n f\"The following fields present differences in their datatypes: {fields_with_different_observed_datatype}.\"\n )\n
Convert intervals into V2G by intersecting with a variant index.
Parameters:
Name Type Description Default variant_indexVariantIndex
Variant index dataset
required
Returns:
Name Type Description V2GV2G
Variant-to-gene evidence dataset
Source code in src/otg/dataset/intervals.py
def v2g(self: Intervals, variant_index: VariantIndex) -> V2G:\n\"\"\"Convert intervals into V2G by intersecting with a variant index.\n\n Args:\n variant_index (VariantIndex): Variant index dataset\n\n Returns:\n V2G: Variant-to-gene evidence dataset\n \"\"\"\n return V2G(\n _df=(\n # TODO: We can include the start and end position as part of the `on` clause in the join\n self.df.alias(\"interval\")\n .join(\n variant_index.df.selectExpr(\n \"chromosome as vi_chromosome\", \"variantId\", \"position\"\n ).alias(\"vi\"),\n on=[\n f.col(\"vi.vi_chromosome\") == f.col(\"interval.chromosome\"),\n f.col(\"vi.position\").between(\n f.col(\"interval.start\"), f.col(\"interval.end\")\n ),\n ],\n how=\"inner\",\n )\n .drop(\"start\", \"end\", \"vi_chromosome\")\n )\n )\n
Annotate LD index with indices starting and stopping at a given interval.
Parameters:
Name Type Description Default ld_radiusint
radius around each position
required
Returns:
Name Type Description LDIndexLDIndex
including start_idx and stop_idx columns
Source code in src/otg/dataset/ld_index.py
def annotate_index_intervals(self: LDIndex, ld_radius: int) -> LDIndex:\n\"\"\"Annotate LD index with indices starting and stopping at a given interval.\n\n Args:\n ld_radius (int): radius around each position\n\n Returns:\n LDIndex: including `start_idx` and `stop_idx` columns\n \"\"\"\n index_with_positions = (\n self._df.drop(\"start_idx\", \"stop_idx\")\n .select(\n \"*\",\n LDIndex._interval_start(\n contig=f.col(\"chromosome\"),\n position=f.col(\"position\"),\n ld_radius=ld_radius,\n ).alias(\"start_pos\"),\n LDIndex._interval_stop(\n contig=f.col(\"chromosome\"),\n position=f.col(\"position\"),\n ld_radius=ld_radius,\n ).alias(\"stop_pos\"),\n )\n .persist()\n )\n\n self.df = (\n index_with_positions.join(\n (\n index_with_positions\n # Given the multiple variants with the same chromosome/position can have different indices, filter for the lowest index:\n .transform(\n lambda df: get_record_with_minimum_value(\n df, [\"chromosome\", \"position\"], \"idx\"\n )\n ).select(\n \"chromosome\",\n f.col(\"position\").alias(\"start_pos\"),\n f.col(\"idx\").alias(\"start_idx\"),\n )\n ),\n on=[\"chromosome\", \"start_pos\"],\n )\n .join(\n (\n index_with_positions\n # Given the multiple variants with the same chromosome/position can have different indices, filter for the highest index:\n .transform(\n lambda df: get_record_with_maximum_value(\n df, [\"chromosome\", \"position\"], \"idx\"\n )\n ).select(\n \"chromosome\",\n f.col(\"position\").alias(\"stop_pos\"),\n f.col(\"idx\").alias(\"stop_idx\"),\n )\n ),\n on=[\"chromosome\", \"stop_pos\"],\n )\n # Filter out variants for which start idx > stop idx due to liftover\n .filter(f.col(\"start_idx\") < f.col(\"stop_idx\"))\n .drop(\"start_pos\", \"stop_pos\")\n )\n\n return self\n
Filter summary statistics based on the provided p-value threshold.
Parameters:
Name Type Description Default pvaluefloat
upper limit of the p-value to be filtered upon.
required
Returns:
Name Type Description SummaryStatisticsSummaryStatistics
summary statistics object containing single point associations with p-values at least as significant as the provided threshold.
Source code in src/otg/dataset/summary_statistics.py
def pvalue_filter(self: SummaryStatistics, pvalue: float) -> SummaryStatistics:\n\"\"\"Filter summary statistics based on the provided p-value threshold.\n\n Args:\n pvalue (float): upper limit of the p-value to be filtered upon.\n\n Returns:\n SummaryStatistics: summary statistics object containing single point associations with p-values at least as significant as the provided threshold.\n \"\"\"\n # Converting p-value to mantissa and exponent:\n (mantissa, exponent) = split_pvalue(pvalue)\n\n # Applying filter:\n df = self._df.filter(\n (f.col(\"pValueExponent\") < exponent)\n | (\n (f.col(\"pValueExponent\") == exponent)\n & (f.col(\"pValueMantissa\") <= mantissa)\n )\n )\n return SummaryStatistics(_df=df)\n
Dataset with variant-level annotations derived from GnomAD.
Source code in src/otg/dataset/variant_annotation.py
@dataclass\nclass VariantAnnotation(Dataset):\n\"\"\"Dataset with variant-level annotations derived from GnomAD.\"\"\"\n\n _schema: StructType = parse_spark_schema(\"variant_annotation.json\")\n\n @classmethod\n def from_parquet(\n cls: type[VariantAnnotation], session: Session, path: str\n ) -> VariantAnnotation:\n\"\"\"Initialise VariantAnnotation from parquet file.\n\n Args:\n session (Session): ETL session\n path (str): Path to parquet file\n\n Returns:\n VariantAnnotation: VariantAnnotation dataset\n \"\"\"\n df = session.read_parquet(path=path, schema=cls._schema)\n return cls(_df=df, _schema=cls._schema)\n\n @classmethod\n def from_gnomad(\n cls: type[VariantAnnotation],\n gnomad_file: str,\n grch38_to_grch37_chain: str,\n populations: list,\n ) -> VariantAnnotation:\n\"\"\"Generate variant annotation dataset from gnomAD.\n\n Some relevant modifications to the original dataset are:\n\n 1. The transcript consequences features provided by VEP are filtered to only refer to the Ensembl canonical transcript.\n 2. Genome coordinates are liftovered from GRCh38 to GRCh37 to keep as annotation.\n 3. Field names are converted to camel case to follow the convention.\n\n Args:\n gnomad_file (str): Path to `gnomad.genomes.vX.X.X.sites.ht` gnomAD dataset\n grch38_to_grch37_chain (str): Path to chain file for liftover\n populations (list): List of populations to include in the dataset\n\n Returns:\n VariantAnnotation: Variant annotation dataset\n \"\"\"\n # Load variants dataset\n ht = hl.read_table(\n gnomad_file,\n _load_refs=False,\n )\n\n # Liftover\n grch37 = hl.get_reference(\"GRCh37\")\n grch38 = hl.get_reference(\"GRCh38\")\n grch38.add_liftover(grch38_to_grch37_chain, grch37)\n\n # Drop non biallelic variants\n ht = ht.filter(ht.alleles.length() == 2)\n # Liftover\n ht = ht.annotate(locus_GRCh37=hl.liftover(ht.locus, \"GRCh37\"))\n # Select relevant fields and nested records to create class\n return cls(\n _df=(\n ht.select(\n gnomad3VariantId=hl.str(\"-\").join(\n [\n ht.locus.contig.replace(\"chr\", \"\"),\n hl.str(ht.locus.position),\n ht.alleles[0],\n ht.alleles[1],\n ]\n ),\n chromosome=ht.locus.contig.replace(\"chr\", \"\"),\n position=convert_gnomad_position_to_ensembl_hail(\n ht.locus.position, ht.alleles[0], ht.alleles[1]\n ),\n variantId=hl.str(\"_\").join(\n [\n ht.locus.contig.replace(\"chr\", \"\"),\n hl.str(\n convert_gnomad_position_to_ensembl_hail(\n ht.locus.position, ht.alleles[0], ht.alleles[1]\n )\n ),\n ht.alleles[0],\n ht.alleles[1],\n ]\n ),\n chromosomeB37=ht.locus_GRCh37.contig.replace(\"chr\", \"\"),\n positionB37=ht.locus_GRCh37.position,\n referenceAllele=ht.alleles[0],\n alternateAllele=ht.alleles[1],\n rsIds=ht.rsid,\n alleleType=ht.allele_info.allele_type,\n cadd=hl.struct(\n phred=ht.cadd.phred,\n raw=ht.cadd.raw_score,\n ),\n alleleFrequencies=hl.set([f\"{pop}-adj\" for pop in populations]).map(\n lambda p: hl.struct(\n populationName=p,\n alleleFrequency=ht.freq[ht.globals.freq_index_dict[p]].AF,\n )\n ),\n vep=hl.struct(\n mostSevereConsequence=ht.vep.most_severe_consequence,\n transcriptConsequences=hl.map(\n lambda x: hl.struct(\n aminoAcids=x.amino_acids,\n consequenceTerms=x.consequence_terms,\n geneId=x.gene_id,\n lof=x.lof,\n polyphenScore=x.polyphen_score,\n polyphenPrediction=x.polyphen_prediction,\n siftScore=x.sift_score,\n siftPrediction=x.sift_prediction,\n ),\n # Only keeping canonical transcripts\n ht.vep.transcript_consequences.filter(\n lambda x: (x.canonical == 1)\n & (x.gene_symbol_source == \"HGNC\")\n ),\n ),\n ),\n )\n .key_by(\"chromosome\", \"position\")\n .drop(\"locus\", \"alleles\")\n .select_globals()\n .to_spark(flatten=False)\n )\n )\n\n def persist(self: VariantAnnotation) -> VariantAnnotation:\n\"\"\"Persist DataFrame included in the Dataset.\"\"\"\n self.df = self._df.persist()\n return self\n\n def max_maf(self: VariantAnnotation) -> Column:\n\"\"\"Maximum minor allele frequency accross all populations.\n\n Returns:\n Column: Maximum minor allele frequency accross all populations.\n \"\"\"\n return f.array_max(\n f.transform(\n self.df.alleleFrequencies,\n lambda af: f.when(\n af.alleleFrequency > 0.5, 1 - af.alleleFrequency\n ).otherwise(af.alleleFrequency),\n )\n )\n\n def filter_by_variant_df(\n self: VariantAnnotation, df: DataFrame, cols: list[str]\n ) -> VariantAnnotation:\n\"\"\"Filter variant annotation dataset by a variant dataframe.\n\n Args:\n df (DataFrame): A dataframe of variants\n cols (List[str]): A list of columns to join on\n\n Returns:\n VariantAnnotation: A filtered variant annotation dataset\n \"\"\"\n self.df = self._df.join(f.broadcast(df.select(cols)), on=cols, how=\"inner\")\n return self\n\n def get_transcript_consequence_df(\n self: VariantAnnotation, filter_by: Optional[GeneIndex] = None\n ) -> DataFrame:\n\"\"\"Dataframe of exploded transcript consequences.\n\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index. Defaults to None.\n\n Returns:\n DataFrame: A dataframe exploded by transcript consequences with the columns variantId, chromosome, transcriptConsequence\n \"\"\"\n # exploding the array removes records without VEP annotation\n transript_consequences = self.df.withColumn(\n \"transcriptConsequence\", f.explode(\"vep.transcriptConsequences\")\n ).select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"transcriptConsequence\",\n f.col(\"transcriptConsequence.geneId\").alias(\"geneId\"),\n )\n if filter_by:\n transript_consequences = transript_consequences.join(\n f.broadcast(filter_by.df),\n on=[\"chromosome\", \"geneId\"],\n )\n return transript_consequences.persist()\n\n def get_most_severe_vep_v2g(\n self: VariantAnnotation,\n vep_consequences: DataFrame,\n filter_by: GeneIndex,\n ) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments based on VEP's predicted consequence on the transcript.\n\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n vep_consequences (DataFrame): A dataframe of VEP consequences\n filter_by (GeneIndex): A gene index to filter by. Defaults to None.\n\n Returns:\n V2G: High and medium severity variant to gene assignments\n \"\"\"\n vep_lut = vep_consequences.select(\n f.element_at(f.split(\"Accession\", r\"/\"), -1).alias(\n \"variantFunctionalConsequenceId\"\n ),\n f.col(\"Term\").alias(\"label\"),\n f.col(\"v2g_score\").cast(\"double\").alias(\"score\"),\n )\n\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n f.col(\"transcriptConsequence.geneId\").alias(\"geneId\"),\n f.explode(\"transcriptConsequence.consequenceTerms\").alias(\"label\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"variantConsequence\").alias(\"datasourceId\"),\n )\n # A variant can have multiple predicted consequences on a transcript, the most severe one is selected\n .join(\n f.broadcast(vep_lut),\n on=\"label\",\n how=\"inner\",\n )\n .filter(f.col(\"score\") != 0)\n .transform(\n lambda df: get_record_with_maximum_value(\n df, [\"variantId\", \"geneId\"], \"score\"\n )\n )\n )\n\n def get_polyphen_v2g(\n self: VariantAnnotation, filter_by: Optional[GeneIndex] = None\n ) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments with a PolyPhen's predicted score on the transcript.\n\n Polyphen informs about the probability that a substitution is damaging. Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by. Defaults to None.\n\n Returns:\n V2G: variant to gene assignments with their polyphen scores\n \"\"\"\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .filter(f.col(\"transcriptConsequence.polyphenScore\").isNotNull())\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"geneId\",\n f.col(\"transcriptConsequence.polyphenScore\").alias(\"score\"),\n f.col(\"transcriptConsequence.polyphenPrediction\").alias(\"label\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"polyphen\").alias(\"datasourceId\"),\n )\n )\n\n def get_sift_v2g(self: VariantAnnotation, filter_by: GeneIndex) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments with a SIFT's predicted score on the transcript.\n\n SIFT informs about the probability that a substitution is tolerated so scores nearer zero are more likely to be deleterious.\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by.\n\n Returns:\n V2G: variant to gene assignments with their SIFT scores\n \"\"\"\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .filter(f.col(\"transcriptConsequence.siftScore\").isNotNull())\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"geneId\",\n f.expr(\"1 - transcriptConsequence.siftScore\").alias(\"score\"),\n f.col(\"transcriptConsequence.siftPrediction\").alias(\"label\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"sift\").alias(\"datasourceId\"),\n )\n )\n\n def get_plof_v2g(self: VariantAnnotation, filter_by: GeneIndex) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments with a flag indicating if the variant is predicted to be a loss-of-function variant by the LOFTEE algorithm.\n\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by.\n\n Returns:\n V2G: variant to gene assignments from the LOFTEE algorithm\n \"\"\"\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .filter(f.col(\"transcriptConsequence.lof\").isNotNull())\n .withColumn(\n \"isHighQualityPlof\",\n f.when(f.col(\"transcriptConsequence.lof\") == \"HC\", True).when(\n f.col(\"transcriptConsequence.lof\") == \"LC\", False\n ),\n )\n .withColumn(\n \"score\",\n f.when(f.col(\"isHighQualityPlof\"), 1.0).when(\n ~f.col(\"isHighQualityPlof\"), 0\n ),\n )\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"geneId\",\n \"isHighQualityPlof\",\n f.col(\"score\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"loftee\").alias(\"datasourceId\"),\n )\n )\n\n def get_distance_to_tss(\n self: VariantAnnotation,\n filter_by: GeneIndex,\n max_distance: int = 500_000,\n ) -> V2G:\n\"\"\"Extracts variant to gene assignments for variants falling within a window of a gene's TSS.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by.\n max_distance (int): The maximum distance from the TSS to consider. Defaults to 500_000.\n\n Returns:\n V2G: variant to gene assignments with their distance to the TSS\n \"\"\"\n return V2G(\n _df=self.df.alias(\"variant\")\n .join(\n f.broadcast(filter_by.locations_lut()).alias(\"gene\"),\n on=[\n f.col(\"variant.chromosome\") == f.col(\"gene.chromosome\"),\n f.abs(f.col(\"variant.position\") - f.col(\"gene.tss\"))\n <= max_distance,\n ],\n how=\"inner\",\n )\n .withColumn(\n \"inverse_distance\",\n max_distance - f.abs(f.col(\"variant.position\") - f.col(\"gene.tss\")),\n )\n .transform(lambda df: normalise_column(df, \"inverse_distance\", \"score\"))\n .select(\n \"variantId\",\n f.col(\"variant.chromosome\").alias(\"chromosome\"),\n \"position\",\n \"geneId\",\n \"score\",\n f.lit(\"distance\").alias(\"datatypeId\"),\n f.lit(\"canonical_tss\").alias(\"datasourceId\"),\n )\n )\n
Extracts variant to gene assignments for variants falling within a window of a gene's TSS.
Parameters:
Name Type Description Default filter_byGeneIndex
A gene index to filter by.
required max_distanceint
The maximum distance from the TSS to consider. Defaults to 500_000.
500000
Returns:
Name Type Description V2GV2G
variant to gene assignments with their distance to the TSS
Source code in src/otg/dataset/variant_annotation.py
def get_distance_to_tss(\n self: VariantAnnotation,\n filter_by: GeneIndex,\n max_distance: int = 500_000,\n) -> V2G:\n\"\"\"Extracts variant to gene assignments for variants falling within a window of a gene's TSS.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by.\n max_distance (int): The maximum distance from the TSS to consider. Defaults to 500_000.\n\n Returns:\n V2G: variant to gene assignments with their distance to the TSS\n \"\"\"\n return V2G(\n _df=self.df.alias(\"variant\")\n .join(\n f.broadcast(filter_by.locations_lut()).alias(\"gene\"),\n on=[\n f.col(\"variant.chromosome\") == f.col(\"gene.chromosome\"),\n f.abs(f.col(\"variant.position\") - f.col(\"gene.tss\"))\n <= max_distance,\n ],\n how=\"inner\",\n )\n .withColumn(\n \"inverse_distance\",\n max_distance - f.abs(f.col(\"variant.position\") - f.col(\"gene.tss\")),\n )\n .transform(lambda df: normalise_column(df, \"inverse_distance\", \"score\"))\n .select(\n \"variantId\",\n f.col(\"variant.chromosome\").alias(\"chromosome\"),\n \"position\",\n \"geneId\",\n \"score\",\n f.lit(\"distance\").alias(\"datatypeId\"),\n f.lit(\"canonical_tss\").alias(\"datasourceId\"),\n )\n )\n
Creates a dataset with variant to gene assignments based on VEP's predicted consequence on the transcript.
Optionally the trancript consequences can be reduced to the universe of a gene index.
Parameters:
Name Type Description Default vep_consequencesDataFrame
A dataframe of VEP consequences
required filter_byGeneIndex
A gene index to filter by. Defaults to None.
required
Returns:
Name Type Description V2GV2G
High and medium severity variant to gene assignments
Source code in src/otg/dataset/variant_annotation.py
def get_most_severe_vep_v2g(\n self: VariantAnnotation,\n vep_consequences: DataFrame,\n filter_by: GeneIndex,\n) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments based on VEP's predicted consequence on the transcript.\n\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n vep_consequences (DataFrame): A dataframe of VEP consequences\n filter_by (GeneIndex): A gene index to filter by. Defaults to None.\n\n Returns:\n V2G: High and medium severity variant to gene assignments\n \"\"\"\n vep_lut = vep_consequences.select(\n f.element_at(f.split(\"Accession\", r\"/\"), -1).alias(\n \"variantFunctionalConsequenceId\"\n ),\n f.col(\"Term\").alias(\"label\"),\n f.col(\"v2g_score\").cast(\"double\").alias(\"score\"),\n )\n\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n f.col(\"transcriptConsequence.geneId\").alias(\"geneId\"),\n f.explode(\"transcriptConsequence.consequenceTerms\").alias(\"label\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"variantConsequence\").alias(\"datasourceId\"),\n )\n # A variant can have multiple predicted consequences on a transcript, the most severe one is selected\n .join(\n f.broadcast(vep_lut),\n on=\"label\",\n how=\"inner\",\n )\n .filter(f.col(\"score\") != 0)\n .transform(\n lambda df: get_record_with_maximum_value(\n df, [\"variantId\", \"geneId\"], \"score\"\n )\n )\n )\n
Creates a dataset with variant to gene assignments with a flag indicating if the variant is predicted to be a loss-of-function variant by the LOFTEE algorithm.
Optionally the trancript consequences can be reduced to the universe of a gene index.
Parameters:
Name Type Description Default filter_byGeneIndex
A gene index to filter by.
required
Returns:
Name Type Description V2GV2G
variant to gene assignments from the LOFTEE algorithm
Source code in src/otg/dataset/variant_annotation.py
def get_plof_v2g(self: VariantAnnotation, filter_by: GeneIndex) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments with a flag indicating if the variant is predicted to be a loss-of-function variant by the LOFTEE algorithm.\n\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by.\n\n Returns:\n V2G: variant to gene assignments from the LOFTEE algorithm\n \"\"\"\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .filter(f.col(\"transcriptConsequence.lof\").isNotNull())\n .withColumn(\n \"isHighQualityPlof\",\n f.when(f.col(\"transcriptConsequence.lof\") == \"HC\", True).when(\n f.col(\"transcriptConsequence.lof\") == \"LC\", False\n ),\n )\n .withColumn(\n \"score\",\n f.when(f.col(\"isHighQualityPlof\"), 1.0).when(\n ~f.col(\"isHighQualityPlof\"), 0\n ),\n )\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"geneId\",\n \"isHighQualityPlof\",\n f.col(\"score\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"loftee\").alias(\"datasourceId\"),\n )\n )\n
Creates a dataset with variant to gene assignments with a PolyPhen's predicted score on the transcript.
Polyphen informs about the probability that a substitution is damaging. Optionally the trancript consequences can be reduced to the universe of a gene index.
Parameters:
Name Type Description Default filter_byGeneIndex
A gene index to filter by. Defaults to None.
None
Returns:
Name Type Description V2GV2G
variant to gene assignments with their polyphen scores
Source code in src/otg/dataset/variant_annotation.py
def get_polyphen_v2g(\n self: VariantAnnotation, filter_by: Optional[GeneIndex] = None\n) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments with a PolyPhen's predicted score on the transcript.\n\n Polyphen informs about the probability that a substitution is damaging. Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by. Defaults to None.\n\n Returns:\n V2G: variant to gene assignments with their polyphen scores\n \"\"\"\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .filter(f.col(\"transcriptConsequence.polyphenScore\").isNotNull())\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"geneId\",\n f.col(\"transcriptConsequence.polyphenScore\").alias(\"score\"),\n f.col(\"transcriptConsequence.polyphenPrediction\").alias(\"label\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"polyphen\").alias(\"datasourceId\"),\n )\n )\n
Creates a dataset with variant to gene assignments with a SIFT's predicted score on the transcript.
SIFT informs about the probability that a substitution is tolerated so scores nearer zero are more likely to be deleterious. Optionally the trancript consequences can be reduced to the universe of a gene index.
Parameters:
Name Type Description Default filter_byGeneIndex
A gene index to filter by.
required
Returns:
Name Type Description V2GV2G
variant to gene assignments with their SIFT scores
Source code in src/otg/dataset/variant_annotation.py
def get_sift_v2g(self: VariantAnnotation, filter_by: GeneIndex) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments with a SIFT's predicted score on the transcript.\n\n SIFT informs about the probability that a substitution is tolerated so scores nearer zero are more likely to be deleterious.\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by.\n\n Returns:\n V2G: variant to gene assignments with their SIFT scores\n \"\"\"\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .filter(f.col(\"transcriptConsequence.siftScore\").isNotNull())\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"geneId\",\n f.expr(\"1 - transcriptConsequence.siftScore\").alias(\"score\"),\n f.col(\"transcriptConsequence.siftPrediction\").alias(\"label\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"sift\").alias(\"datasourceId\"),\n )\n )\n
Optionally the trancript consequences can be reduced to the universe of a gene index.
Parameters:
Name Type Description Default filter_byGeneIndex
A gene index. Defaults to None.
None
Returns:
Name Type Description DataFrameDataFrame
A dataframe exploded by transcript consequences with the columns variantId, chromosome, transcriptConsequence
Source code in src/otg/dataset/variant_annotation.py
def get_transcript_consequence_df(\n self: VariantAnnotation, filter_by: Optional[GeneIndex] = None\n) -> DataFrame:\n\"\"\"Dataframe of exploded transcript consequences.\n\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index. Defaults to None.\n\n Returns:\n DataFrame: A dataframe exploded by transcript consequences with the columns variantId, chromosome, transcriptConsequence\n \"\"\"\n # exploding the array removes records without VEP annotation\n transript_consequences = self.df.withColumn(\n \"transcriptConsequence\", f.explode(\"vep.transcriptConsequences\")\n ).select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"transcriptConsequence\",\n f.col(\"transcriptConsequence.geneId\").alias(\"geneId\"),\n )\n if filter_by:\n transript_consequences = transript_consequences.join(\n f.broadcast(filter_by.df),\n on=[\"chromosome\", \"geneId\"],\n )\n return transript_consequences.persist()\n
Source code in src/otg/dataset/variant_annotation.py
def persist(self: VariantAnnotation) -> VariantAnnotation:\n\"\"\"Persist DataFrame included in the Dataset.\"\"\"\n self.df = self._df.persist()\n return self\n
def persist(self: VariantIndex) -> VariantIndex:\n\"\"\"Persist DataFrame included in the Dataset.\"\"\"\n self.df = self._df.persist()\n return self\n
"},{"location":"components/dataset/variant_to_gene/","title":"Variant to gene","text":"
Bases: Dataset
Variant-to-gene (V2G) evidence dataset.
A variant-to-gene (V2G) evidence is understood as any piece of evidence that supports the association of a variant with a likely causal gene. The evidence can sometimes be context-specific and refer to specific biofeatures (e.g. cell types)
Source code in src/otg/dataset/v2g.py
@dataclass\nclass V2G(Dataset):\n\"\"\"Variant-to-gene (V2G) evidence dataset.\n\n A variant-to-gene (V2G) evidence is understood as any piece of evidence that supports the association of a variant with a likely causal gene. The evidence can sometimes be context-specific and refer to specific `biofeatures` (e.g. cell types)\n \"\"\"\n\n _schema: StructType = parse_spark_schema(\"v2g.json\")\n\n @classmethod\n def from_parquet(cls: type[V2G], session: Session, path: str) -> V2G:\n\"\"\"Initialise V2G from parquet file.\n\n Args:\n session (Session): ETL session\n path (str): Path to parquet file\n\n Returns:\n V2G: V2G dataset\n \"\"\"\n df = session.read_parquet(path=path, schema=cls._schema)\n return cls(_df=df, _schema=cls._schema)\n\n def filter_by_genes(self: V2G, genes: GeneIndex) -> V2G:\n\"\"\"Filter by V2G dataset by genes.\n\n Args:\n genes (GeneIndex): Gene index dataset to filter by\n\n Returns:\n V2G: V2G dataset filtered by genes\n \"\"\"\n self.df = self._df.join(genes.df.select(\"geneId\"), on=\"geneId\", how=\"inner\")\n return self\n
A study index dataset captures all the metadata for all studies including GWAS and Molecular QTL.
Source code in src/otg/dataset/study_index.py
@dataclass\nclass StudyIndex(Dataset):\n\"\"\"Study index dataset.\n\n A study index dataset captures all the metadata for all studies including GWAS and Molecular QTL.\n \"\"\"\n\n _schema: StructType = parse_spark_schema(\"studies.json\")\n\n @classmethod\n def from_parquet(cls: type[StudyIndex], session: Session, path: str) -> StudyIndex:\n\"\"\"Initialise StudyIndex from parquet file.\n\n Args:\n session (Session): ETL session\n path (str): Path to parquet file\n\n Returns:\n StudyIndex: Study index dataset\n \"\"\"\n df = session.read_parquet(path=path, schema=cls._schema)\n return cls(_df=df, _schema=cls._schema)\n\n def study_type_lut(self: StudyIndex) -> DataFrame:\n\"\"\"Return a lookup table of study type.\n\n Returns:\n DataFrame: A dataframe containing `studyId` and `studyType` columns.\n \"\"\"\n return self.df.select(\"studyId\", \"studyType\")\n
A dataframe containing studyId and studyType columns.
Source code in src/otg/dataset/study_index.py
def study_type_lut(self: StudyIndex) -> DataFrame:\n\"\"\"Return a lookup table of study type.\n\n Returns:\n DataFrame: A dataframe containing `studyId` and `studyType` columns.\n \"\"\"\n return self.df.select(\"studyId\", \"studyType\")\n
"},{"location":"components/dataset/study_index/study_index_finngen/","title":"Study index finngen","text":"
Bases: StudyIndex
Study index dataset from FinnGen.
The following information is aggregated/extracted:
Study ID in the special format (FINNGEN_R9_*)
Trait name (for example, Amoebiasis)
Number of cases and controls
Link to the summary statistics location
Some fields are also populated as constants, such as study type and the initial sample size.
Source code in src/otg/dataset/study_index.py
@dataclass\nclass StudyIndexFinnGen(StudyIndex):\n\"\"\"Study index dataset from FinnGen.\n\n The following information is aggregated/extracted:\n\n - Study ID in the special format (FINNGEN_R9_*)\n - Trait name (for example, Amoebiasis)\n - Number of cases and controls\n - Link to the summary statistics location\n\n Some fields are also populated as constants, such as study type and the initial sample size.\n \"\"\"\n\n @classmethod\n def from_source(\n cls: type[StudyIndexFinnGen],\n finngen_studies: DataFrame,\n finngen_release_prefix: str,\n finngen_sumstat_url_prefix: str,\n finngen_sumstat_url_suffix: str,\n ) -> StudyIndexFinnGen:\n\"\"\"This function ingests study level metadata from FinnGen.\n\n Args:\n finngen_studies (DataFrame): FinnGen raw study table\n finngen_release_prefix (str): Release prefix pattern.\n finngen_sumstat_url_prefix (str): URL prefix for summary statistics location.\n finngen_sumstat_url_suffix (str): URL prefix suffix for summary statistics location.\n\n Returns:\n StudyIndexFinnGen: Parsed and annotated FinnGen study table.\n \"\"\"\n return cls(\n _df=(\n # Read FinnGen raw data.\n finngen_studies.select(\n # Select the desired columns.\n f.concat(\n f.lit(finngen_release_prefix + \"_\"), f.col(\"phenocode\")\n ).alias(\"studyId\"),\n f.col(\"phenostring\").alias(\"traitFromSource\"),\n f.col(\"num_cases\").alias(\"nCases\"),\n f.col(\"num_controls\").alias(\"nControls\"),\n # Set constant value columns.\n f.lit(finngen_release_prefix).alias(\"projectId\"),\n f.lit(\"gwas\").alias(\"studyType\"),\n f.lit(True).alias(\"hasSumstats\"),\n f.lit(\"377,277 (210,870 females and 166,407 males)\").alias(\n \"initialSampleSize\"\n ),\n )\n .withColumn(\"nSamples\", f.col(\"nCases\") + f.col(\"nControls\"))\n .withColumn(\n \"summarystatsLocation\",\n f.concat(\n f.lit(finngen_sumstat_url_prefix),\n f.col(\"studyId\"),\n f.lit(finngen_sumstat_url_suffix),\n ),\n )\n )\n )\n
This dataset captures associations between study/traits and a genetic loci as provided by finemapping methods.
Source code in src/otg/dataset/study_locus.py
@dataclass\nclass StudyLocus(Dataset):\n\"\"\"Study-Locus dataset.\n\n This dataset captures associations between study/traits and a genetic loci as provided by finemapping methods.\n \"\"\"\n\n _schema: StructType = parse_spark_schema(\"study_locus.json\")\n\n @staticmethod\n def _overlapping_peaks(credset_to_overlap: DataFrame) -> DataFrame:\n\"\"\"Calculate overlapping signals (study-locus) between GWAS-GWAS and GWAS-Molecular trait.\n\n Args:\n credset_to_overlap (DataFrame): DataFrame containing at least `studyLocusId`, `studyType`, `chromosome` and `tagVariantId` columns.\n\n Returns:\n DataFrame: containing `left_studyLocusId`, `right_studyLocusId` and `chromosome` columns.\n \"\"\"\n # Reduce columns to the minimum to reduce the size of the dataframe\n credset_to_overlap = credset_to_overlap.select(\n \"studyLocusId\", \"studyType\", \"chromosome\", \"tagVariantId\"\n )\n return (\n credset_to_overlap.alias(\"left\")\n .filter(f.col(\"studyType\") == \"gwas\")\n # Self join with complex condition. Left it's all gwas and right can be gwas or molecular trait\n .join(\n credset_to_overlap.alias(\"right\"),\n on=[\n f.col(\"left.chromosome\") == f.col(\"right.chromosome\"),\n f.col(\"left.tagVariantId\") == f.col(\"right.tagVariantId\"),\n (f.col(\"right.studyType\") != \"gwas\")\n | (f.col(\"left.studyLocusId\") > f.col(\"right.studyLocusId\")),\n ],\n how=\"inner\",\n )\n .select(\n f.col(\"left.studyLocusId\").alias(\"left_studyLocusId\"),\n f.col(\"right.studyLocusId\").alias(\"right_studyLocusId\"),\n f.col(\"left.chromosome\").alias(\"chromosome\"),\n )\n .distinct()\n .repartition(\"chromosome\")\n .persist()\n )\n\n @staticmethod\n def _align_overlapping_tags(\n credset_to_overlap: DataFrame, peak_overlaps: DataFrame\n ) -> StudyLocusOverlap:\n\"\"\"Align overlapping tags in pairs of overlapping study-locus, keeping all tags in both loci.\n\n Args:\n credset_to_overlap (DataFrame): containing `studyLocusId`, `studyType`, `chromosome`, `tagVariantId`, `logABF` and `posteriorProbability` columns.\n peak_overlaps (DataFrame): containing `left_studyLocusId`, `right_studyLocusId` and `chromosome` columns.\n\n Returns:\n StudyLocusOverlap: Pairs of overlapping study-locus with aligned tags.\n \"\"\"\n # Complete information about all tags in the left study-locus of the overlap\n overlapping_left = credset_to_overlap.select(\n f.col(\"chromosome\"),\n f.col(\"tagVariantId\"),\n f.col(\"studyLocusId\").alias(\"left_studyLocusId\"),\n f.col(\"logABF\").alias(\"left_logABF\"),\n f.col(\"posteriorProbability\").alias(\"left_posteriorProbability\"),\n ).join(peak_overlaps, on=[\"chromosome\", \"left_studyLocusId\"], how=\"inner\")\n\n # Complete information about all tags in the right study-locus of the overlap\n overlapping_right = credset_to_overlap.select(\n f.col(\"chromosome\"),\n f.col(\"tagVariantId\"),\n f.col(\"studyLocusId\").alias(\"right_studyLocusId\"),\n f.col(\"logABF\").alias(\"right_logABF\"),\n f.col(\"posteriorProbability\").alias(\"right_posteriorProbability\"),\n ).join(peak_overlaps, on=[\"chromosome\", \"right_studyLocusId\"], how=\"inner\")\n\n # Include information about all tag variants in both study-locus aligned by tag variant id\n return StudyLocusOverlap(\n _df=overlapping_left.join(\n overlapping_right,\n on=[\n \"chromosome\",\n \"right_studyLocusId\",\n \"left_studyLocusId\",\n \"tagVariantId\",\n ],\n how=\"outer\",\n )\n # ensures nullable=false for following columns\n .fillna(\n value=\"unknown\",\n subset=[\n \"chromosome\",\n \"right_studyLocusId\",\n \"left_studyLocusId\",\n \"tagVariantId\",\n ],\n )\n )\n\n @staticmethod\n def _update_quality_flag(\n qc: Column, flag_condition: Column, flag_text: StudyLocusQualityCheck\n ) -> Column:\n\"\"\"Update the provided quality control list with a new flag if condition is met.\n\n Args:\n qc (Column): Array column with the current list of qc flags.\n flag_condition (Column): This is a column of booleans, signing which row should be flagged\n flag_text (StudyLocusQualityCheck): Text for the new quality control flag\n\n Returns:\n Column: Array column with the updated list of qc flags.\n \"\"\"\n qc = f.when(qc.isNull(), f.array()).otherwise(qc)\n return f.when(\n flag_condition,\n f.array_union(qc, f.array(f.lit(flag_text.value))),\n ).otherwise(qc)\n\n @classmethod\n def from_parquet(cls: type[StudyLocus], session: Session, path: str) -> StudyLocus:\n\"\"\"Initialise StudyLocus from parquet file.\n\n Args:\n session (Session): spark session\n path (str): Path to parquet file\n\n Returns:\n StudyLocus: Study-locus dataset\n \"\"\"\n df = session.read_parquet(path=path, schema=cls._schema)\n return cls(_df=df, _schema=cls._schema)\n\n def credible_set(\n self: StudyLocus,\n credible_interval: CredibleInterval,\n ) -> StudyLocus:\n\"\"\"Filter study-locus tag variants based on given credible interval.\n\n Args:\n credible_interval (CredibleInterval): Credible interval to filter for.\n\n Returns:\n StudyLocus: Filtered study-locus dataset.\n \"\"\"\n self.df = self._df.withColumn(\n \"credibleSet\",\n f.expr(f\"filter(credibleSet, tag -> (tag.{credible_interval.value}))\"),\n )\n return self\n\n def overlaps(self: StudyLocus, study_index: StudyIndex) -> StudyLocusOverlap:\n\"\"\"Calculate overlapping study-locus.\n\n Find overlapping study-locus that share at least one tagging variant. All GWAS-GWAS and all GWAS-Molecular traits are computed with the Molecular traits always\n appearing on the right side.\n\n Args:\n study_index (StudyIndex): Study index to resolve study types.\n\n Returns:\n StudyLocusOverlap: Pairs of overlapping study-locus with aligned tags.\n \"\"\"\n credset_to_overlap = (\n self.df.join(study_index.study_type_lut(), on=\"studyId\", how=\"inner\")\n .withColumn(\"credibleSet\", f.explode(\"credibleSet\"))\n .select(\n \"studyLocusId\",\n \"studyType\",\n \"chromosome\",\n f.col(\"credibleSet.tagVariantId\").alias(\"tagVariantId\"),\n f.col(\"credibleSet.logABF\").alias(\"logABF\"),\n f.col(\"credibleSet.posteriorProbability\").alias(\"posteriorProbability\"),\n )\n .persist()\n )\n\n # overlapping study-locus\n peak_overlaps = self._overlapping_peaks(credset_to_overlap)\n\n # study-locus overlap by aligning overlapping variants\n return self._align_overlapping_tags(credset_to_overlap, peak_overlaps)\n\n def unique_lead_tag_variants(self: StudyLocus) -> DataFrame:\n\"\"\"All unique lead and tag variants contained in the `StudyLocus` dataframe.\n\n Returns:\n DataFrame: A dataframe containing `variantId` and `chromosome` columns.\n \"\"\"\n lead_tags = (\n self.df.select(\n f.col(\"variantId\"),\n f.col(\"chromosome\"),\n f.explode(\"credibleSet.tagVariantId\").alias(\"tagVariantId\"),\n )\n .repartition(\"chromosome\")\n .persist()\n )\n return (\n lead_tags.select(\"variantId\", \"chromosome\")\n .union(\n lead_tags.select(f.col(\"tagVariantId\").alias(\"variantId\"), \"chromosome\")\n )\n .distinct()\n )\n\n def unique_study_locus_ancestries(\n self: StudyLocus, studies: StudyIndexGWASCatalog\n ) -> DataFrame:\n\"\"\"All unique lead variant and ancestries contained in the `StudyLocus`.\n\n Args:\n studies (StudyIndexGWASCatalog): Metadata about studies in the `StudyLocus`.\n\n Returns:\n DataFrame: unique [\"variantId\", \"studyId\", \"gnomadPopulation\", \"chromosome\", \"relativeSampleSize\"]\n\n Note:\n This method is only available for GWAS Catalog studies.\n \"\"\"\n return (\n self.df.join(\n studies.get_gnomad_ancestry_sample_sizes(), on=\"studyId\", how=\"left\"\n )\n .filter(f.col(\"position\").isNotNull())\n .select(\n \"variantId\",\n \"chromosome\",\n \"studyId\",\n \"gnomadPopulation\",\n \"relativeSampleSize\",\n )\n .distinct()\n )\n\n def neglog_pvalue(self: StudyLocus) -> Column:\n\"\"\"Returns the negative log p-value.\n\n Returns:\n Column: Negative log p-value\n \"\"\"\n return calculate_neglog_pvalue(\n self.df.pValueMantissa,\n self.df.pValueExponent,\n )\n\n def annotate_credible_sets(self: StudyLocus) -> StudyLocus:\n\"\"\"Annotate study-locus dataset with credible set flags.\n\n Sorts the array in the `credibleSet` column elements by their `posteriorProbability` values in descending order and adds\n `is95CredibleSet` and `is99CredibleSet` fields to the elements, indicating which are the tagging variants whose cumulative sum\n of their `posteriorProbability` values is below 0.95 and 0.99, respectively.\n\n Returns:\n StudyLocus: including annotation on `is95CredibleSet` and `is99CredibleSet`.\n \"\"\"\n self.df = self.df.withColumn(\n # Sort credible set by posterior probability in descending order\n \"credibleSet\",\n f.when(\n f.size(f.col(\"credibleSet\")) > 0,\n order_array_of_structs_by_field(\"credibleSet\", \"posteriorProbability\"),\n ).when(f.size(f.col(\"credibleSet\")) == 0, f.col(\"credibleSet\")),\n ).withColumn(\n # Calculate array of cumulative sums of posterior probabilities to determine which variants are in the 95% and 99% credible sets\n # and zip the cumulative sums array with the credible set array to add the flags\n \"credibleSet\",\n f.when(\n f.size(f.col(\"credibleSet\")) > 0,\n f.zip_with(\n f.col(\"credibleSet\"),\n f.transform(\n f.sequence(f.lit(1), f.size(f.col(\"credibleSet\"))),\n lambda index: f.aggregate(\n f.slice(\n # By using `index - 1` we introduce a value of `0.0` in the cumulative sums array. to ensure that the last variant\n # that exceeds the 0.95 threshold is included in the cumulative sum, as its probability is necessary to satisfy the threshold.\n f.col(\"credibleSet.posteriorProbability\"),\n 1,\n index - 1,\n ),\n f.lit(0.0),\n lambda acc, el: acc + el,\n ),\n ),\n lambda struct_e, acc: struct_e.withField(\n CredibleInterval.IS95.value, acc < 0.95\n ).withField(CredibleInterval.IS99.value, acc < 0.99),\n ),\n ).when(f.size(f.col(\"credibleSet\")) == 0, f.col(\"credibleSet\")),\n )\n return self\n\n def clump(self: StudyLocus) -> StudyLocus:\n\"\"\"Perform LD clumping of the studyLocus.\n\n Evaluates whether a lead variant is linked to a tag (with lowest p-value) in the same studyLocus dataset.\n\n Returns:\n StudyLocus: with empty credible sets for linked variants and QC flag.\n \"\"\"\n self.df = (\n self.df.withColumn(\n \"is_lead_linked\",\n LDclumping._is_lead_linked(\n self.df.studyId,\n self.df.variantId,\n self.df.pValueExponent,\n self.df.pValueMantissa,\n self.df.credibleSet,\n ),\n )\n .withColumn(\n \"credibleSet\",\n f.when(f.col(\"is_lead_linked\"), f.array()).otherwise(\n f.col(\"credibleSet\")\n ),\n )\n .withColumn(\n \"qualityControls\",\n StudyLocus._update_quality_flag(\n f.col(\"qualityControls\"),\n f.col(\"is_lead_linked\"),\n StudyLocusQualityCheck.LD_CLUMPED,\n ),\n )\n .drop(\"is_lead_linked\")\n )\n return self\n
Annotate study-locus dataset with credible set flags.
Sorts the array in the credibleSet column elements by their posteriorProbability values in descending order and adds is95CredibleSet and is99CredibleSet fields to the elements, indicating which are the tagging variants whose cumulative sum of their posteriorProbability values is below 0.95 and 0.99, respectively.
Returns:
Name Type Description StudyLocusStudyLocus
including annotation on is95CredibleSet and is99CredibleSet.
Source code in src/otg/dataset/study_locus.py
def annotate_credible_sets(self: StudyLocus) -> StudyLocus:\n\"\"\"Annotate study-locus dataset with credible set flags.\n\n Sorts the array in the `credibleSet` column elements by their `posteriorProbability` values in descending order and adds\n `is95CredibleSet` and `is99CredibleSet` fields to the elements, indicating which are the tagging variants whose cumulative sum\n of their `posteriorProbability` values is below 0.95 and 0.99, respectively.\n\n Returns:\n StudyLocus: including annotation on `is95CredibleSet` and `is99CredibleSet`.\n \"\"\"\n self.df = self.df.withColumn(\n # Sort credible set by posterior probability in descending order\n \"credibleSet\",\n f.when(\n f.size(f.col(\"credibleSet\")) > 0,\n order_array_of_structs_by_field(\"credibleSet\", \"posteriorProbability\"),\n ).when(f.size(f.col(\"credibleSet\")) == 0, f.col(\"credibleSet\")),\n ).withColumn(\n # Calculate array of cumulative sums of posterior probabilities to determine which variants are in the 95% and 99% credible sets\n # and zip the cumulative sums array with the credible set array to add the flags\n \"credibleSet\",\n f.when(\n f.size(f.col(\"credibleSet\")) > 0,\n f.zip_with(\n f.col(\"credibleSet\"),\n f.transform(\n f.sequence(f.lit(1), f.size(f.col(\"credibleSet\"))),\n lambda index: f.aggregate(\n f.slice(\n # By using `index - 1` we introduce a value of `0.0` in the cumulative sums array. to ensure that the last variant\n # that exceeds the 0.95 threshold is included in the cumulative sum, as its probability is necessary to satisfy the threshold.\n f.col(\"credibleSet.posteriorProbability\"),\n 1,\n index - 1,\n ),\n f.lit(0.0),\n lambda acc, el: acc + el,\n ),\n ),\n lambda struct_e, acc: struct_e.withField(\n CredibleInterval.IS95.value, acc < 0.95\n ).withField(CredibleInterval.IS99.value, acc < 0.99),\n ),\n ).when(f.size(f.col(\"credibleSet\")) == 0, f.col(\"credibleSet\")),\n )\n return self\n
Find overlapping study-locus that share at least one tagging variant. All GWAS-GWAS and all GWAS-Molecular traits are computed with the Molecular traits always appearing on the right side.
Parameters:
Name Type Description Default study_indexStudyIndex
Study index to resolve study types.
required
Returns:
Name Type Description StudyLocusOverlapStudyLocusOverlap
Pairs of overlapping study-locus with aligned tags.
Source code in src/otg/dataset/study_locus.py
def overlaps(self: StudyLocus, study_index: StudyIndex) -> StudyLocusOverlap:\n\"\"\"Calculate overlapping study-locus.\n\n Find overlapping study-locus that share at least one tagging variant. All GWAS-GWAS and all GWAS-Molecular traits are computed with the Molecular traits always\n appearing on the right side.\n\n Args:\n study_index (StudyIndex): Study index to resolve study types.\n\n Returns:\n StudyLocusOverlap: Pairs of overlapping study-locus with aligned tags.\n \"\"\"\n credset_to_overlap = (\n self.df.join(study_index.study_type_lut(), on=\"studyId\", how=\"inner\")\n .withColumn(\"credibleSet\", f.explode(\"credibleSet\"))\n .select(\n \"studyLocusId\",\n \"studyType\",\n \"chromosome\",\n f.col(\"credibleSet.tagVariantId\").alias(\"tagVariantId\"),\n f.col(\"credibleSet.logABF\").alias(\"logABF\"),\n f.col(\"credibleSet.posteriorProbability\").alias(\"posteriorProbability\"),\n )\n .persist()\n )\n\n # overlapping study-locus\n peak_overlaps = self._overlapping_peaks(credset_to_overlap)\n\n # study-locus overlap by aligning overlapping variants\n return self._align_overlapping_tags(credset_to_overlap, peak_overlaps)\n
Study-Locus quality control options listing concerns on the quality of the association.
Attributes:
Name Type Description SUBSIGNIFICANT_FLAGstr
p-value below significance threshold
NO_GENOMIC_LOCATION_FLAGstr
Incomplete genomic mapping
COMPOSITE_FLAGstr
Composite association due to variant x variant interactions
VARIANT_INCONSISTENCY_FLAGstr
Inconsistencies in the reported variants
NON_MAPPED_VARIANT_FLAGstr
Variant not mapped to GnomAd
PALINDROMIC_ALLELE_FLAGstr
Alleles are palindromic - cannot harmonize
AMBIGUOUS_STUDYstr
Association with ambiguous study
UNRESOLVED_LDstr
Variant not found in LD reference
LD_CLUMPEDstr
Explained by a more significant variant in high LD (clumped)
Source code in src/otg/dataset/study_locus.py
class StudyLocusQualityCheck(Enum):\n\"\"\"Study-Locus quality control options listing concerns on the quality of the association.\n\n Attributes:\n SUBSIGNIFICANT_FLAG (str): p-value below significance threshold\n NO_GENOMIC_LOCATION_FLAG (str): Incomplete genomic mapping\n COMPOSITE_FLAG (str): Composite association due to variant x variant interactions\n VARIANT_INCONSISTENCY_FLAG (str): Inconsistencies in the reported variants\n NON_MAPPED_VARIANT_FLAG (str): Variant not mapped to GnomAd\n PALINDROMIC_ALLELE_FLAG (str): Alleles are palindromic - cannot harmonize\n AMBIGUOUS_STUDY (str): Association with ambiguous study\n UNRESOLVED_LD (str): Variant not found in LD reference\n LD_CLUMPED (str): Explained by a more significant variant in high LD (clumped)\n \"\"\"\n\n SUBSIGNIFICANT_FLAG = \"Subsignificant p-value\"\n NO_GENOMIC_LOCATION_FLAG = \"Incomplete genomic mapping\"\n COMPOSITE_FLAG = \"Composite association\"\n INCONSISTENCY_FLAG = \"Variant inconsistency\"\n NON_MAPPED_VARIANT_FLAG = \"No mapping in GnomAd\"\n PALINDROMIC_ALLELE_FLAG = \"Palindrome alleles - cannot harmonize\"\n AMBIGUOUS_STUDY = \"Association with ambiguous study\"\n UNRESOLVED_LD = \"Variant not found in LD reference\"\n LD_CLUMPED = \"Explained by a more significant variant in high LD (clumped)\"\n
Interval within which an unobserved parameter value falls with a particular probability.
Attributes:
Name Type Description IS95str
95% credible interval
IS99str
99% credible interval
Source code in src/otg/dataset/study_locus.py
class CredibleInterval(Enum):\n\"\"\"Credible interval enum.\n\n Interval within which an unobserved parameter value falls with a particular probability.\n\n Attributes:\n IS95 (str): 95% credible interval\n IS99 (str): 99% credible interval\n \"\"\"\n\n IS95 = \"is95CredibleSet\"\n IS99 = \"is99CredibleSet\"\n
"},{"location":"components/dataset/study_locus/study_locus_gwas_catalog/","title":"Study locus gwas catalog","text":"
Bases: StudyLocus
Study-locus dataset derived from GWAS Catalog.
Source code in src/otg/dataset/study_locus.py
class StudyLocusGWASCatalog(StudyLocus):\n\"\"\"Study-locus dataset derived from GWAS Catalog.\"\"\"\n\n @staticmethod\n def _parse_pvalue(pvalue: Column) -> tuple[Column, Column]:\n\"\"\"Parse p-value column.\n\n Args:\n pvalue (Column): p-value [string]\n\n Returns:\n tuple[Column, Column]: p-value mantissa and exponent\n\n Example:\n >>> import pyspark.sql.types as t\n >>> d = [(\"1.0\"), (\"0.5\"), (\"1E-20\"), (\"3E-3\"), (\"1E-1000\")]\n >>> df = spark.createDataFrame(d, t.StringType())\n >>> df.select('value',*StudyLocusGWASCatalog._parse_pvalue(f.col('value'))).show()\n +-------+--------------+--------------+\n | value|pValueMantissa|pValueExponent|\n +-------+--------------+--------------+\n | 1.0| 1.0| 1|\n | 0.5| 0.5| 1|\n | 1E-20| 1.0| -20|\n | 3E-3| 3.0| -3|\n |1E-1000| 1.0| -1000|\n +-------+--------------+--------------+\n <BLANKLINE>\n\n \"\"\"\n split = f.split(pvalue, \"E\")\n return split.getItem(0).cast(\"float\").alias(\"pValueMantissa\"), f.coalesce(\n split.getItem(1).cast(\"integer\"), f.lit(1)\n ).alias(\"pValueExponent\")\n\n @staticmethod\n def _normalise_pvaluetext(p_value_text: Column) -> Column:\n\"\"\"Normalised p-value text column to a standardised format.\n\n For cases where there is no mapping, the value is set to null.\n\n Args:\n p_value_text (Column): `pValueText` column from GWASCatalog\n\n Returns:\n Column: Array column after using GWAS Catalog mappings. There might be multiple mappings for a single p-value text.\n\n Example:\n >>> import pyspark.sql.types as t\n >>> d = [(\"European Ancestry\"), (\"African ancestry\"), (\"Alzheimer\u2019s Disease\"), (\"(progression)\"), (\"\"), (None)]\n >>> df = spark.createDataFrame(d, t.StringType())\n >>> df.withColumn('normalised', StudyLocusGWASCatalog._normalise_pvaluetext(f.col('value'))).show()\n +-------------------+----------+\n | value|normalised|\n +-------------------+----------+\n | European Ancestry| [EA]|\n | African ancestry| [AA]|\n |Alzheimer\u2019s Disease| [AD]|\n | (progression)| null|\n | | null|\n | null| null|\n +-------------------+----------+\n <BLANKLINE>\n\n \"\"\"\n # GWAS Catalog to p-value mapping\n json_dict = json.loads(\n pkg_resources.read_text(data, \"gwas_pValueText_map.json\", encoding=\"utf-8\")\n )\n map_expr = f.create_map(*[f.lit(x) for x in chain(*json_dict.items())])\n\n splitted_col = f.split(f.regexp_replace(p_value_text, r\"[\\(\\)]\", \"\"), \",\")\n mapped_col = f.transform(splitted_col, lambda x: map_expr[x])\n return f.when(f.forall(mapped_col, lambda x: x.isNull()), None).otherwise(\n mapped_col\n )\n\n @staticmethod\n def _normalise_risk_allele(risk_allele: Column) -> Column:\n\"\"\"Normalised risk allele column to a standardised format.\n\n If multiple risk alleles are present, the first one is returned.\n\n Args:\n risk_allele (Column): `riskAllele` column from GWASCatalog\n\n Returns:\n Column: mapped using GWAS Catalog mapping\n\n Example:\n >>> import pyspark.sql.types as t\n >>> d = [(\"rs1234-A-G\"), (\"rs1234-A\"), (\"rs1234-A; rs1235-G\")]\n >>> df = spark.createDataFrame(d, t.StringType())\n >>> df.withColumn('normalised', StudyLocusGWASCatalog._normalise_risk_allele(f.col('value'))).show()\n +------------------+----------+\n | value|normalised|\n +------------------+----------+\n | rs1234-A-G| A|\n | rs1234-A| A|\n |rs1234-A; rs1235-G| A|\n +------------------+----------+\n <BLANKLINE>\n\n \"\"\"\n # GWAS Catalog to risk allele mapping\n return f.split(f.split(risk_allele, \"; \").getItem(0), \"-\").getItem(1)\n\n @staticmethod\n def _collect_rsids(\n snp_id: Column, snp_id_current: Column, risk_allele: Column\n ) -> Column:\n\"\"\"It takes three columns, and returns an array of distinct values from those columns.\n\n Args:\n snp_id (Column): The original snp id from the GWAS catalog.\n snp_id_current (Column): The current snp id field is just a number at the moment (stored as a string). Adding 'rs' prefix if looks good.\n risk_allele (Column): The risk allele for the SNP.\n\n Returns:\n An array of distinct values.\n \"\"\"\n # The current snp id field is just a number at the moment (stored as a string). Adding 'rs' prefix if looks good.\n snp_id_current = f.when(\n snp_id_current.rlike(\"^[0-9]*$\"),\n f.format_string(\"rs%s\", snp_id_current),\n )\n # Cleaning risk allele:\n risk_allele = f.split(risk_allele, \"-\").getItem(0)\n\n # Collecting all values:\n return f.array_distinct(f.array(snp_id, snp_id_current, risk_allele))\n\n @staticmethod\n def _map_to_variant_annotation_variants(\n gwas_associations: DataFrame, variant_annotation: VariantAnnotation\n ) -> DataFrame:\n\"\"\"Add variant metadata in associations.\n\n Args:\n gwas_associations (DataFrame): raw GWAS Catalog associations\n variant_annotation (VariantAnnotation): variant annotation dataset\n\n Returns:\n DataFrame: GWAS Catalog associations data including `variantId`, `referenceAllele`,\n `alternateAllele`, `chromosome`, `position` with variant metadata\n \"\"\"\n # Subset of GWAS Catalog associations required for resolving variant IDs:\n gwas_associations_subset = gwas_associations.select(\n \"studyLocusId\",\n f.col(\"CHR_ID\").alias(\"chromosome\"),\n f.col(\"CHR_POS\").cast(IntegerType()).alias(\"position\"),\n # List of all SNPs associated with the variant\n StudyLocusGWASCatalog._collect_rsids(\n f.split(f.col(\"SNPS\"), \"; \").getItem(0),\n f.col(\"SNP_ID_CURRENT\"),\n f.split(f.col(\"STRONGEST SNP-RISK ALLELE\"), \"; \").getItem(0),\n ).alias(\"rsIdsGwasCatalog\"),\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ).alias(\"riskAllele\"),\n )\n\n # Subset of variant annotation required for GWAS Catalog annotations:\n va_subset = variant_annotation.df.select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n f.col(\"rsIds\").alias(\"rsIdsGnomad\"),\n \"referenceAllele\",\n \"alternateAllele\",\n \"alleleFrequencies\",\n variant_annotation.max_maf().alias(\"maxMaf\"),\n ).join(\n f.broadcast(\n gwas_associations_subset.select(\"chromosome\", \"position\").distinct()\n ),\n on=[\"chromosome\", \"position\"],\n how=\"inner\",\n )\n\n # Semi-resolved ids (still contains duplicates when conclusion was not possible to make\n # based on rsIds or allele concordance)\n filtered_associations = (\n gwas_associations_subset.join(\n f.broadcast(va_subset),\n on=[\"chromosome\", \"position\"],\n how=\"left\",\n )\n .withColumn(\n \"rsIdFilter\",\n StudyLocusGWASCatalog._flag_mappings_to_retain(\n f.col(\"studyLocusId\"),\n StudyLocusGWASCatalog._compare_rsids(\n f.col(\"rsIdsGnomad\"), f.col(\"rsIdsGwasCatalog\")\n ),\n ),\n )\n .withColumn(\n \"concordanceFilter\",\n StudyLocusGWASCatalog._flag_mappings_to_retain(\n f.col(\"studyLocusId\"),\n StudyLocusGWASCatalog._check_concordance(\n f.col(\"riskAllele\"),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n ),\n ),\n )\n .filter(\n # Filter out rows where GWAS Catalog rsId does not match with GnomAD rsId,\n # but there is corresponding variant for the same association\n f.col(\"rsIdFilter\")\n # or filter out rows where GWAS Catalog alleles are not concordant with GnomAD alleles,\n # but there is corresponding variant for the same association\n | f.col(\"concordanceFilter\")\n )\n )\n\n # Keep only highest maxMaf variant per studyLocusId\n fully_mapped_associations = get_record_with_maximum_value(\n filtered_associations, grouping_col=\"studyLocusId\", sorting_col=\"maxMaf\"\n ).select(\n \"studyLocusId\",\n \"variantId\",\n \"referenceAllele\",\n \"alternateAllele\",\n \"chromosome\",\n \"position\",\n )\n\n return gwas_associations.join(\n fully_mapped_associations, on=\"studyLocusId\", how=\"left\"\n )\n\n @staticmethod\n def _compare_rsids(gnomad: Column, gwas: Column) -> Column:\n\"\"\"If the intersection of the two arrays is greater than 0, return True, otherwise return False.\n\n Args:\n gnomad (Column): rsids from gnomad\n gwas (Column): rsids from the GWAS Catalog\n\n Returns:\n A boolean column that is true if the GnomAD rsIDs can be found in the GWAS rsIDs.\n\n Examples:\n >>> d = [\n ... (1, [\"rs123\", \"rs523\"], [\"rs123\"]),\n ... (2, [], [\"rs123\"]),\n ... (3, [\"rs123\", \"rs523\"], []),\n ... (4, [], []),\n ... ]\n >>> df = spark.createDataFrame(d, ['associationId', 'gnomad', 'gwas'])\n >>> df.withColumn(\"rsid_matches\", StudyLocusGWASCatalog._compare_rsids(f.col(\"gnomad\"),f.col('gwas'))).show()\n +-------------+--------------+-------+------------+\n |associationId| gnomad| gwas|rsid_matches|\n +-------------+--------------+-------+------------+\n | 1|[rs123, rs523]|[rs123]| true|\n | 2| []|[rs123]| false|\n | 3|[rs123, rs523]| []| false|\n | 4| []| []| false|\n +-------------+--------------+-------+------------+\n <BLANKLINE>\n\n \"\"\"\n return f.when(f.size(f.array_intersect(gnomad, gwas)) > 0, True).otherwise(\n False\n )\n\n @staticmethod\n def _flag_mappings_to_retain(\n association_id: Column, filter_column: Column\n ) -> Column:\n\"\"\"Flagging mappings to drop for each association.\n\n Some associations have multiple mappings. Some has matching rsId others don't. We only\n want to drop the non-matching mappings, when a matching is available for the given association.\n This logic can be generalised for other measures eg. allele concordance.\n\n Args:\n association_id (Column): association identifier column\n filter_column (Column): boolean col indicating to keep a mapping\n\n Returns:\n A column with a boolean value.\n\n Examples:\n >>> d = [\n ... (1, False),\n ... (1, False),\n ... (2, False),\n ... (2, True),\n ... (3, True),\n ... (3, True),\n ... ]\n >>> df = spark.createDataFrame(d, ['associationId', 'filter'])\n >>> df.withColumn(\"isConcordant\", StudyLocusGWASCatalog._flag_mappings_to_retain(f.col(\"associationId\"),f.col('filter'))).show()\n +-------------+------+------------+\n |associationId|filter|isConcordant|\n +-------------+------+------------+\n | 1| false| true|\n | 1| false| true|\n | 2| false| false|\n | 2| true| true|\n | 3| true| true|\n | 3| true| true|\n +-------------+------+------------+\n <BLANKLINE>\n\n \"\"\"\n w = Window.partitionBy(association_id)\n\n # Generating a boolean column informing if the filter column contains true anywhere for the association:\n aggregated_filter = f.when(\n f.array_contains(f.collect_set(filter_column).over(w), True), True\n ).otherwise(False)\n\n # Generate a filter column:\n return f.when(aggregated_filter & (~filter_column), False).otherwise(True)\n\n @staticmethod\n def _check_concordance(\n risk_allele: Column, reference_allele: Column, alternate_allele: Column\n ) -> Column:\n\"\"\"A function to check if the risk allele is concordant with the alt or ref allele.\n\n If the risk allele is the same as the reference or alternate allele, or if the reverse complement of\n the risk allele is the same as the reference or alternate allele, then the allele is concordant.\n If no mapping is available (ref/alt is null), the function returns True.\n\n Args:\n risk_allele (Column): The allele that is associated with the risk of the disease.\n reference_allele (Column): The reference allele from the GWAS catalog\n alternate_allele (Column): The alternate allele of the variant.\n\n Returns:\n A boolean column that is True if the risk allele is the same as the reference or alternate allele,\n or if the reverse complement of the risk allele is the same as the reference or alternate allele.\n\n Examples:\n >>> d = [\n ... ('A', 'A', 'G'),\n ... ('A', 'T', 'G'),\n ... ('A', 'C', 'G'),\n ... ('A', 'A', '?'),\n ... (None, None, 'A'),\n ... ]\n >>> df = spark.createDataFrame(d, ['riskAllele', 'referenceAllele', 'alternateAllele'])\n >>> df.withColumn(\"isConcordant\", StudyLocusGWASCatalog._check_concordance(f.col(\"riskAllele\"),f.col('referenceAllele'), f.col('alternateAllele'))).show()\n +----------+---------------+---------------+------------+\n |riskAllele|referenceAllele|alternateAllele|isConcordant|\n +----------+---------------+---------------+------------+\n | A| A| G| true|\n | A| T| G| true|\n | A| C| G| false|\n | A| A| ?| true|\n | null| null| A| true|\n +----------+---------------+---------------+------------+\n <BLANKLINE>\n\n \"\"\"\n # Calculating the reverse complement of the risk allele:\n risk_allele_reverse_complement = f.when(\n risk_allele.rlike(r\"^[ACTG]+$\"),\n f.reverse(f.translate(risk_allele, \"ACTG\", \"TGAC\")),\n ).otherwise(risk_allele)\n\n # OK, is the risk allele or the reverse complent is the same as the mapped alleles:\n return (\n f.when(\n (risk_allele == reference_allele) | (risk_allele == alternate_allele),\n True,\n )\n # If risk allele is found on the negative strand:\n .when(\n (risk_allele_reverse_complement == reference_allele)\n | (risk_allele_reverse_complement == alternate_allele),\n True,\n )\n # If risk allele is ambiguous, still accepted: < This condition could be reconsidered\n .when(risk_allele == \"?\", True)\n # If the association could not be mapped we keep it:\n .when(reference_allele.isNull(), True)\n # Allele is discordant:\n .otherwise(False)\n )\n\n @staticmethod\n def _get_reverse_complement(allele_col: Column) -> Column:\n\"\"\"A function to return the reverse complement of an allele column.\n\n It takes a string and returns the reverse complement of that string if it's a DNA sequence,\n otherwise it returns the original string. Assumes alleles in upper case.\n\n Args:\n allele_col (Column): The column containing the allele to reverse complement.\n\n Returns:\n A column that is the reverse complement of the allele column.\n\n Examples:\n >>> d = [{\"allele\": 'A'}, {\"allele\": 'T'},{\"allele\": 'G'}, {\"allele\": 'C'},{\"allele\": 'AC'}, {\"allele\": 'GTaatc'},{\"allele\": '?'}, {\"allele\": None}]\n >>> df = spark.createDataFrame(d)\n >>> df.withColumn(\"revcom_allele\", StudyLocusGWASCatalog._get_reverse_complement(f.col(\"allele\"))).show()\n +------+-------------+\n |allele|revcom_allele|\n +------+-------------+\n | A| T|\n | T| A|\n | G| C|\n | C| G|\n | AC| GT|\n |GTaatc| GATTAC|\n | ?| ?|\n | null| null|\n +------+-------------+\n <BLANKLINE>\n\n \"\"\"\n allele_col = f.upper(allele_col)\n return f.when(\n allele_col.rlike(\"[ACTG]+\"),\n f.reverse(f.translate(allele_col, \"ACTG\", \"TGAC\")),\n ).otherwise(allele_col)\n\n @staticmethod\n def _effect_needs_harmonisation(\n risk_allele: Column, reference_allele: Column\n ) -> Column:\n\"\"\"A function to check if the effect allele needs to be harmonised.\n\n Args:\n risk_allele (Column): Risk allele column\n reference_allele (Column): Effect allele column\n\n Returns:\n A boolean column indicating if the effect allele needs to be harmonised.\n\n Examples:\n >>> d = [{\"risk\": 'A', \"reference\": 'A'}, {\"risk\": 'A', \"reference\": 'T'}, {\"risk\": 'AT', \"reference\": 'TA'}, {\"risk\": 'AT', \"reference\": 'AT'}]\n >>> df = spark.createDataFrame(d)\n >>> df.withColumn(\"needs_harmonisation\", StudyLocusGWASCatalog._effect_needs_harmonisation(f.col(\"risk\"), f.col(\"reference\"))).show()\n +---------+----+-------------------+\n |reference|risk|needs_harmonisation|\n +---------+----+-------------------+\n | A| A| true|\n | T| A| true|\n | TA| AT| false|\n | AT| AT| true|\n +---------+----+-------------------+\n <BLANKLINE>\n\n \"\"\"\n return (risk_allele == reference_allele) | (\n risk_allele\n == StudyLocusGWASCatalog._get_reverse_complement(reference_allele)\n )\n\n @staticmethod\n def _are_alleles_palindromic(\n reference_allele: Column, alternate_allele: Column\n ) -> Column:\n\"\"\"A function to check if the alleles are palindromic.\n\n Args:\n reference_allele (Column): Reference allele column\n alternate_allele (Column): Alternate allele column\n\n Returns:\n A boolean column indicating if the alleles are palindromic.\n\n Examples:\n >>> d = [{\"reference\": 'A', \"alternate\": 'T'}, {\"reference\": 'AT', \"alternate\": 'AG'}, {\"reference\": 'AT', \"alternate\": 'AT'}, {\"reference\": 'CATATG', \"alternate\": 'CATATG'}, {\"reference\": '-', \"alternate\": None}]\n >>> df = spark.createDataFrame(d)\n >>> df.withColumn(\"is_palindromic\", StudyLocusGWASCatalog._are_alleles_palindromic(f.col(\"reference\"), f.col(\"alternate\"))).show()\n +---------+---------+--------------+\n |alternate|reference|is_palindromic|\n +---------+---------+--------------+\n | T| A| true|\n | AG| AT| false|\n | AT| AT| true|\n | CATATG| CATATG| true|\n | null| -| false|\n +---------+---------+--------------+\n <BLANKLINE>\n\n \"\"\"\n revcomp = StudyLocusGWASCatalog._get_reverse_complement(alternate_allele)\n return (\n f.when(reference_allele == revcomp, True)\n .when(revcomp.isNull(), False)\n .otherwise(False)\n )\n\n @staticmethod\n def _harmonise_beta(\n risk_allele: Column,\n reference_allele: Column,\n alternate_allele: Column,\n effect_size: Column,\n confidence_interval: Column,\n ) -> Column:\n\"\"\"A function to extract the beta value from the effect size and confidence interval.\n\n If the confidence interval contains the word \"increase\" or \"decrease\" it indicates, we are dealing with betas.\n If it's \"increase\" and the effect size needs to be harmonized, then multiply the effect size by -1\n\n Args:\n risk_allele (Column): Risk allele column\n reference_allele (Column): Reference allele column\n alternate_allele (Column): Alternate allele column\n effect_size (Column): GWAS Catalog effect size column\n confidence_interval (Column): GWAS Catalog confidence interval column\n\n Returns:\n A column containing the beta value.\n \"\"\"\n return (\n f.when(\n StudyLocusGWASCatalog._are_alleles_palindromic(\n reference_allele, alternate_allele\n ),\n None,\n )\n .when(\n (\n StudyLocusGWASCatalog._effect_needs_harmonisation(\n risk_allele, reference_allele\n )\n & confidence_interval.contains(\"increase\")\n )\n | (\n ~StudyLocusGWASCatalog._effect_needs_harmonisation(\n risk_allele, reference_allele\n )\n & confidence_interval.contains(\"decrease\")\n ),\n -effect_size,\n )\n .otherwise(effect_size)\n .cast(DoubleType())\n )\n\n @staticmethod\n def _harmonise_beta_ci(\n risk_allele: Column,\n reference_allele: Column,\n alternate_allele: Column,\n effect_size: Column,\n confidence_interval: Column,\n p_value: Column,\n direction: str,\n ) -> Column:\n\"\"\"Calculating confidence intervals for beta values.\n\n Args:\n risk_allele (Column): Risk allele column\n reference_allele (Column): Reference allele column\n alternate_allele (Column): Alternate allele column\n effect_size (Column): GWAS Catalog effect size column\n confidence_interval (Column): GWAS Catalog confidence interval column\n p_value (Column): GWAS Catalog p-value column\n direction (str): This is the direction of the confidence interval. It can be either \"upper\" or \"lower\".\n\n Returns:\n The upper and lower bounds of the confidence interval for the beta coefficient.\n \"\"\"\n zscore_95 = f.lit(1.96)\n beta = StudyLocusGWASCatalog._harmonise_beta(\n risk_allele,\n reference_allele,\n alternate_allele,\n effect_size,\n confidence_interval,\n )\n zscore = pvalue_to_zscore(p_value)\n return (\n f.when(f.lit(direction) == \"upper\", beta + f.abs(zscore_95 * beta) / zscore)\n .when(f.lit(direction) == \"lower\", beta - f.abs(zscore_95 * beta) / zscore)\n .otherwise(None)\n )\n\n @staticmethod\n def _harmonise_odds_ratio(\n risk_allele: Column,\n reference_allele: Column,\n alternate_allele: Column,\n effect_size: Column,\n confidence_interval: Column,\n ) -> Column:\n\"\"\"Harmonizing odds ratio.\n\n Args:\n risk_allele (Column): Risk allele column\n reference_allele (Column): Reference allele column\n alternate_allele (Column): Alternate allele column\n effect_size (Column): GWAS Catalog effect size column\n confidence_interval (Column): GWAS Catalog confidence interval column\n\n Returns:\n A column with the odds ratio, or 1/odds_ratio if harmonization required.\n \"\"\"\n return (\n f.when(\n StudyLocusGWASCatalog._are_alleles_palindromic(\n reference_allele, alternate_allele\n ),\n None,\n )\n .when(\n (\n StudyLocusGWASCatalog._effect_needs_harmonisation(\n risk_allele, reference_allele\n )\n & ~confidence_interval.rlike(\"|\".join([\"decrease\", \"increase\"]))\n ),\n 1 / effect_size,\n )\n .otherwise(effect_size)\n .cast(DoubleType())\n )\n\n @staticmethod\n def _harmonise_odds_ratio_ci(\n risk_allele: Column,\n reference_allele: Column,\n alternate_allele: Column,\n effect_size: Column,\n confidence_interval: Column,\n p_value: Column,\n direction: str,\n ) -> Column:\n\"\"\"Calculating confidence intervals for beta values.\n\n Args:\n risk_allele (Column): Risk allele column\n reference_allele (Column): Reference allele column\n alternate_allele (Column): Alternate allele column\n effect_size (Column): GWAS Catalog effect size column\n confidence_interval (Column): GWAS Catalog confidence interval column\n p_value (Column): GWAS Catalog p-value column\n direction (str): This is the direction of the confidence interval. It can be either \"upper\" or \"lower\".\n\n Returns:\n The upper and lower bounds of the 95% confidence interval for the odds ratio.\n \"\"\"\n zscore_95 = f.lit(1.96)\n odds_ratio = StudyLocusGWASCatalog._harmonise_odds_ratio(\n risk_allele,\n reference_allele,\n alternate_allele,\n effect_size,\n confidence_interval,\n )\n odds_ratio_estimate = f.log(odds_ratio)\n zscore = pvalue_to_zscore(p_value)\n odds_ratio_se = odds_ratio_estimate / zscore\n return f.when(\n f.lit(direction) == \"upper\",\n f.exp(odds_ratio_estimate + f.abs(zscore_95 * odds_ratio_se)),\n ).when(\n f.lit(direction) == \"lower\",\n f.exp(odds_ratio_estimate - f.abs(zscore_95 * odds_ratio_se)),\n )\n\n @staticmethod\n def _concatenate_substudy_description(\n association_trait: Column, pvalue_text: Column, mapped_trait_uri: Column\n ) -> Column:\n\"\"\"Substudy description parsing. Complex string containing metadata about the substudy (e.g. QTL, specific EFO, etc.).\n\n Args:\n association_trait (Column): GWAS Catalog association trait column\n pvalue_text (Column): GWAS Catalog p-value text column\n mapped_trait_uri (Column): GWAS Catalog mapped trait URI column\n\n Returns:\n A column with the substudy description in the shape trait|pvaluetext1_pvaluetext2|EFO1_EFO2.\n\n Examples:\n >>> df = spark.createDataFrame([\n ... (\"Height\", \"http://www.ebi.ac.uk/efo/EFO_0000001,http://www.ebi.ac.uk/efo/EFO_0000002\", \"European Ancestry\"),\n ... (\"Schizophrenia\", \"http://www.ebi.ac.uk/efo/MONDO_0005090\", None)],\n ... [\"association_trait\", \"mapped_trait_uri\", \"pvalue_text\"]\n ... )\n >>> df.withColumn('substudy_description', StudyLocusGWASCatalog._concatenate_substudy_description(df.association_trait, df.pvalue_text, df.mapped_trait_uri)).show(truncate=False)\n +-----------------+-------------------------------------------------------------------------+-----------------+------------------------------------------+\n |association_trait|mapped_trait_uri |pvalue_text |substudy_description |\n +-----------------+-------------------------------------------------------------------------+-----------------+------------------------------------------+\n |Height |http://www.ebi.ac.uk/efo/EFO_0000001,http://www.ebi.ac.uk/efo/EFO_0000002|European Ancestry|Height|EA|EFO_0000001/EFO_0000002 |\n |Schizophrenia |http://www.ebi.ac.uk/efo/MONDO_0005090 |null |Schizophrenia|no_pvalue_text|MONDO_0005090|\n +-----------------+-------------------------------------------------------------------------+-----------------+------------------------------------------+\n <BLANKLINE>\n \"\"\"\n p_value_text = f.coalesce(\n StudyLocusGWASCatalog._normalise_pvaluetext(pvalue_text),\n f.array(f.lit(\"no_pvalue_text\")),\n )\n return f.concat_ws(\n \"|\",\n association_trait,\n f.concat_ws(\n \"/\",\n p_value_text,\n ),\n f.concat_ws(\n \"/\",\n parse_efos(mapped_trait_uri),\n ),\n )\n\n @staticmethod\n def _qc_all(\n qc: Column,\n chromosome: Column,\n position: Column,\n reference_allele: Column,\n alternate_allele: Column,\n strongest_snp_risk_allele: Column,\n p_value_mantissa: Column,\n p_value_exponent: Column,\n p_value_cutoff: float,\n ) -> Column:\n\"\"\"Flag associations that fail any QC.\n\n Args:\n qc (Column): QC column\n chromosome (Column): Chromosome column\n position (Column): Position column\n reference_allele (Column): Reference allele column\n alternate_allele (Column): Alternate allele column\n strongest_snp_risk_allele (Column): Strongest SNP risk allele column\n p_value_mantissa (Column): P-value mantissa column\n p_value_exponent (Column): P-value exponent column\n p_value_cutoff (float): P-value cutoff\n\n Returns:\n Column: Updated QC column with flag.\n \"\"\"\n qc = StudyLocusGWASCatalog._qc_variant_interactions(\n qc, strongest_snp_risk_allele\n )\n qc = StudyLocusGWASCatalog._qc_subsignificant_associations(\n qc, p_value_mantissa, p_value_exponent, p_value_cutoff\n )\n qc = StudyLocusGWASCatalog._qc_genomic_location(qc, chromosome, position)\n qc = StudyLocusGWASCatalog._qc_variant_inconsistencies(\n qc, chromosome, position, strongest_snp_risk_allele\n )\n qc = StudyLocusGWASCatalog._qc_unmapped_variants(qc, alternate_allele)\n qc = StudyLocusGWASCatalog._qc_palindromic_alleles(\n qc, reference_allele, alternate_allele\n )\n return qc\n\n @staticmethod\n def _qc_variant_interactions(\n qc: Column, strongest_snp_risk_allele: Column\n ) -> Column:\n\"\"\"Flag associations based on variant x variant interactions.\n\n Args:\n qc (Column): QC column\n strongest_snp_risk_allele (Column): Column with the strongest SNP risk allele\n\n Returns:\n Column: Updated QC column with flag.\n \"\"\"\n return StudyLocusGWASCatalog._update_quality_flag(\n qc,\n strongest_snp_risk_allele.contains(\";\"),\n StudyLocusQualityCheck.COMPOSITE_FLAG,\n )\n\n @staticmethod\n def _qc_subsignificant_associations(\n qc: Column,\n p_value_mantissa: Column,\n p_value_exponent: Column,\n pvalue_cutoff: float,\n ) -> Column:\n\"\"\"Flag associations below significant threshold.\n\n Args:\n qc (Column): QC column\n p_value_mantissa (Column): P-value mantissa column\n p_value_exponent (Column): P-value exponent column\n pvalue_cutoff (float): association p-value cut-off\n\n Returns:\n Column: Updated QC column with flag.\n\n Examples:\n >>> import pyspark.sql.types as t\n >>> d = [{'qc': None, 'p_value_mantissa': 1, 'p_value_exponent': -7}, {'qc': None, 'p_value_mantissa': 1, 'p_value_exponent': -8}, {'qc': None, 'p_value_mantissa': 5, 'p_value_exponent': -8}, {'qc': None, 'p_value_mantissa': 1, 'p_value_exponent': -9}]\n >>> df = spark.createDataFrame(d, t.StructType([t.StructField('qc', t.ArrayType(t.StringType()), True), t.StructField('p_value_mantissa', t.IntegerType()), t.StructField('p_value_exponent', t.IntegerType())]))\n >>> df.withColumn('qc', StudyLocusGWASCatalog._qc_subsignificant_associations(f.col(\"qc\"), f.col(\"p_value_mantissa\"), f.col(\"p_value_exponent\"), 5e-8)).show(truncate = False)\n +------------------------+----------------+----------------+\n |qc |p_value_mantissa|p_value_exponent|\n +------------------------+----------------+----------------+\n |[Subsignificant p-value]|1 |-7 |\n |[] |1 |-8 |\n |[] |5 |-8 |\n |[] |1 |-9 |\n +------------------------+----------------+----------------+\n <BLANKLINE>\n\n \"\"\"\n return StudyLocus._update_quality_flag(\n qc,\n calculate_neglog_pvalue(p_value_mantissa, p_value_exponent)\n < f.lit(-np.log10(pvalue_cutoff)),\n StudyLocusQualityCheck.SUBSIGNIFICANT_FLAG,\n )\n\n @staticmethod\n def _qc_genomic_location(\n qc: Column, chromosome: Column, position: Column\n ) -> Column:\n\"\"\"Flag associations without genomic location in GWAS Catalog.\n\n Args:\n qc (Column): QC column\n chromosome (Column): Chromosome column in GWAS Catalog\n position (Column): Position column in GWAS Catalog\n\n Returns:\n Column: Updated QC column with flag.\n\n Examples:\n >>> import pyspark.sql.types as t\n >>> d = [{'qc': None, 'chromosome': None, 'position': None}, {'qc': None, 'chromosome': '1', 'position': None}, {'qc': None, 'chromosome': None, 'position': 1}, {'qc': None, 'chromosome': '1', 'position': 1}]\n >>> df = spark.createDataFrame(d, schema=t.StructType([t.StructField('qc', t.ArrayType(t.StringType()), True), t.StructField('chromosome', t.StringType()), t.StructField('position', t.IntegerType())]))\n >>> df.withColumn('qc', StudyLocusGWASCatalog._qc_genomic_location(df.qc, df.chromosome, df.position)).show(truncate=False)\n +----------------------------+----------+--------+\n |qc |chromosome|position|\n +----------------------------+----------+--------+\n |[Incomplete genomic mapping]|null |null |\n |[Incomplete genomic mapping]|1 |null |\n |[Incomplete genomic mapping]|null |1 |\n |[] |1 |1 |\n +----------------------------+----------+--------+\n <BLANKLINE>\n\n \"\"\"\n return StudyLocus._update_quality_flag(\n qc,\n position.isNull() | chromosome.isNull(),\n StudyLocusQualityCheck.NO_GENOMIC_LOCATION_FLAG,\n )\n\n @staticmethod\n def _qc_variant_inconsistencies(\n qc: Column,\n chromosome: Column,\n position: Column,\n strongest_snp_risk_allele: Column,\n ) -> Column:\n\"\"\"Flag associations with inconsistencies in the variant annotation.\n\n Args:\n qc (Column): QC column\n chromosome (Column): Chromosome column in GWAS Catalog\n position (Column): Position column in GWAS Catalog\n strongest_snp_risk_allele (Column): Strongest SNP risk allele column in GWAS Catalog\n\n Returns:\n Column: Updated QC column with flag.\n \"\"\"\n return StudyLocusGWASCatalog._update_quality_flag(\n qc,\n # Number of chromosomes does not correspond to the number of positions:\n (f.size(f.split(chromosome, \";\")) != f.size(f.split(position, \";\")))\n # Number of chromosome values different from riskAllele values:\n | (\n f.size(f.split(chromosome, \";\"))\n != f.size(f.split(strongest_snp_risk_allele, \";\"))\n ),\n StudyLocusQualityCheck.INCONSISTENCY_FLAG,\n )\n\n @staticmethod\n def _qc_unmapped_variants(qc: Column, alternate_allele: Column) -> Column:\n\"\"\"Flag associations with variants not mapped to variantAnnotation.\n\n Args:\n qc (Column): QC column\n alternate_allele (Column): alternate allele\n\n Returns:\n Column: Updated QC column with flag.\n\n Example:\n >>> import pyspark.sql.types as t\n >>> d = [{'alternate_allele': 'A', 'qc': None}, {'alternate_allele': None, 'qc': None}]\n >>> schema = t.StructType([t.StructField('alternate_allele', t.StringType(), True), t.StructField('qc', t.ArrayType(t.StringType()), True)])\n >>> df = spark.createDataFrame(data=d, schema=schema)\n >>> df.withColumn(\"new_qc\", StudyLocusGWASCatalog._qc_unmapped_variants(f.col(\"qc\"), f.col(\"alternate_allele\"))).show()\n +----------------+----+--------------------+\n |alternate_allele| qc| new_qc|\n +----------------+----+--------------------+\n | A|null| []|\n | null|null|[No mapping in Gn...|\n +----------------+----+--------------------+\n <BLANKLINE>\n\n \"\"\"\n return StudyLocus._update_quality_flag(\n qc,\n alternate_allele.isNull(),\n StudyLocusQualityCheck.NON_MAPPED_VARIANT_FLAG,\n )\n\n @staticmethod\n def _qc_palindromic_alleles(\n qc: Column, reference_allele: Column, alternate_allele: Column\n ) -> Column:\n\"\"\"Flag associations with palindromic variants which effects can not be harmonised.\n\n Args:\n qc (Column): QC column\n reference_allele (Column): reference allele\n alternate_allele (Column): alternate allele\n\n Returns:\n Column: Updated QC column with flag.\n\n Example:\n >>> import pyspark.sql.types as t\n >>> schema = t.StructType([t.StructField('reference_allele', t.StringType(), True), t.StructField('alternate_allele', t.StringType(), True), t.StructField('qc', t.ArrayType(t.StringType()), True)])\n >>> d = [{'reference_allele': 'A', 'alternate_allele': 'T', 'qc': None}, {'reference_allele': 'AT', 'alternate_allele': 'TA', 'qc': None}, {'reference_allele': 'AT', 'alternate_allele': 'AT', 'qc': None}]\n >>> df = spark.createDataFrame(data=d, schema=schema)\n >>> df.withColumn(\"qc\", StudyLocusGWASCatalog._qc_palindromic_alleles(f.col(\"qc\"), f.col(\"reference_allele\"), f.col(\"alternate_allele\"))).show(truncate=False)\n +----------------+----------------+---------------------------------------+\n |reference_allele|alternate_allele|qc |\n +----------------+----------------+---------------------------------------+\n |A |T |[Palindrome alleles - cannot harmonize]|\n |AT |TA |[] |\n |AT |AT |[Palindrome alleles - cannot harmonize]|\n +----------------+----------------+---------------------------------------+\n <BLANKLINE>\n\n \"\"\"\n return StudyLocus._update_quality_flag(\n qc,\n StudyLocusGWASCatalog._are_alleles_palindromic(\n reference_allele, alternate_allele\n ),\n StudyLocusQualityCheck.PALINDROMIC_ALLELE_FLAG,\n )\n\n @classmethod\n def from_source(\n cls: type[StudyLocusGWASCatalog],\n gwas_associations: DataFrame,\n variant_annotation: VariantAnnotation,\n pvalue_threshold: float = 5e-8,\n ) -> StudyLocusGWASCatalog:\n\"\"\"Read GWASCatalog associations.\n\n It reads the GWAS Catalog association dataset, selects and renames columns, casts columns, and\n applies some pre-defined filters on the data:\n\n Args:\n gwas_associations (DataFrame): GWAS Catalog raw associations dataset\n variant_annotation (VariantAnnotation): Variant annotation dataset\n pvalue_threshold (float): P-value threshold for flagging associations\n\n Returns:\n StudyLocusGWASCatalog: StudyLocusGWASCatalog dataset\n \"\"\"\n return cls(\n _df=gwas_associations.withColumn(\n \"studyLocusId\", f.monotonically_increasing_id().cast(LongType())\n )\n .transform(\n # Map/harmonise variants to variant annotation dataset:\n # This function adds columns: variantId, referenceAllele, alternateAllele, chromosome, position\n lambda df: StudyLocusGWASCatalog._map_to_variant_annotation_variants(\n df, variant_annotation\n )\n )\n .withColumn(\n # Perform all quality control checks:\n \"qualityControls\",\n StudyLocusGWASCatalog._qc_all(\n f.array().alias(\"qualityControls\"),\n f.col(\"CHR_ID\"),\n f.col(\"CHR_POS\").cast(IntegerType()),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"STRONGEST SNP-RISK ALLELE\"),\n *StudyLocusGWASCatalog._parse_pvalue(f.col(\"P-VALUE\")),\n pvalue_threshold,\n ),\n )\n .select(\n # INSIDE STUDY-LOCUS SCHEMA:\n \"studyLocusId\",\n \"variantId\",\n # Mapped genomic location of the variant (; separated list)\n \"chromosome\",\n \"position\",\n f.col(\"STUDY ACCESSION\").alias(\"studyId\"),\n # beta value of the association\n StudyLocusGWASCatalog._harmonise_beta(\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"OR or BETA\"),\n f.col(\"95% CI (TEXT)\"),\n ).alias(\"beta\"),\n # odds ratio of the association\n StudyLocusGWASCatalog._harmonise_odds_ratio(\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"OR or BETA\"),\n f.col(\"95% CI (TEXT)\"),\n ).alias(\"oddsRatio\"),\n # CI lower of the beta value\n StudyLocusGWASCatalog._harmonise_beta_ci(\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"OR or BETA\"),\n f.col(\"95% CI (TEXT)\"),\n f.col(\"P-VALUE\"),\n \"lower\",\n ).alias(\"betaConfidenceIntervalLower\"),\n # CI upper for the beta value\n StudyLocusGWASCatalog._harmonise_beta_ci(\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"OR or BETA\"),\n f.col(\"95% CI (TEXT)\"),\n f.col(\"P-VALUE\"),\n \"upper\",\n ).alias(\"betaConfidenceIntervalUpper\"),\n # CI lower of the odds ratio value\n StudyLocusGWASCatalog._harmonise_odds_ratio_ci(\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"OR or BETA\"),\n f.col(\"95% CI (TEXT)\"),\n f.col(\"P-VALUE\"),\n \"lower\",\n ).alias(\"oddsRatioConfidenceIntervalLower\"),\n # CI upper of the odds ratio value\n StudyLocusGWASCatalog._harmonise_odds_ratio_ci(\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"OR or BETA\"),\n f.col(\"95% CI (TEXT)\"),\n f.col(\"P-VALUE\"),\n \"upper\",\n ).alias(\"oddsRatioConfidenceIntervalUpper\"),\n # p-value of the association, string: split into exponent and mantissa.\n *StudyLocusGWASCatalog._parse_pvalue(f.col(\"P-VALUE\")),\n # Capturing phenotype granularity at the association level\n StudyLocusGWASCatalog._concatenate_substudy_description(\n f.col(\"DISEASE/TRAIT\"),\n f.col(\"P-VALUE (TEXT)\"),\n f.col(\"MAPPED_TRAIT_URI\"),\n ).alias(\"subStudyDescription\"),\n # Quality controls (array of strings)\n \"qualityControls\",\n )\n )\n\n def update_study_id(\n self: StudyLocusGWASCatalog, study_annotation: DataFrame\n ) -> StudyLocusGWASCatalog:\n\"\"\"Update studyId with a dataframe containing study.\n\n Args:\n study_annotation (DataFrame): Dataframe containing `updatedStudyId` and key columns `studyId` and `subStudyDescription`.\n\n Returns:\n StudyLocusGWASCatalog: Updated study locus.\n \"\"\"\n self.df = (\n self._df.join(\n study_annotation, on=[\"studyId\", \"subStudyDescription\"], how=\"left\"\n )\n .withColumn(\"studyId\", f.coalesce(\"updatedStudyId\", \"studyId\"))\n .drop(\"subStudyDescription\", \"updatedStudyId\")\n )\n return self\n\n def annotate_ld(\n self: StudyLocusGWASCatalog,\n session: Session,\n studies: StudyIndexGWASCatalog,\n ld_populations: list[str],\n ld_index_template: str,\n ld_matrix_template: str,\n min_r2: float,\n ) -> StudyLocus:\n\"\"\"Annotate LD set for every studyLocus using gnomAD.\n\n Args:\n session (Session): Session\n studies (StudyIndexGWASCatalog): Study index containing ancestry information\n ld_populations (list[str]): List of populations to annotate\n ld_index_template (str): Template path of the LD matrix index containing `{POP}` where the population is expected\n ld_matrix_template (str): Template path of the LD matrix containing `{POP}` where the population is expected\n min_r2 (float): Minimum r2 to include in the LD set\n\n Returns:\n StudyLocus: Study-locus with an annotated credible set.\n \"\"\"\n # TODO: call unique_study_locus_ancestries here so that it is not duplicated with ld_annotation_by_locus_ancestry\n # LD annotation for all unique lead variants in all populations (study independent).\n ld_r = LDAnnotatorGnomad.ld_annotation_by_locus_ancestry(\n session,\n self,\n studies,\n ld_populations,\n ld_index_template,\n ld_matrix_template,\n min_r2,\n ).coalesce(400)\n\n ld_set = (\n self.unique_study_locus_ancestries(studies)\n .join(ld_r, on=[\"chromosome\", \"variantId\", \"gnomadPopulation\"], how=\"left\")\n .withColumn(\"r2\", f.pow(f.col(\"r\"), f.lit(2)))\n .withColumn(\n \"r2Overall\",\n LDAnnotatorGnomad.weighted_r_overall(\n f.col(\"chromosome\"),\n f.col(\"studyId\"),\n f.col(\"variantId\"),\n f.col(\"tagVariantId\"),\n f.col(\"relativeSampleSize\"),\n f.col(\"r2\"),\n ),\n )\n .groupBy(\"chromosome\", \"studyId\", \"variantId\")\n .agg(\n f.collect_set(\n f.when(\n f.col(\"tagVariantId\").isNotNull(),\n f.struct(\"tagVariantId\", \"r2Overall\"),\n )\n ).alias(\"credibleSet\")\n )\n )\n\n self.df = self.df.join(\n ld_set, on=[\"chromosome\", \"studyId\", \"variantId\"], how=\"left\"\n )\n\n return self._qc_unresolved_ld()\n\n def _qc_ambiguous_study(self: StudyLocusGWASCatalog) -> StudyLocusGWASCatalog:\n\"\"\"Flag associations with variants that can not be unambiguously associated with one study.\n\n Returns:\n StudyLocusGWASCatalog: Updated study locus.\n \"\"\"\n assoc_ambiguity_window = Window.partitionBy(\n f.col(\"studyId\"), f.col(\"variantId\")\n )\n\n self._df.withColumn(\n \"qualityControls\",\n StudyLocus._update_quality_flag(\n f.col(\"qualityControls\"),\n f.count(f.col(\"variantId\")).over(assoc_ambiguity_window) > 1,\n StudyLocusQualityCheck.AMBIGUOUS_STUDY,\n ),\n )\n return self\n\n def _qc_unresolved_ld(self: StudyLocusGWASCatalog) -> StudyLocusGWASCatalog:\n\"\"\"Flag associations with variants that are not found in the LD reference.\n\n Returns:\n StudyLocusGWASCatalog: Updated study locus.\n \"\"\"\n self._df.withColumn(\n \"qualityControls\",\n StudyLocus._update_quality_flag(\n f.col(\"qualityControls\"),\n f.col(\"credibleSet\").isNull(),\n StudyLocusQualityCheck.UNRESOLVED_LD,\n ),\n )\n return self\n
Annotate LD set for every studyLocus using gnomAD.
Parameters:
Name Type Description Default sessionSession
Session
required studiesStudyIndexGWASCatalog
Study index containing ancestry information
required ld_populationslist[str]
List of populations to annotate
required ld_index_templatestr
Template path of the LD matrix index containing {POP} where the population is expected
required ld_matrix_templatestr
Template path of the LD matrix containing {POP} where the population is expected
required min_r2float
Minimum r2 to include in the LD set
required
Returns:
Name Type Description StudyLocusStudyLocus
Study-locus with an annotated credible set.
Source code in src/otg/dataset/study_locus.py
def annotate_ld(\n self: StudyLocusGWASCatalog,\n session: Session,\n studies: StudyIndexGWASCatalog,\n ld_populations: list[str],\n ld_index_template: str,\n ld_matrix_template: str,\n min_r2: float,\n) -> StudyLocus:\n\"\"\"Annotate LD set for every studyLocus using gnomAD.\n\n Args:\n session (Session): Session\n studies (StudyIndexGWASCatalog): Study index containing ancestry information\n ld_populations (list[str]): List of populations to annotate\n ld_index_template (str): Template path of the LD matrix index containing `{POP}` where the population is expected\n ld_matrix_template (str): Template path of the LD matrix containing `{POP}` where the population is expected\n min_r2 (float): Minimum r2 to include in the LD set\n\n Returns:\n StudyLocus: Study-locus with an annotated credible set.\n \"\"\"\n # TODO: call unique_study_locus_ancestries here so that it is not duplicated with ld_annotation_by_locus_ancestry\n # LD annotation for all unique lead variants in all populations (study independent).\n ld_r = LDAnnotatorGnomad.ld_annotation_by_locus_ancestry(\n session,\n self,\n studies,\n ld_populations,\n ld_index_template,\n ld_matrix_template,\n min_r2,\n ).coalesce(400)\n\n ld_set = (\n self.unique_study_locus_ancestries(studies)\n .join(ld_r, on=[\"chromosome\", \"variantId\", \"gnomadPopulation\"], how=\"left\")\n .withColumn(\"r2\", f.pow(f.col(\"r\"), f.lit(2)))\n .withColumn(\n \"r2Overall\",\n LDAnnotatorGnomad.weighted_r_overall(\n f.col(\"chromosome\"),\n f.col(\"studyId\"),\n f.col(\"variantId\"),\n f.col(\"tagVariantId\"),\n f.col(\"relativeSampleSize\"),\n f.col(\"r2\"),\n ),\n )\n .groupBy(\"chromosome\", \"studyId\", \"variantId\")\n .agg(\n f.collect_set(\n f.when(\n f.col(\"tagVariantId\").isNotNull(),\n f.struct(\"tagVariantId\", \"r2Overall\"),\n )\n ).alias(\"credibleSet\")\n )\n )\n\n self.df = self.df.join(\n ld_set, on=[\"chromosome\", \"studyId\", \"variantId\"], how=\"left\"\n )\n\n return self._qc_unresolved_ld()\n
Clumping is a commonly used post-processing method that allows for identification of independent association signals from GWAS summary statistics and curated associations. This process is critical because of the complex linkage disequilibrium (LD) structure in human populations, which can result in multiple statistically significant associations within the same genomic region. Clumping methods help reduce redundancy in GWAS results and ensure that each reported association represents an independent signal.
We have implemented 2 clumping methods:
"},{"location":"components/method/clumping/#clumping-based-on-linkage-disequilibrium-ld","title":"Clumping based on Linkage Disequilibrium (LD)","text":"
LD clumping reports the most significant genetic associations in a region in terms of a smaller number of \u201cclumps\u201d of genetically linked SNPs.
Source code in src/otg/method/clump.py
class LDclumping:\n\"\"\"LD clumping reports the most significant genetic associations in a region in terms of a smaller number of \u201cclumps\u201d of genetically linked SNPs.\"\"\"\n\n @staticmethod\n def _is_lead_linked(\n study_id: Column,\n variant_id: Column,\n p_value_exponent: Column,\n p_value_mantissa: Column,\n credible_set: Column,\n ) -> Column:\n\"\"\"Evaluates whether a lead variant is linked to a tag (with lowest p-value) in the same studyLocus dataset.\n\n Args:\n study_id (Column): studyId\n variant_id (Column): Lead variant id\n p_value_exponent (Column): p-value exponent\n p_value_mantissa (Column): p-value mantissa\n credible_set (Column): Credible set <array of structs>\n\n Returns:\n Column: Boolean in which True indicates that the lead is linked to another tag in the same dataset.\n \"\"\"\n leads_in_study = f.collect_set(variant_id).over(Window.partitionBy(study_id))\n tags_in_studylocus = f.array_union(\n # Get all tag variants from the credible set per studyLocusId\n f.transform(credible_set, lambda x: x.tagVariantId),\n # And append the lead variant so that the intersection is the same for all studyLocusIds in a study\n f.array(f.col(\"variantId\")),\n )\n intersect_lead_tags = f.array_sort(\n f.array_intersect(leads_in_study, tags_in_studylocus)\n )\n return (\n # If the lead is in the credible set, we rank the peaks by p-value\n f.when(\n f.size(intersect_lead_tags) > 0,\n f.row_number().over(\n Window.partitionBy(study_id, intersect_lead_tags).orderBy(\n p_value_exponent, p_value_mantissa\n )\n )\n > 1,\n )\n # If the intersection is empty (lead is not in the credible set or cred set is empty), the association is not linked\n .otherwise(f.lit(False))\n )\n\n @classmethod\n def clump(cls: type[LDclumping], associations: StudyLocus) -> StudyLocus:\n\"\"\"Perform clumping on studyLocus dataset.\n\n Args:\n associations (StudyLocus): StudyLocus dataset\n\n Returns:\n StudyLocus: including flag and removing credibleSet information for LD clumped loci.\n \"\"\"\n return associations.clump()\n
Calculate bayesian colocalisation based on overlapping signals from credible sets.
Based on the R COLOC package, which uses the Bayes factors from the credible set to estimate the posterior probability of colocalisation. This method makes the simplifying assumption that only one single causal variant exists for any given trait in any genomic region.
Hypothesis Description H0 no association with either trait in the region H1 association with trait 1 only H2 association with trait 2 only H3 both traits are associated, but have different single causal variants H4 both traits are associated and share the same single causal variant
Approximate Bayes factors required
Coloc requires the availability of approximate Bayes factors (ABF) for each variant in the credible set (logABF column).
Source code in src/otg/method/colocalisation.py
class Coloc:\n\"\"\"Calculate bayesian colocalisation based on overlapping signals from credible sets.\n\n Based on the [R COLOC package](https://github.com/chr1swallace/coloc/blob/main/R/claudia.R), which uses the Bayes factors from the credible set to estimate the posterior probability of colocalisation. This method makes the simplifying assumption that **only one single causal variant** exists for any given trait in any genomic region.\n\n | Hypothesis | Description |\n | ------------- | --------------------------------------------------------------------- |\n | H<sub>0</sub> | no association with either trait in the region |\n | H<sub>1</sub> | association with trait 1 only |\n | H<sub>2</sub> | association with trait 2 only |\n | H<sub>3</sub> | both traits are associated, but have different single causal variants |\n | H<sub>4</sub> | both traits are associated and share the same single causal variant |\n\n !!! warning \"Approximate Bayes factors required\"\n Coloc requires the availability of approximate Bayes factors (ABF) for each variant in the credible set (`logABF` column).\n\n \"\"\"\n\n @staticmethod\n def _get_logsum(log_abf: ndarray) -> float:\n\"\"\"Calculates logsum of vector.\n\n This function calculates the log of the sum of the exponentiated\n logs taking out the max, i.e. insuring that the sum is not Inf\n\n Args:\n log_abf (ndarray): log approximate bayes factor\n\n Returns:\n float: logsum\n\n Example:\n >>> l = [0.2, 0.1, 0.05, 0]\n >>> round(Coloc._get_logsum(l), 6)\n 1.476557\n \"\"\"\n themax = np.max(log_abf)\n result = themax + np.log(np.sum(np.exp(log_abf - themax)))\n return float(result)\n\n @staticmethod\n def _get_posteriors(all_abfs: ndarray) -> DenseVector:\n\"\"\"Calculate posterior probabilities for each hypothesis.\n\n Args:\n all_abfs (ndarray): h0-h4 bayes factors\n\n Returns:\n DenseVector: Posterior\n\n Example:\n >>> l = np.array([0.2, 0.1, 0.05, 0])\n >>> Coloc._get_posteriors(l)\n DenseVector([0.279, 0.2524, 0.2401, 0.2284])\n \"\"\"\n diff = all_abfs - Coloc._get_logsum(all_abfs)\n abfs_posteriors = np.exp(diff)\n return Vectors.dense(abfs_posteriors)\n\n @classmethod\n def colocalise(\n cls: type[Coloc],\n overlapping_signals: StudyLocusOverlap,\n priorc1: float = 1e-4,\n priorc2: float = 1e-4,\n priorc12: float = 1e-5,\n ) -> Colocalisation:\n\"\"\"Calculate bayesian colocalisation based on overlapping signals.\n\n Args:\n overlapping_signals (StudyLocusOverlap): overlapping peaks\n priorc1 (float): Prior on variant being causal for trait 1. Defaults to 1e-4.\n priorc2 (float): Prior on variant being causal for trait 2. Defaults to 1e-4.\n priorc12 (float): Prior on variant being causal for traits 1 and 2. Defaults to 1e-5.\n\n Returns:\n Colocalisation: Colocalisation results\n \"\"\"\n # register udfs\n logsum = f.udf(Coloc._get_logsum, DoubleType())\n posteriors = f.udf(Coloc._get_posteriors, VectorUDT())\n return Colocalisation(\n _df=(\n overlapping_signals.df\n # Before summing log_abf columns nulls need to be filled with 0:\n .fillna(0, subset=[\"left_logABF\", \"right_logABF\"])\n # Sum of log_abfs for each pair of signals\n .withColumn(\"sum_log_abf\", f.col(\"left_logABF\") + f.col(\"right_logABF\"))\n # Group by overlapping peak and generating dense vectors of log_abf:\n .groupBy(\"chromosome\", \"left_studyLocusId\", \"right_studyLocusId\")\n .agg(\n f.count(\"*\").alias(\"coloc_n_vars\"),\n fml.array_to_vector(f.collect_list(f.col(\"left_logABF\"))).alias(\n \"left_logABF\"\n ),\n fml.array_to_vector(f.collect_list(f.col(\"right_logABF\"))).alias(\n \"right_logABF\"\n ),\n fml.array_to_vector(f.collect_list(f.col(\"sum_log_abf\"))).alias(\n \"sum_log_abf\"\n ),\n )\n .withColumn(\"logsum1\", logsum(f.col(\"left_logABF\")))\n .withColumn(\"logsum2\", logsum(f.col(\"right_logABF\")))\n .withColumn(\"logsum12\", logsum(f.col(\"sum_log_abf\")))\n .drop(\"left_logABF\", \"right_logABF\", \"sum_log_abf\")\n # Add priors\n # priorc1 Prior on variant being causal for trait 1\n .withColumn(\"priorc1\", f.lit(priorc1))\n # priorc2 Prior on variant being causal for trait 2\n .withColumn(\"priorc2\", f.lit(priorc2))\n # priorc12 Prior on variant being causal for traits 1 and 2\n .withColumn(\"priorc12\", f.lit(priorc12))\n # h0-h2\n .withColumn(\"lH0abf\", f.lit(0))\n .withColumn(\"lH1abf\", f.log(f.col(\"priorc1\")) + f.col(\"logsum1\"))\n .withColumn(\"lH2abf\", f.log(f.col(\"priorc2\")) + f.col(\"logsum2\"))\n # h3\n .withColumn(\"sumlogsum\", f.col(\"logsum1\") + f.col(\"logsum2\"))\n # exclude null H3/H4s: due to sumlogsum == logsum12\n .filter(f.col(\"sumlogsum\") != f.col(\"logsum12\"))\n .withColumn(\"max\", f.greatest(\"sumlogsum\", \"logsum12\"))\n .withColumn(\n \"logdiff\",\n (\n f.col(\"max\")\n + f.log(\n f.exp(f.col(\"sumlogsum\") - f.col(\"max\"))\n - f.exp(f.col(\"logsum12\") - f.col(\"max\"))\n )\n ),\n )\n .withColumn(\n \"lH3abf\",\n f.log(f.col(\"priorc1\"))\n + f.log(f.col(\"priorc2\"))\n + f.col(\"logdiff\"),\n )\n .drop(\"right_logsum\", \"left_logsum\", \"sumlogsum\", \"max\", \"logdiff\")\n # h4\n .withColumn(\"lH4abf\", f.log(f.col(\"priorc12\")) + f.col(\"logsum12\"))\n # cleaning\n .drop(\n \"priorc1\", \"priorc2\", \"priorc12\", \"logsum1\", \"logsum2\", \"logsum12\"\n )\n # posteriors\n .withColumn(\n \"allABF\",\n fml.array_to_vector(\n f.array(\n f.col(\"lH0abf\"),\n f.col(\"lH1abf\"),\n f.col(\"lH2abf\"),\n f.col(\"lH3abf\"),\n f.col(\"lH4abf\"),\n )\n ),\n )\n .withColumn(\n \"posteriors\", fml.vector_to_array(posteriors(f.col(\"allABF\")))\n )\n .withColumn(\"coloc_h0\", f.col(\"posteriors\").getItem(0))\n .withColumn(\"coloc_h1\", f.col(\"posteriors\").getItem(1))\n .withColumn(\"coloc_h2\", f.col(\"posteriors\").getItem(2))\n .withColumn(\"coloc_h3\", f.col(\"posteriors\").getItem(3))\n .withColumn(\"coloc_h4\", f.col(\"posteriors\").getItem(4))\n .withColumn(\"coloc_h4_h3\", f.col(\"coloc_h4\") / f.col(\"coloc_h3\"))\n .withColumn(\"coloc_log2_h4_h3\", f.log2(f.col(\"coloc_h4_h3\")))\n # clean up\n .drop(\n \"posteriors\",\n \"allABF\",\n \"coloc_h4_h3\",\n \"lH0abf\",\n \"lH1abf\",\n \"lH2abf\",\n \"lH3abf\",\n \"lH4abf\",\n )\n .withColumn(\"colocalisationMethod\", f.lit(\"COLOC\"))\n )\n )\n
It extends CAVIAR\u00a0framework to explicitly estimate the posterior probability that the same variant is causal in 2 studies while accounting for the uncertainty of LD. eCAVIAR computes the colocalization posterior probability (CLPP) by utilizing the marginal posterior probabilities. This framework allows for multiple variants to be causal in a single locus.
Source code in src/otg/method/colocalisation.py
class ECaviar:\n\"\"\"ECaviar-based colocalisation analysis.\n\n It extends [CAVIAR](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5142122/#bib18)\u00a0framework to explicitly estimate the posterior probability that the same variant is causal in 2 studies while accounting for the uncertainty of LD. eCAVIAR computes the colocalization posterior probability (**CLPP**) by utilizing the marginal posterior probabilities. This framework allows for **multiple variants to be causal** in a single locus.\n \"\"\"\n\n @staticmethod\n def _get_clpp(left_pp: Column, right_pp: Column) -> Column:\n\"\"\"Calculate the colocalisation posterior probability (CLPP).\n\n If the fact that the same variant is found causal for two studies are independent events,\n CLPP is defined as the product of posterior porbabilities that a variant is causal in both studies.\n\n Args:\n left_pp (Column): left posterior probability\n right_pp (Column): right posterior probability\n\n Returns:\n Column: CLPP\n\n Examples:\n >>> d = [{\"left_pp\": 0.5, \"right_pp\": 0.5}, {\"left_pp\": 0.25, \"right_pp\": 0.75}]\n >>> df = spark.createDataFrame(d)\n >>> df.withColumn(\"clpp\", ECaviar._get_clpp(f.col(\"left_pp\"), f.col(\"right_pp\"))).show()\n +-------+--------+------+\n |left_pp|right_pp| clpp|\n +-------+--------+------+\n | 0.5| 0.5| 0.25|\n | 0.25| 0.75|0.1875|\n +-------+--------+------+\n <BLANKLINE>\n\n \"\"\"\n return left_pp * right_pp\n\n @classmethod\n def colocalise(\n cls: type[ECaviar], overlapping_signals: StudyLocusOverlap\n ) -> Colocalisation:\n\"\"\"Calculate bayesian colocalisation based on overlapping signals.\n\n Args:\n overlapping_signals (StudyLocusOverlap): overlapping signals.\n\n Returns:\n Colocalisation: colocalisation results based on eCAVIAR.\n \"\"\"\n return Colocalisation(\n _df=(\n overlapping_signals.df.withColumn(\n \"clpp\",\n ECaviar._get_clpp(\n f.col(\"left_posteriorProbability\"),\n f.col(\"right_posteriorProbability\"),\n ),\n )\n .groupBy(\"left_studyLocusId\", \"right_studyLocusId\", \"chromosome\")\n .agg(\n f.count(\"*\").alias(\"coloc_n_vars\"),\n f.sum(f.col(\"clpp\")).alias(\"clpp\"),\n )\n .withColumn(\"colocalisationMethod\", f.lit(\"eCAVIAR\"))\n )\n )\n
@classmethod\ndef ld_annotation_by_locus_ancestry(\n cls: type[LDAnnotatorGnomad],\n session: Session,\n associations: StudyLocusGWASCatalog,\n studies: StudyIndexGWASCatalog,\n ld_populations: list[str],\n ld_index_template: str,\n ld_matrix_template: str,\n min_r2: float,\n) -> DataFrame:\n\"\"\"LD information for all locus and ancestries.\n\n Args:\n session (Session): Session\n associations (StudyLocusGWASCatalog): GWAS associations\n studies (StudyIndexGWASCatalog): study metadata of the associations\n ld_populations (list[str]): List of populations to annotate\n ld_index_template (str): Template path of the LD matrix index containing `{POP}` where the population is expected\n ld_matrix_template (str): Template path of the LD matrix containing `{POP}` where the population is expected\n min_r2 (float): minimum r2 to keep\n\n Returns:\n DataFrame: LD annotation [\"variantId\", \"chromosome\", \"gnomadPopulation\", \"tagVariantId\", \"r\"]\n \"\"\"\n # Unique lead - population pairs:\n locus_ancestry = (\n associations.unique_study_locus_ancestries(studies)\n # Ignoring study information / relativeSampleSize to get unique lead-ancestry pairs\n .drop(\"studyId\", \"relativeSampleSize\")\n .distinct()\n .persist()\n )\n\n # All gnomad populations captured in associations:\n assoc_populations = locus_ancestry.rdd.map(\n lambda x: x.gnomadPopulation\n ).collect()\n\n # Retrieve LD information from gnomAD\n ld_annotated_assocs = []\n for population in ld_populations:\n if population in assoc_populations:\n pop_parsed_ldindex_path = ld_index_template.format(POP=population)\n pop_matrix_path = ld_matrix_template.format(POP=population)\n ld_index = LDIndex.from_parquet(session, pop_parsed_ldindex_path)\n ld_matrix = BlockMatrix.read(pop_matrix_path)\n ld_annotated_assocs.append(\n LDAnnotatorGnomad.get_ld_annotated_assocs_for_population(\n population,\n ld_index,\n ld_matrix,\n locus_ancestry,\n min_r2,\n ).coalesce(400)\n )\n return reduce(DataFrame.unionByName, ld_annotated_assocs)\n
Probabilistic Identification of Causal SNPs (PICS), an algorithm estimating the probability that an individual variant is causal considering the haplotype structure and observed pattern of association at the genetic locus.
Source code in src/otg/method/pics.py
class PICS:\n\"\"\"Probabilistic Identification of Causal SNPs (PICS), an algorithm estimating the probability that an individual variant is causal considering the haplotype structure and observed pattern of association at the genetic locus.\"\"\"\n\n @staticmethod\n def _pics_relative_posterior_probability(\n neglog_p: float, pics_snp_mu: float, pics_snp_std: float\n ) -> float:\n\"\"\"Compute the PICS posterior probability for a given SNP.\n\n !!! info \"This probability needs to be scaled to take into account the probabilities of the other variants in the locus.\"\n\n Args:\n neglog_p (float): Negative log p-value of the lead variant\n pics_snp_mu (float): Mean P value of the association between a SNP and a trait\n pics_snp_std (float): Standard deviation for the P value of the association between a SNP and a trait\n\n Returns:\n Relative posterior probability of a SNP being causal in a locus\n\n Examples:\n >>> rel_prob = PICS._pics_relative_posterior_probability(neglog_p=10.0, pics_snp_mu=1.0, pics_snp_std=10.0)\n >>> round(rel_prob, 3)\n 0.368\n \"\"\"\n return float(norm(pics_snp_mu, pics_snp_std).sf(neglog_p) * 2)\n\n @staticmethod\n def _pics_standard_deviation(neglog_p: float, r2: float, k: float) -> float | None:\n\"\"\"Compute the PICS standard deviation.\n\n This distribution is obtained after a series of permutation tests described in the PICS method, and it is only\n valid when the SNP is highly linked with the lead (r2 > 0.5).\n\n Args:\n neglog_p (float): Negative log p-value of the lead variant\n r2 (float): LD score between a given SNP and the lead variant\n k (float): Empiric constant that can be adjusted to fit the curve, 6.4 recommended.\n\n Returns:\n Standard deviation for the P value of the association between a SNP and a trait\n\n Examples:\n >>> PICS._pics_standard_deviation(neglog_p=1.0, r2=1.0, k=6.4)\n 0.0\n >>> round(PICS._pics_standard_deviation(neglog_p=10.0, r2=0.5, k=6.4), 3)\n 0.143\n >>> print(PICS._pics_standard_deviation(neglog_p=1.0, r2=0.0, k=6.4))\n None\n \"\"\"\n return (\n (1 - abs(r2) ** 0.5**k) ** 0.5 * (neglog_p) ** 0.5 / 2\n if r2 >= 0.5\n else None\n )\n\n @staticmethod\n def _pics_mu(neglog_p: float, r2: float) -> float | None:\n\"\"\"Compute the PICS mu that estimates the probability of association between a given SNP and the trait.\n\n This distribution is obtained after a series of permutation tests described in the PICS method, and it is only\n valid when the SNP is highly linked with the lead (r2 > 0.5).\n\n Args:\n neglog_p (float): Negative log p-value of the lead variant\n r2 (float): LD score between a given SNP and the lead variant\n\n Returns:\n Mean P value of the association between a SNP and a trait\n\n Examples:\n >>> PICS._pics_mu(neglog_p=1.0, r2=1.0)\n 1.0\n >>> PICS._pics_mu(neglog_p=10.0, r2=0.5)\n 5.0\n >>> print(PICS._pics_mu(neglog_p=10.0, r2=0.3))\n None\n \"\"\"\n return neglog_p * r2 if r2 >= 0.5 else None\n\n @staticmethod\n def _finemap(\n credible_set: list[Row], lead_neglog_p: float, k: float\n ) -> list | None:\n\"\"\"Calculates the probability of a variant being causal in a study-locus context by applying the PICS method.\n\n It is intended to be applied as an UDF in `PICS.finemap`, where each row is a StudyLocus association.\n The function iterates over every SNP in the `credibleSet` array, and it returns an updated credibleSet with\n its association signal and causality probability as of PICS.\n\n Args:\n credible_set (list): list of tagging variants after expanding the locus\n lead_neglog_p (float): P value of the association signal between the lead variant and the study in the form of -log10.\n k (float): Empiric constant that can be adjusted to fit the curve, 6.4 recommended.\n\n Returns:\n List of tagging variants with an estimation of the association signal and their posterior probability as of PICS.\n \"\"\"\n if credible_set is None:\n return None\n elif not credible_set:\n return []\n\n tmp_credible_set = []\n new_credible_set = []\n # First iteration: calculation of mu, standard deviation, and the relative posterior probability\n for tag_struct in credible_set:\n tag_dict = (\n tag_struct.asDict()\n ) # tag_struct is of type pyspark.Row, we'll represent it as a dict\n if (\n not tag_dict[\"r2Overall\"]\n or tag_dict[\"r2Overall\"] < 0.5\n or not lead_neglog_p\n ):\n # If PICS cannot be calculated, we'll return the original credible set\n new_credible_set.append(tag_dict)\n continue\n pics_snp_mu = PICS._pics_mu(lead_neglog_p, tag_dict[\"r2Overall\"])\n pics_snp_std = PICS._pics_standard_deviation(\n lead_neglog_p, tag_dict[\"r2Overall\"], k\n )\n pics_snp_std = 0.001 if pics_snp_std == 0 else pics_snp_std\n if pics_snp_mu is not None and pics_snp_std is not None:\n posterior_probability = PICS._pics_relative_posterior_probability(\n lead_neglog_p, pics_snp_mu, pics_snp_std\n )\n tag_dict[\"tagPValue\"] = 10**-pics_snp_mu\n tag_dict[\"tagStandardError\"] = 10**-pics_snp_std\n tag_dict[\"relativePosteriorProbability\"] = posterior_probability\n\n tmp_credible_set.append(tag_dict)\n\n # Second iteration: calculation of the sum of all the posteriors in each study-locus, so that we scale them between 0-1\n total_posteriors = sum(\n tag_dict.get(\"relativePosteriorProbability\", 0)\n for tag_dict in tmp_credible_set\n )\n\n # Third iteration: calculation of the final posteriorProbability\n for tag_dict in tmp_credible_set:\n if total_posteriors != 0:\n tag_dict[\"posteriorProbability\"] = float(\n tag_dict.get(\"relativePosteriorProbability\", 0) / total_posteriors\n )\n tag_dict.pop(\"relativePosteriorProbability\")\n new_credible_set.append(tag_dict)\n return new_credible_set\n\n @classmethod\n def finemap(\n cls: type[PICS], associations: StudyLocus, k: float = 6.4\n ) -> StudyLocus:\n\"\"\"Run PICS on a study locus.\n\n !!! info \"Study locus needs to be LD annotated\"\n The study locus needs to be LD annotated before PICS can be calculated.\n\n Args:\n associations (StudyLocus): Study locus to finemap using PICS\n k (float): Empiric constant that can be adjusted to fit the curve, 6.4 recommended.\n\n Returns:\n StudyLocus: Study locus with PICS results\n \"\"\"\n # Register UDF by defining the structure of the output credibleSet array of structs\n credset_schema = t.ArrayType(\n [field.dataType.elementType for field in associations.schema if field.name == \"credibleSet\"][0] # type: ignore\n )\n _finemap_udf = f.udf(\n lambda credible_set, neglog_p: PICS._finemap(credible_set, neglog_p, k),\n credset_schema,\n )\n\n associations.df = (\n associations.df.withColumn(\"neglog_pvalue\", associations.neglog_pvalue())\n .withColumn(\n \"credibleSet\",\n f.when(\n f.col(\"credibleSet\").isNotNull(),\n _finemap_udf(f.col(\"credibleSet\"), f.col(\"neglog_pvalue\")),\n ),\n )\n .drop(\"neglog_pvalue\")\n )\n return associations\n
The study locus needs to be LD annotated before PICS can be calculated.
Parameters:
Name Type Description Default associationsStudyLocus
Study locus to finemap using PICS
required kfloat
Empiric constant that can be adjusted to fit the curve, 6.4 recommended.
6.4
Returns:
Name Type Description StudyLocusStudyLocus
Study locus with PICS results
Source code in src/otg/method/pics.py
@classmethod\ndef finemap(\n cls: type[PICS], associations: StudyLocus, k: float = 6.4\n) -> StudyLocus:\n\"\"\"Run PICS on a study locus.\n\n !!! info \"Study locus needs to be LD annotated\"\n The study locus needs to be LD annotated before PICS can be calculated.\n\n Args:\n associations (StudyLocus): Study locus to finemap using PICS\n k (float): Empiric constant that can be adjusted to fit the curve, 6.4 recommended.\n\n Returns:\n StudyLocus: Study locus with PICS results\n \"\"\"\n # Register UDF by defining the structure of the output credibleSet array of structs\n credset_schema = t.ArrayType(\n [field.dataType.elementType for field in associations.schema if field.name == \"credibleSet\"][0] # type: ignore\n )\n _finemap_udf = f.udf(\n lambda credible_set, neglog_p: PICS._finemap(credible_set, neglog_p, k),\n credset_schema,\n )\n\n associations.df = (\n associations.df.withColumn(\"neglog_pvalue\", associations.neglog_pvalue())\n .withColumn(\n \"credibleSet\",\n f.when(\n f.col(\"credibleSet\").isNotNull(),\n _finemap_udf(f.col(\"credibleSet\"), f.col(\"neglog_pvalue\")),\n ),\n )\n .drop(\"neglog_pvalue\")\n )\n return associations\n
This workflow runs colocalization analyses that assess the degree to which independent signals of the association share the same causal variant in a region of the genome, typically limited by linkage disequilibrium (LD).
Source code in src/otg/colocalisation.py
@dataclass\nclass ColocalisationStep(ColocalisationStepConfig):\n\"\"\"Colocalisation step.\n\n This workflow runs colocalization analyses that assess the degree to which independent signals of the association share the same causal variant in a region of the genome, typically limited by linkage disequilibrium (LD).\n \"\"\"\n\n session: Session = Session()\n\n def run(self: ColocalisationStep) -> None:\n\"\"\"Run colocalisation step.\"\"\"\n # Study-locus information\n sl = StudyLocus.from_parquet(self.session, self.study_locus_path)\n si = StudyIndex.from_parquet(self.session, self.study_index_path)\n\n # Study-locus overlaps for 95% credible sets\n sl_overlaps = sl.credible_set(CredibleInterval.IS95).overlaps(si)\n\n coloc_results = Coloc.colocalise(\n sl_overlaps, self.priorc1, self.priorc2, self.priorc12\n )\n ecaviar_results = ECaviar.colocalise(sl_overlaps)\n\n coloc_results.df.unionByName(ecaviar_results.df, allowMissingColumns=True)\n\n coloc_results.df.write.mode(self.session.write_mode).parquet(self.coloc_path)\n
Colocalisation step requirements.
Attributes:
Name Type Description study_locus_pathDictConfig
Input Study-locus path.
coloc_pathDictConfig
Output Colocalisation path.
priorc1float
Prior on variant being causal for trait 1.
priorc2float
Prior on variant being causal for trait 2.
priorc12float
Prior on variant being causal for traits 1 and 2.
Source code in src/otg/config.py
@dataclass\nclass ColocalisationStepConfig:\n\"\"\"Colocalisation step requirements.\n\n Attributes:\n study_locus_path (DictConfig): Input Study-locus path.\n coloc_path (DictConfig): Output Colocalisation path.\n priorc1 (float): Prior on variant being causal for trait 1.\n priorc2 (float): Prior on variant being causal for trait 2.\n priorc12 (float): Prior on variant being causal for traits 1 and 2.\n \"\"\"\n\n _target_: str = \"otg.colocalisation.ColocalisationStep\"\n study_locus_path: str = MISSING\n study_index_path: str = MISSING\n coloc_path: str = MISSING\n priorc1: float = 1e-4\n priorc2: float = 1e-4\n priorc12: float = 1e-5\n
Variant annotation step produces a dataset of the type VariantAnnotation derived from gnomADs gnomad.genomes.vX.X.X.sites.ht Hail's table. This dataset is used to validate variants and as a source of annotation.
Source code in src/otg/variant_annotation.py
@dataclass\nclass VariantAnnotationStep(VariantAnnotationStepConfig):\n\"\"\"Variant annotation step.\n\n Variant annotation step produces a dataset of the type `VariantAnnotation` derived from gnomADs `gnomad.genomes.vX.X.X.sites.ht` Hail's table. This dataset is used to validate variants and as a source of annotation.\n \"\"\"\n\n session: Session = Session()\n\n def run(self: VariantAnnotationStep) -> None:\n\"\"\"Run variant annotation step.\"\"\"\n # init hail session\n hl.init(sc=self.session.spark.sparkContext, log=\"/dev/null\")\n\n\"\"\"Run variant annotation step.\"\"\"\n variant_annotation = VariantAnnotation.from_gnomad(\n self.gnomad_genomes,\n self.chain_38_to_37,\n self.populations,\n )\n # Writing data partitioned by chromosome and position:\n (\n variant_annotation.df.repartition(400, \"chromosome\")\n .sortWithinPartitions(\"chromosome\", \"position\")\n .write.partitionBy(\"chromosome\")\n .mode(self.session.write_mode)\n .parquet(self.variant_annotation_path)\n )\n
Using a VariantAnnotation dataset as a reference, this step creates and writes a dataset of the type VariantIndex that includes only variants that have disease-association data with a reduced set of annotations.
Source code in src/otg/variant_index.py
@dataclass\nclass VariantIndexStep(VariantIndexStepConfig):\n\"\"\"Variant index step.\n\n Using a `VariantAnnotation` dataset as a reference, this step creates and writes a dataset of the type `VariantIndex` that includes only variants that have disease-association data with a reduced set of annotations.\n \"\"\"\n\n session: Session = Session()\n\n def run(self: VariantIndexStep) -> None:\n\"\"\"Run variant index step.\"\"\"\n # Variant annotation dataset\n va = VariantAnnotation.from_parquet(self.session, self.variant_annotation_path)\n\n # Study-locus dataset\n study_locus = StudyLocus.from_parquet(self.session, self.study_locus_path)\n\n # Reduce scope of variant annotation dataset to only variants in study-locus sets:\n va_slimmed = va.filter_by_variant_df(\n study_locus.unique_lead_tag_variants(), [\"id\", \"chromosome\"]\n )\n\n # Generate variant index ussing a subset of the variant annotation dataset\n vi = VariantIndex.from_variant_annotation(va_slimmed)\n\n # Write data:\n # self.etl.logger.info(\n # f\"Writing invalid variants from the credible set to: {self.variant_invalid}\"\n # )\n # vi.invalid_variants.write.mode(self.etl.write_mode).parquet(\n # self.variant_invalid\n # )\n\n self.session.logger.info(f\"Writing variant index to: {self.variant_index_path}\")\n (\n vi.df.write.partitionBy(\"chromosome\")\n .mode(self.session.write_mode)\n .parquet(self.variant_index_path)\n )\n
This step aims to generate a dataset that contains multiple pieces of evidence supporting the functional association of specific variants with genes. Some of the evidence types include:
Chromatin interaction experiments, e.g. Promoter Capture Hi-C (PCHi-C).
In silico functional predictions, e.g. Variant Effect Predictor (VEP) from Ensembl.
Distance between the variant and each gene's canonical transcription start site (TSS).
Source code in src/otg/v2g.py
@dataclass\nclass V2GStep(V2GStepConfig):\n\"\"\"Variant-to-gene (V2G) step.\n\n This step aims to generate a dataset that contains multiple pieces of evidence supporting the functional association of specific variants with genes. Some of the evidence types include:\n\n 1. Chromatin interaction experiments, e.g. Promoter Capture Hi-C (PCHi-C).\n 2. In silico functional predictions, e.g. Variant Effect Predictor (VEP) from Ensembl.\n 3. Distance between the variant and each gene's canonical transcription start site (TSS).\n\n \"\"\"\n\n session: Session = Session()\n\n def run(self: V2GStep) -> None:\n\"\"\"Run V2G dataset generation.\"\"\"\n # Filter gene index by approved biotypes to define V2G gene universe\n gene_index_filtered = GeneIndex.from_parquet(\n self.session, self.gene_index_path\n ).filter_by_biotypes(self.approved_biotypes)\n\n vi = VariantIndex.from_parquet(self.session, self.variant_index_path).persist()\n va = VariantAnnotation.from_parquet(self.session, self.variant_annotation_path)\n vep_consequences = self.session.spark.read.csv(\n self.vep_consequences_path, sep=\"\\t\", header=True\n )\n\n # Variant annotation reduced to the variant index to define V2G variant universe\n va_slimmed = va.filter_by_variant_df(vi.df, [\"id\", \"chromosome\"]).persist()\n\n # lift over variants to hg38\n lift = LiftOverSpark(\n self.liftover_chain_file_path, self.liftover_max_length_difference\n )\n\n v2g_datasets = [\n va_slimmed.get_distance_to_tss(gene_index_filtered, self.max_distance),\n # variant effects\n va_slimmed.get_most_severe_vep_v2g(vep_consequences, gene_index_filtered),\n va_slimmed.get_polyphen_v2g(gene_index_filtered),\n va_slimmed.get_sift_v2g(gene_index_filtered),\n va_slimmed.get_plof_v2g(gene_index_filtered),\n # intervals\n Intervals.parse_andersson(\n self.session, self.anderson_path, gene_index_filtered, lift\n ).v2g(vi),\n Intervals.parse_javierre(\n self.session, self.javierre_path, gene_index_filtered, lift\n ).v2g(vi),\n Intervals.parse_jung(\n self.session, self.jung_path, gene_index_filtered, lift\n ).v2g(vi),\n Intervals.parse_thurman(\n self.session, self.thurnman_path, gene_index_filtered, lift\n ).v2g(vi),\n ]\n\n # merge all V2G datasets\n v2g = V2G(\n _df=reduce(\n lambda x, y: x.unionByName(y, allowMissingColumns=True),\n [dataset.df for dataset in v2g_datasets],\n ).repartition(\"chromosome\")\n )\n # write V2G dataset\n (\n v2g.df.write.partitionBy(\"chromosome\")\n .mode(self.session.write_mode)\n .parquet(self.v2g_path)\n )\n
"}]}
\ No newline at end of file
+{"config":{"lang":["en"],"separator":"[\\s\\-]+","pipeline":["stopWordFilter"]},"docs":[{"location":"","title":"Home","text":"
Ingestion and analysis of genetic and functional genomic data for the identification and prioritisation of drug targets.
This project is still in experimental phase. Please refer to the roadmap section for more information.
For information on how to configure the development environment, run the code, or contribute changes, see the contributing section. For known technical issues and solutions to them, see the troubleshooting section.
"},{"location":"contributing/","title":"Environment configuration and contributing changes","text":""},{"location":"contributing/#one-time-configuration","title":"One-time configuration","text":"
The steps in this section only ever need to be done once on any particular system.
Google Cloud configuration: 1. Install Google Cloud SDK: https://cloud.google.com/sdk/docs/install. 1. Log in to your work Google Account: run gcloud auth login and follow instructions. 1. Obtain Google application credentials: run gcloud auth application-default login and follow instructions.
Check that you have the make utility installed, and if not (which is unlikely), install it using your system package manager.
Run make setup-dev to install/update the necessary packages and activate the development environment. You need to do this every time you open a new shell.
It is recommended to use VS Code as an IDE for development.
"},{"location":"contributing/#how-to-run-the-code","title":"How to run the code","text":"
All pipelines in this repository are intended to be run in Google Dataproc. Running them locally is not currently supported.
In order to run the code:
Manually edit your local workflow/dag.yaml file and comment out the steps you do not want to run.
Manually edit your local pyproject.toml file and modify the version of the code.
This must be different from the version used by any other people working on the repository to avoid any deployment conflicts, so it's a good idea to use your name, for example: 1.2.3+jdoe.
You can also add a brief branch description, for example: 1.2.3+jdoe.myfeature.
Note that the version must comply with PEP440 conventions, otherwise Poetry will not allow it to be deployed.
Do not use underscores or hyphens in your version name. When building the WHL file, they will be automatically converted to dots, which means the file name will no longer match the version and the build will fail. Use dots instead.
Run make build.
This will create a bundle containing the neccessary code, configuration and dependencies to run the ETL pipeline, and then upload this bundle to Google Cloud.
A version specific subpath is used, so uploading the code will not affect any branches but your own.
If there was already a code bundle uploaded with the same version number, it will be replaced.
Submit the Dataproc job with poetry run python workflow/workflow_template.py
You will need to specify additional parameters, some are mandatory and some are optional. Run with --help to see usage.
The script will provision the cluster and submit the job.
The cluster will take a few minutes to get provisioned and running, during which the script will not output anything, this is normal.
Once submitted, you can monitor the progress of your job on this page: https://console.cloud.google.com/dataproc/jobs?project=open-targets-genetics-dev.
On completion (whether successful or a failure), the cluster will be automatically removed, so you don't have to worry about shutting it down to avoid incurring charges.
"},{"location":"contributing/#how-to-generate-a-local-copy-of-the-documentation","title":"How to generate a local copy of the documentation","text":"
Run poetry run mkdocs serve. This will generate the local copy of the documentation and will start a local server to browse it (URL will be printed, usually http://127.0.0.1:8000/).
"},{"location":"contributing/#how-to-run-the-tests","title":"How to run the tests","text":"
When making changes, and especially when implementing a new module or feature, it's essential to ensure that all relevant sections of the code base are modified.
If during development you had a question which wasn't covered in the documentation, and someone explained it to you, add it to the documentation. The same applies if you encountered any instructions in the documentation which were obsolete or incorrect.
Documentation autogeneration expressions start with :::. They will automatically generate sections of the documentation based on class and method docstrings. Be sure to update them for:
Dataset definitions in docs/reference/dataset (example: docs/reference/dataset/study_index/study_index_finngen.md)
Step definitions in docs/reference/step (example: docs/reference/step/finngen.md)
If you see errors related to BLAS/LAPACK libraries, see this StackOverflow post for guidance.
"},{"location":"troubleshooting/#pyenv-and-poetry","title":"Pyenv and Poetry","text":"
If you see various errors thrown by Pyenv or Poetry, they can be hard to specifically diagnose and resolve. In this case, it often helps to remove those tools from the system completely. Follow these steps:
Close your currently activated environment, if any: exit
Officially, PySpark requires Java version 8 (a.k.a. 1.8) or above to work. However, if you have a very recent version of Java, you may experience issues, as it may introduce breaking changes that PySpark hasn't had time to integrate. For example, as of May 2023, PySpark did not work with Java 20.
If you are encountering problems with initialising a Spark session, try using Java 11.
If you see an error message thrown by pre-commit, which looks like this (SyntaxError: Unexpected token '?'), followed by a JavaScript traceback, the issue is likely with your system NodeJS version.
One solution which can help in this case is to upgrade your system NodeJS version. However, this may not always be possible. For example, Ubuntu repository is several major versions behind the latest version as of July 2023.
Another solution which helps is to remove Node, NodeJS, and npm from your system entirely. In this case, pre-commit will not try to rely on a system version of NodeJS and will install its own, suitable one.
On Ubuntu, this can be done using sudo apt remove node nodejs npm, followed by sudo apt autoremove. But in some cases, depending on your existing installation, you may need to also manually remove some files. See this StackOverflow answer for guidance.
After running these commands, you are advised to open a fresh shell, and then also reinstall Pyenv and Poetry to make sure they pick up the changes (see relevant section above).
Dataset is a wrapper around a Spark DataFrame with a predefined schema. Schemas for each child dataset are described in the json.schemas module.
Source code in src/otg/dataset/dataset.py
@dataclass\nclass Dataset:\n\"\"\"Open Targets Genetics Dataset.\n\n `Dataset` is a wrapper around a Spark DataFrame with a predefined schema. Schemas for each child dataset are described in the `json.schemas` module.\n \"\"\"\n\n _df: DataFrame\n _schema: StructType\n\n def __post_init__(self: Dataset) -> None:\n\"\"\"Post init.\"\"\"\n self.validate_schema()\n\n @property\n def df(self: Dataset) -> DataFrame:\n\"\"\"Dataframe included in the Dataset.\"\"\"\n return self._df\n\n @df.setter\n def df(self: Dataset, new_df: DataFrame) -> None: # noqa: CCE001\n\"\"\"Dataframe setter.\"\"\"\n self._df = new_df\n self.validate_schema()\n\n @property\n def schema(self: Dataset) -> StructType:\n\"\"\"Dataframe expected schema.\"\"\"\n return self._schema\n\n @classmethod\n def from_parquet(\n cls: type[Dataset], session: Session, path: str, schema: StructType\n ) -> Dataset:\n\"\"\"Reads a parquet file into a Dataset with a given schema.\n\n Args:\n session (Session): ETL session\n path (str): Path to parquet file\n schema (StructType): Schema to use\n\n Returns:\n Dataset: Dataset with given schema\n \"\"\"\n df = session.read_parquet(path=path, schema=schema)\n return cls(_df=df, _schema=schema)\n\n def validate_schema(self: Dataset) -> None: # sourcery skip: invert-any-all\n\"\"\"Validate DataFrame schema against expected class schema.\n\n Raises:\n ValueError: DataFrame schema is not valid\n \"\"\"\n expected_schema = self._schema\n expected_fields = flatten_schema(expected_schema)\n observed_schema = self._df.schema\n observed_fields = flatten_schema(observed_schema)\n\n # Unexpected fields in dataset\n if unexpected_struct_fields := [\n x for x in observed_fields if x not in expected_fields\n ]:\n raise ValueError(\n f\"The {unexpected_struct_fields} fields are not included in DataFrame schema: {expected_fields}\"\n )\n\n # Required fields not in dataset\n required_fields = [x.name for x in expected_schema if not x.nullable]\n if missing_required_fields := [\n req\n for req in required_fields\n if not any(field.name == req for field in observed_fields)\n ]:\n raise ValueError(\n f\"The {missing_required_fields} fields are required but missing: {required_fields}\"\n )\n\n # Fields with duplicated names\n if duplicated_fields := [\n x for x in set(observed_fields) if observed_fields.count(x) > 1\n ]:\n raise ValueError(\n f\"The following fields are duplicated in DataFrame schema: {duplicated_fields}\"\n )\n\n # Fields with different datatype\n if fields_with_different_observed_datatype := [\n field\n for field in set(observed_fields)\n if observed_fields.count(field) != expected_fields.count(field)\n ]:\n raise ValueError(\n f\"The following fields present differences in their datatypes: {fields_with_different_observed_datatype}.\"\n )\n
Validate DataFrame schema against expected class schema.
Raises:
Type Description ValueError
DataFrame schema is not valid
Source code in src/otg/dataset/dataset.py
def validate_schema(self: Dataset) -> None: # sourcery skip: invert-any-all\n\"\"\"Validate DataFrame schema against expected class schema.\n\n Raises:\n ValueError: DataFrame schema is not valid\n \"\"\"\n expected_schema = self._schema\n expected_fields = flatten_schema(expected_schema)\n observed_schema = self._df.schema\n observed_fields = flatten_schema(observed_schema)\n\n # Unexpected fields in dataset\n if unexpected_struct_fields := [\n x for x in observed_fields if x not in expected_fields\n ]:\n raise ValueError(\n f\"The {unexpected_struct_fields} fields are not included in DataFrame schema: {expected_fields}\"\n )\n\n # Required fields not in dataset\n required_fields = [x.name for x in expected_schema if not x.nullable]\n if missing_required_fields := [\n req\n for req in required_fields\n if not any(field.name == req for field in observed_fields)\n ]:\n raise ValueError(\n f\"The {missing_required_fields} fields are required but missing: {required_fields}\"\n )\n\n # Fields with duplicated names\n if duplicated_fields := [\n x for x in set(observed_fields) if observed_fields.count(x) > 1\n ]:\n raise ValueError(\n f\"The following fields are duplicated in DataFrame schema: {duplicated_fields}\"\n )\n\n # Fields with different datatype\n if fields_with_different_observed_datatype := [\n field\n for field in set(observed_fields)\n if observed_fields.count(field) != expected_fields.count(field)\n ]:\n raise ValueError(\n f\"The following fields present differences in their datatypes: {fields_with_different_observed_datatype}.\"\n )\n
Convert intervals into V2G by intersecting with a variant index.
Parameters:
Name Type Description Default variant_indexVariantIndex
Variant index dataset
required
Returns:
Name Type Description V2GV2G
Variant-to-gene evidence dataset
Source code in src/otg/dataset/intervals.py
def v2g(self: Intervals, variant_index: VariantIndex) -> V2G:\n\"\"\"Convert intervals into V2G by intersecting with a variant index.\n\n Args:\n variant_index (VariantIndex): Variant index dataset\n\n Returns:\n V2G: Variant-to-gene evidence dataset\n \"\"\"\n return V2G(\n _df=(\n # TODO: We can include the start and end position as part of the `on` clause in the join\n self.df.alias(\"interval\")\n .join(\n variant_index.df.selectExpr(\n \"chromosome as vi_chromosome\", \"variantId\", \"position\"\n ).alias(\"vi\"),\n on=[\n f.col(\"vi.vi_chromosome\") == f.col(\"interval.chromosome\"),\n f.col(\"vi.position\").between(\n f.col(\"interval.start\"), f.col(\"interval.end\")\n ),\n ],\n how=\"inner\",\n )\n .drop(\"start\", \"end\", \"vi_chromosome\")\n )\n )\n
Annotate LD index with indices starting and stopping at a given interval.
Parameters:
Name Type Description Default ld_radiusint
radius around each position
required
Returns:
Name Type Description LDIndexLDIndex
including start_idx and stop_idx columns
Source code in src/otg/dataset/ld_index.py
def annotate_index_intervals(self: LDIndex, ld_radius: int) -> LDIndex:\n\"\"\"Annotate LD index with indices starting and stopping at a given interval.\n\n Args:\n ld_radius (int): radius around each position\n\n Returns:\n LDIndex: including `start_idx` and `stop_idx` columns\n \"\"\"\n index_with_positions = (\n self._df.drop(\"start_idx\", \"stop_idx\")\n .select(\n \"*\",\n LDIndex._interval_start(\n contig=f.col(\"chromosome\"),\n position=f.col(\"position\"),\n ld_radius=ld_radius,\n ).alias(\"start_pos\"),\n LDIndex._interval_stop(\n contig=f.col(\"chromosome\"),\n position=f.col(\"position\"),\n ld_radius=ld_radius,\n ).alias(\"stop_pos\"),\n )\n .persist()\n )\n\n self.df = (\n index_with_positions.join(\n (\n index_with_positions\n # Given the multiple variants with the same chromosome/position can have different indices, filter for the lowest index:\n .transform(\n lambda df: get_record_with_minimum_value(\n df, [\"chromosome\", \"position\"], \"idx\"\n )\n ).select(\n \"chromosome\",\n f.col(\"position\").alias(\"start_pos\"),\n f.col(\"idx\").alias(\"start_idx\"),\n )\n ),\n on=[\"chromosome\", \"start_pos\"],\n )\n .join(\n (\n index_with_positions\n # Given the multiple variants with the same chromosome/position can have different indices, filter for the highest index:\n .transform(\n lambda df: get_record_with_maximum_value(\n df, [\"chromosome\", \"position\"], \"idx\"\n )\n ).select(\n \"chromosome\",\n f.col(\"position\").alias(\"stop_pos\"),\n f.col(\"idx\").alias(\"stop_idx\"),\n )\n ),\n on=[\"chromosome\", \"stop_pos\"],\n )\n # Filter out variants for which start idx > stop idx due to liftover\n .filter(f.col(\"start_idx\") < f.col(\"stop_idx\"))\n .drop(\"start_pos\", \"stop_pos\")\n )\n\n return self\n
Filter summary statistics based on the provided p-value threshold.
Parameters:
Name Type Description Default pvaluefloat
upper limit of the p-value to be filtered upon.
required
Returns:
Name Type Description SummaryStatisticsSummaryStatistics
summary statistics object containing single point associations with p-values at least as significant as the provided threshold.
Source code in src/otg/dataset/summary_statistics.py
def pvalue_filter(self: SummaryStatistics, pvalue: float) -> SummaryStatistics:\n\"\"\"Filter summary statistics based on the provided p-value threshold.\n\n Args:\n pvalue (float): upper limit of the p-value to be filtered upon.\n\n Returns:\n SummaryStatistics: summary statistics object containing single point associations with p-values at least as significant as the provided threshold.\n \"\"\"\n # Converting p-value to mantissa and exponent:\n (mantissa, exponent) = split_pvalue(pvalue)\n\n # Applying filter:\n df = self._df.filter(\n (f.col(\"pValueExponent\") < exponent)\n | (\n (f.col(\"pValueExponent\") == exponent)\n & (f.col(\"pValueMantissa\") <= mantissa)\n )\n )\n return SummaryStatistics(_df=df)\n
Dataset with variant-level annotations derived from GnomAD.
Source code in src/otg/dataset/variant_annotation.py
@dataclass\nclass VariantAnnotation(Dataset):\n\"\"\"Dataset with variant-level annotations derived from GnomAD.\"\"\"\n\n _schema: StructType = parse_spark_schema(\"variant_annotation.json\")\n\n @classmethod\n def from_parquet(\n cls: type[VariantAnnotation], session: Session, path: str\n ) -> VariantAnnotation:\n\"\"\"Initialise VariantAnnotation from parquet file.\n\n Args:\n session (Session): ETL session\n path (str): Path to parquet file\n\n Returns:\n VariantAnnotation: VariantAnnotation dataset\n \"\"\"\n df = session.read_parquet(path=path, schema=cls._schema)\n return cls(_df=df, _schema=cls._schema)\n\n @classmethod\n def from_gnomad(\n cls: type[VariantAnnotation],\n gnomad_file: str,\n grch38_to_grch37_chain: str,\n populations: list,\n ) -> VariantAnnotation:\n\"\"\"Generate variant annotation dataset from gnomAD.\n\n Some relevant modifications to the original dataset are:\n\n 1. The transcript consequences features provided by VEP are filtered to only refer to the Ensembl canonical transcript.\n 2. Genome coordinates are liftovered from GRCh38 to GRCh37 to keep as annotation.\n 3. Field names are converted to camel case to follow the convention.\n\n Args:\n gnomad_file (str): Path to `gnomad.genomes.vX.X.X.sites.ht` gnomAD dataset\n grch38_to_grch37_chain (str): Path to chain file for liftover\n populations (list): List of populations to include in the dataset\n\n Returns:\n VariantAnnotation: Variant annotation dataset\n \"\"\"\n # Load variants dataset\n ht = hl.read_table(\n gnomad_file,\n _load_refs=False,\n )\n\n # Liftover\n grch37 = hl.get_reference(\"GRCh37\")\n grch38 = hl.get_reference(\"GRCh38\")\n grch38.add_liftover(grch38_to_grch37_chain, grch37)\n\n # Drop non biallelic variants\n ht = ht.filter(ht.alleles.length() == 2)\n # Liftover\n ht = ht.annotate(locus_GRCh37=hl.liftover(ht.locus, \"GRCh37\"))\n # Select relevant fields and nested records to create class\n return cls(\n _df=(\n ht.select(\n gnomad3VariantId=hl.str(\"-\").join(\n [\n ht.locus.contig.replace(\"chr\", \"\"),\n hl.str(ht.locus.position),\n ht.alleles[0],\n ht.alleles[1],\n ]\n ),\n chromosome=ht.locus.contig.replace(\"chr\", \"\"),\n position=convert_gnomad_position_to_ensembl_hail(\n ht.locus.position, ht.alleles[0], ht.alleles[1]\n ),\n variantId=hl.str(\"_\").join(\n [\n ht.locus.contig.replace(\"chr\", \"\"),\n hl.str(\n convert_gnomad_position_to_ensembl_hail(\n ht.locus.position, ht.alleles[0], ht.alleles[1]\n )\n ),\n ht.alleles[0],\n ht.alleles[1],\n ]\n ),\n chromosomeB37=ht.locus_GRCh37.contig.replace(\"chr\", \"\"),\n positionB37=ht.locus_GRCh37.position,\n referenceAllele=ht.alleles[0],\n alternateAllele=ht.alleles[1],\n rsIds=ht.rsid,\n alleleType=ht.allele_info.allele_type,\n cadd=hl.struct(\n phred=ht.cadd.phred,\n raw=ht.cadd.raw_score,\n ),\n alleleFrequencies=hl.set([f\"{pop}-adj\" for pop in populations]).map(\n lambda p: hl.struct(\n populationName=p,\n alleleFrequency=ht.freq[ht.globals.freq_index_dict[p]].AF,\n )\n ),\n vep=hl.struct(\n mostSevereConsequence=ht.vep.most_severe_consequence,\n transcriptConsequences=hl.map(\n lambda x: hl.struct(\n aminoAcids=x.amino_acids,\n consequenceTerms=x.consequence_terms,\n geneId=x.gene_id,\n lof=x.lof,\n polyphenScore=x.polyphen_score,\n polyphenPrediction=x.polyphen_prediction,\n siftScore=x.sift_score,\n siftPrediction=x.sift_prediction,\n ),\n # Only keeping canonical transcripts\n ht.vep.transcript_consequences.filter(\n lambda x: (x.canonical == 1)\n & (x.gene_symbol_source == \"HGNC\")\n ),\n ),\n ),\n )\n .key_by(\"chromosome\", \"position\")\n .drop(\"locus\", \"alleles\")\n .select_globals()\n .to_spark(flatten=False)\n )\n )\n\n def persist(self: VariantAnnotation) -> VariantAnnotation:\n\"\"\"Persist DataFrame included in the Dataset.\"\"\"\n self.df = self._df.persist()\n return self\n\n def max_maf(self: VariantAnnotation) -> Column:\n\"\"\"Maximum minor allele frequency accross all populations.\n\n Returns:\n Column: Maximum minor allele frequency accross all populations.\n \"\"\"\n return f.array_max(\n f.transform(\n self.df.alleleFrequencies,\n lambda af: f.when(\n af.alleleFrequency > 0.5, 1 - af.alleleFrequency\n ).otherwise(af.alleleFrequency),\n )\n )\n\n def filter_by_variant_df(\n self: VariantAnnotation, df: DataFrame, cols: list[str]\n ) -> VariantAnnotation:\n\"\"\"Filter variant annotation dataset by a variant dataframe.\n\n Args:\n df (DataFrame): A dataframe of variants\n cols (List[str]): A list of columns to join on\n\n Returns:\n VariantAnnotation: A filtered variant annotation dataset\n \"\"\"\n self.df = self._df.join(f.broadcast(df.select(cols)), on=cols, how=\"inner\")\n return self\n\n def get_transcript_consequence_df(\n self: VariantAnnotation, filter_by: Optional[GeneIndex] = None\n ) -> DataFrame:\n\"\"\"Dataframe of exploded transcript consequences.\n\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index. Defaults to None.\n\n Returns:\n DataFrame: A dataframe exploded by transcript consequences with the columns variantId, chromosome, transcriptConsequence\n \"\"\"\n # exploding the array removes records without VEP annotation\n transript_consequences = self.df.withColumn(\n \"transcriptConsequence\", f.explode(\"vep.transcriptConsequences\")\n ).select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"transcriptConsequence\",\n f.col(\"transcriptConsequence.geneId\").alias(\"geneId\"),\n )\n if filter_by:\n transript_consequences = transript_consequences.join(\n f.broadcast(filter_by.df),\n on=[\"chromosome\", \"geneId\"],\n )\n return transript_consequences.persist()\n\n def get_most_severe_vep_v2g(\n self: VariantAnnotation,\n vep_consequences: DataFrame,\n filter_by: GeneIndex,\n ) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments based on VEP's predicted consequence on the transcript.\n\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n vep_consequences (DataFrame): A dataframe of VEP consequences\n filter_by (GeneIndex): A gene index to filter by. Defaults to None.\n\n Returns:\n V2G: High and medium severity variant to gene assignments\n \"\"\"\n vep_lut = vep_consequences.select(\n f.element_at(f.split(\"Accession\", r\"/\"), -1).alias(\n \"variantFunctionalConsequenceId\"\n ),\n f.col(\"Term\").alias(\"label\"),\n f.col(\"v2g_score\").cast(\"double\").alias(\"score\"),\n )\n\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n f.col(\"transcriptConsequence.geneId\").alias(\"geneId\"),\n f.explode(\"transcriptConsequence.consequenceTerms\").alias(\"label\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"variantConsequence\").alias(\"datasourceId\"),\n )\n # A variant can have multiple predicted consequences on a transcript, the most severe one is selected\n .join(\n f.broadcast(vep_lut),\n on=\"label\",\n how=\"inner\",\n )\n .filter(f.col(\"score\") != 0)\n .transform(\n lambda df: get_record_with_maximum_value(\n df, [\"variantId\", \"geneId\"], \"score\"\n )\n )\n )\n\n def get_polyphen_v2g(\n self: VariantAnnotation, filter_by: Optional[GeneIndex] = None\n ) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments with a PolyPhen's predicted score on the transcript.\n\n Polyphen informs about the probability that a substitution is damaging. Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by. Defaults to None.\n\n Returns:\n V2G: variant to gene assignments with their polyphen scores\n \"\"\"\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .filter(f.col(\"transcriptConsequence.polyphenScore\").isNotNull())\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"geneId\",\n f.col(\"transcriptConsequence.polyphenScore\").alias(\"score\"),\n f.col(\"transcriptConsequence.polyphenPrediction\").alias(\"label\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"polyphen\").alias(\"datasourceId\"),\n )\n )\n\n def get_sift_v2g(self: VariantAnnotation, filter_by: GeneIndex) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments with a SIFT's predicted score on the transcript.\n\n SIFT informs about the probability that a substitution is tolerated so scores nearer zero are more likely to be deleterious.\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by.\n\n Returns:\n V2G: variant to gene assignments with their SIFT scores\n \"\"\"\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .filter(f.col(\"transcriptConsequence.siftScore\").isNotNull())\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"geneId\",\n f.expr(\"1 - transcriptConsequence.siftScore\").alias(\"score\"),\n f.col(\"transcriptConsequence.siftPrediction\").alias(\"label\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"sift\").alias(\"datasourceId\"),\n )\n )\n\n def get_plof_v2g(self: VariantAnnotation, filter_by: GeneIndex) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments with a flag indicating if the variant is predicted to be a loss-of-function variant by the LOFTEE algorithm.\n\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by.\n\n Returns:\n V2G: variant to gene assignments from the LOFTEE algorithm\n \"\"\"\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .filter(f.col(\"transcriptConsequence.lof\").isNotNull())\n .withColumn(\n \"isHighQualityPlof\",\n f.when(f.col(\"transcriptConsequence.lof\") == \"HC\", True).when(\n f.col(\"transcriptConsequence.lof\") == \"LC\", False\n ),\n )\n .withColumn(\n \"score\",\n f.when(f.col(\"isHighQualityPlof\"), 1.0).when(\n ~f.col(\"isHighQualityPlof\"), 0\n ),\n )\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"geneId\",\n \"isHighQualityPlof\",\n f.col(\"score\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"loftee\").alias(\"datasourceId\"),\n )\n )\n\n def get_distance_to_tss(\n self: VariantAnnotation,\n filter_by: GeneIndex,\n max_distance: int = 500_000,\n ) -> V2G:\n\"\"\"Extracts variant to gene assignments for variants falling within a window of a gene's TSS.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by.\n max_distance (int): The maximum distance from the TSS to consider. Defaults to 500_000.\n\n Returns:\n V2G: variant to gene assignments with their distance to the TSS\n \"\"\"\n return V2G(\n _df=self.df.alias(\"variant\")\n .join(\n f.broadcast(filter_by.locations_lut()).alias(\"gene\"),\n on=[\n f.col(\"variant.chromosome\") == f.col(\"gene.chromosome\"),\n f.abs(f.col(\"variant.position\") - f.col(\"gene.tss\"))\n <= max_distance,\n ],\n how=\"inner\",\n )\n .withColumn(\n \"inverse_distance\",\n max_distance - f.abs(f.col(\"variant.position\") - f.col(\"gene.tss\")),\n )\n .transform(lambda df: normalise_column(df, \"inverse_distance\", \"score\"))\n .select(\n \"variantId\",\n f.col(\"variant.chromosome\").alias(\"chromosome\"),\n \"position\",\n \"geneId\",\n \"score\",\n f.lit(\"distance\").alias(\"datatypeId\"),\n f.lit(\"canonical_tss\").alias(\"datasourceId\"),\n )\n )\n
Extracts variant to gene assignments for variants falling within a window of a gene's TSS.
Parameters:
Name Type Description Default filter_byGeneIndex
A gene index to filter by.
required max_distanceint
The maximum distance from the TSS to consider. Defaults to 500_000.
500000
Returns:
Name Type Description V2GV2G
variant to gene assignments with their distance to the TSS
Source code in src/otg/dataset/variant_annotation.py
def get_distance_to_tss(\n self: VariantAnnotation,\n filter_by: GeneIndex,\n max_distance: int = 500_000,\n) -> V2G:\n\"\"\"Extracts variant to gene assignments for variants falling within a window of a gene's TSS.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by.\n max_distance (int): The maximum distance from the TSS to consider. Defaults to 500_000.\n\n Returns:\n V2G: variant to gene assignments with their distance to the TSS\n \"\"\"\n return V2G(\n _df=self.df.alias(\"variant\")\n .join(\n f.broadcast(filter_by.locations_lut()).alias(\"gene\"),\n on=[\n f.col(\"variant.chromosome\") == f.col(\"gene.chromosome\"),\n f.abs(f.col(\"variant.position\") - f.col(\"gene.tss\"))\n <= max_distance,\n ],\n how=\"inner\",\n )\n .withColumn(\n \"inverse_distance\",\n max_distance - f.abs(f.col(\"variant.position\") - f.col(\"gene.tss\")),\n )\n .transform(lambda df: normalise_column(df, \"inverse_distance\", \"score\"))\n .select(\n \"variantId\",\n f.col(\"variant.chromosome\").alias(\"chromosome\"),\n \"position\",\n \"geneId\",\n \"score\",\n f.lit(\"distance\").alias(\"datatypeId\"),\n f.lit(\"canonical_tss\").alias(\"datasourceId\"),\n )\n )\n
Creates a dataset with variant to gene assignments based on VEP's predicted consequence on the transcript.
Optionally the trancript consequences can be reduced to the universe of a gene index.
Parameters:
Name Type Description Default vep_consequencesDataFrame
A dataframe of VEP consequences
required filter_byGeneIndex
A gene index to filter by. Defaults to None.
required
Returns:
Name Type Description V2GV2G
High and medium severity variant to gene assignments
Source code in src/otg/dataset/variant_annotation.py
def get_most_severe_vep_v2g(\n self: VariantAnnotation,\n vep_consequences: DataFrame,\n filter_by: GeneIndex,\n) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments based on VEP's predicted consequence on the transcript.\n\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n vep_consequences (DataFrame): A dataframe of VEP consequences\n filter_by (GeneIndex): A gene index to filter by. Defaults to None.\n\n Returns:\n V2G: High and medium severity variant to gene assignments\n \"\"\"\n vep_lut = vep_consequences.select(\n f.element_at(f.split(\"Accession\", r\"/\"), -1).alias(\n \"variantFunctionalConsequenceId\"\n ),\n f.col(\"Term\").alias(\"label\"),\n f.col(\"v2g_score\").cast(\"double\").alias(\"score\"),\n )\n\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n f.col(\"transcriptConsequence.geneId\").alias(\"geneId\"),\n f.explode(\"transcriptConsequence.consequenceTerms\").alias(\"label\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"variantConsequence\").alias(\"datasourceId\"),\n )\n # A variant can have multiple predicted consequences on a transcript, the most severe one is selected\n .join(\n f.broadcast(vep_lut),\n on=\"label\",\n how=\"inner\",\n )\n .filter(f.col(\"score\") != 0)\n .transform(\n lambda df: get_record_with_maximum_value(\n df, [\"variantId\", \"geneId\"], \"score\"\n )\n )\n )\n
Creates a dataset with variant to gene assignments with a flag indicating if the variant is predicted to be a loss-of-function variant by the LOFTEE algorithm.
Optionally the trancript consequences can be reduced to the universe of a gene index.
Parameters:
Name Type Description Default filter_byGeneIndex
A gene index to filter by.
required
Returns:
Name Type Description V2GV2G
variant to gene assignments from the LOFTEE algorithm
Source code in src/otg/dataset/variant_annotation.py
def get_plof_v2g(self: VariantAnnotation, filter_by: GeneIndex) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments with a flag indicating if the variant is predicted to be a loss-of-function variant by the LOFTEE algorithm.\n\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by.\n\n Returns:\n V2G: variant to gene assignments from the LOFTEE algorithm\n \"\"\"\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .filter(f.col(\"transcriptConsequence.lof\").isNotNull())\n .withColumn(\n \"isHighQualityPlof\",\n f.when(f.col(\"transcriptConsequence.lof\") == \"HC\", True).when(\n f.col(\"transcriptConsequence.lof\") == \"LC\", False\n ),\n )\n .withColumn(\n \"score\",\n f.when(f.col(\"isHighQualityPlof\"), 1.0).when(\n ~f.col(\"isHighQualityPlof\"), 0\n ),\n )\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"geneId\",\n \"isHighQualityPlof\",\n f.col(\"score\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"loftee\").alias(\"datasourceId\"),\n )\n )\n
Creates a dataset with variant to gene assignments with a PolyPhen's predicted score on the transcript.
Polyphen informs about the probability that a substitution is damaging. Optionally the trancript consequences can be reduced to the universe of a gene index.
Parameters:
Name Type Description Default filter_byGeneIndex
A gene index to filter by. Defaults to None.
None
Returns:
Name Type Description V2GV2G
variant to gene assignments with their polyphen scores
Source code in src/otg/dataset/variant_annotation.py
def get_polyphen_v2g(\n self: VariantAnnotation, filter_by: Optional[GeneIndex] = None\n) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments with a PolyPhen's predicted score on the transcript.\n\n Polyphen informs about the probability that a substitution is damaging. Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by. Defaults to None.\n\n Returns:\n V2G: variant to gene assignments with their polyphen scores\n \"\"\"\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .filter(f.col(\"transcriptConsequence.polyphenScore\").isNotNull())\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"geneId\",\n f.col(\"transcriptConsequence.polyphenScore\").alias(\"score\"),\n f.col(\"transcriptConsequence.polyphenPrediction\").alias(\"label\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"polyphen\").alias(\"datasourceId\"),\n )\n )\n
Creates a dataset with variant to gene assignments with a SIFT's predicted score on the transcript.
SIFT informs about the probability that a substitution is tolerated so scores nearer zero are more likely to be deleterious. Optionally the trancript consequences can be reduced to the universe of a gene index.
Parameters:
Name Type Description Default filter_byGeneIndex
A gene index to filter by.
required
Returns:
Name Type Description V2GV2G
variant to gene assignments with their SIFT scores
Source code in src/otg/dataset/variant_annotation.py
def get_sift_v2g(self: VariantAnnotation, filter_by: GeneIndex) -> V2G:\n\"\"\"Creates a dataset with variant to gene assignments with a SIFT's predicted score on the transcript.\n\n SIFT informs about the probability that a substitution is tolerated so scores nearer zero are more likely to be deleterious.\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index to filter by.\n\n Returns:\n V2G: variant to gene assignments with their SIFT scores\n \"\"\"\n return V2G(\n _df=self.get_transcript_consequence_df(filter_by)\n .filter(f.col(\"transcriptConsequence.siftScore\").isNotNull())\n .select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"geneId\",\n f.expr(\"1 - transcriptConsequence.siftScore\").alias(\"score\"),\n f.col(\"transcriptConsequence.siftPrediction\").alias(\"label\"),\n f.lit(\"vep\").alias(\"datatypeId\"),\n f.lit(\"sift\").alias(\"datasourceId\"),\n )\n )\n
Optionally the trancript consequences can be reduced to the universe of a gene index.
Parameters:
Name Type Description Default filter_byGeneIndex
A gene index. Defaults to None.
None
Returns:
Name Type Description DataFrameDataFrame
A dataframe exploded by transcript consequences with the columns variantId, chromosome, transcriptConsequence
Source code in src/otg/dataset/variant_annotation.py
def get_transcript_consequence_df(\n self: VariantAnnotation, filter_by: Optional[GeneIndex] = None\n) -> DataFrame:\n\"\"\"Dataframe of exploded transcript consequences.\n\n Optionally the trancript consequences can be reduced to the universe of a gene index.\n\n Args:\n filter_by (GeneIndex): A gene index. Defaults to None.\n\n Returns:\n DataFrame: A dataframe exploded by transcript consequences with the columns variantId, chromosome, transcriptConsequence\n \"\"\"\n # exploding the array removes records without VEP annotation\n transript_consequences = self.df.withColumn(\n \"transcriptConsequence\", f.explode(\"vep.transcriptConsequences\")\n ).select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n \"transcriptConsequence\",\n f.col(\"transcriptConsequence.geneId\").alias(\"geneId\"),\n )\n if filter_by:\n transript_consequences = transript_consequences.join(\n f.broadcast(filter_by.df),\n on=[\"chromosome\", \"geneId\"],\n )\n return transript_consequences.persist()\n
Source code in src/otg/dataset/variant_annotation.py
def persist(self: VariantAnnotation) -> VariantAnnotation:\n\"\"\"Persist DataFrame included in the Dataset.\"\"\"\n self.df = self._df.persist()\n return self\n
def persist(self: VariantIndex) -> VariantIndex:\n\"\"\"Persist DataFrame included in the Dataset.\"\"\"\n self.df = self._df.persist()\n return self\n
"},{"location":"components/dataset/variant_to_gene/","title":"Variant to gene","text":"
Bases: Dataset
Variant-to-gene (V2G) evidence dataset.
A variant-to-gene (V2G) evidence is understood as any piece of evidence that supports the association of a variant with a likely causal gene. The evidence can sometimes be context-specific and refer to specific biofeatures (e.g. cell types)
Source code in src/otg/dataset/v2g.py
@dataclass\nclass V2G(Dataset):\n\"\"\"Variant-to-gene (V2G) evidence dataset.\n\n A variant-to-gene (V2G) evidence is understood as any piece of evidence that supports the association of a variant with a likely causal gene. The evidence can sometimes be context-specific and refer to specific `biofeatures` (e.g. cell types)\n \"\"\"\n\n _schema: StructType = parse_spark_schema(\"v2g.json\")\n\n @classmethod\n def from_parquet(cls: type[V2G], session: Session, path: str) -> V2G:\n\"\"\"Initialise V2G from parquet file.\n\n Args:\n session (Session): ETL session\n path (str): Path to parquet file\n\n Returns:\n V2G: V2G dataset\n \"\"\"\n df = session.read_parquet(path=path, schema=cls._schema)\n return cls(_df=df, _schema=cls._schema)\n\n def filter_by_genes(self: V2G, genes: GeneIndex) -> V2G:\n\"\"\"Filter by V2G dataset by genes.\n\n Args:\n genes (GeneIndex): Gene index dataset to filter by\n\n Returns:\n V2G: V2G dataset filtered by genes\n \"\"\"\n self.df = self._df.join(genes.df.select(\"geneId\"), on=\"geneId\", how=\"inner\")\n return self\n
A study index dataset captures all the metadata for all studies including GWAS and Molecular QTL.
Source code in src/otg/dataset/study_index.py
@dataclass\nclass StudyIndex(Dataset):\n\"\"\"Study index dataset.\n\n A study index dataset captures all the metadata for all studies including GWAS and Molecular QTL.\n \"\"\"\n\n _schema: StructType = parse_spark_schema(\"studies.json\")\n\n @classmethod\n def from_parquet(cls: type[StudyIndex], session: Session, path: str) -> StudyIndex:\n\"\"\"Initialise StudyIndex from parquet file.\n\n Args:\n session (Session): ETL session\n path (str): Path to parquet file\n\n Returns:\n StudyIndex: Study index dataset\n \"\"\"\n df = session.read_parquet(path=path, schema=cls._schema)\n return cls(_df=df, _schema=cls._schema)\n\n def study_type_lut(self: StudyIndex) -> DataFrame:\n\"\"\"Return a lookup table of study type.\n\n Returns:\n DataFrame: A dataframe containing `studyId` and `studyType` columns.\n \"\"\"\n return self.df.select(\"studyId\", \"studyType\")\n
A dataframe containing studyId and studyType columns.
Source code in src/otg/dataset/study_index.py
def study_type_lut(self: StudyIndex) -> DataFrame:\n\"\"\"Return a lookup table of study type.\n\n Returns:\n DataFrame: A dataframe containing `studyId` and `studyType` columns.\n \"\"\"\n return self.df.select(\"studyId\", \"studyType\")\n
"},{"location":"components/dataset/study_index/study_index_finngen/","title":"Study index finngen","text":"
Bases: StudyIndex
Study index dataset from FinnGen.
The following information is aggregated/extracted:
Study ID in the special format (FINNGEN_R9_*)
Trait name (for example, Amoebiasis)
Number of cases and controls
Link to the summary statistics location
Some fields are also populated as constants, such as study type and the initial sample size.
Source code in src/otg/dataset/study_index.py
@dataclass\nclass StudyIndexFinnGen(StudyIndex):\n\"\"\"Study index dataset from FinnGen.\n\n The following information is aggregated/extracted:\n\n - Study ID in the special format (FINNGEN_R9_*)\n - Trait name (for example, Amoebiasis)\n - Number of cases and controls\n - Link to the summary statistics location\n\n Some fields are also populated as constants, such as study type and the initial sample size.\n \"\"\"\n\n @classmethod\n def from_source(\n cls: type[StudyIndexFinnGen],\n finngen_studies: DataFrame,\n finngen_release_prefix: str,\n finngen_sumstat_url_prefix: str,\n finngen_sumstat_url_suffix: str,\n ) -> StudyIndexFinnGen:\n\"\"\"This function ingests study level metadata from FinnGen.\n\n Args:\n finngen_studies (DataFrame): FinnGen raw study table\n finngen_release_prefix (str): Release prefix pattern.\n finngen_sumstat_url_prefix (str): URL prefix for summary statistics location.\n finngen_sumstat_url_suffix (str): URL prefix suffix for summary statistics location.\n\n Returns:\n StudyIndexFinnGen: Parsed and annotated FinnGen study table.\n \"\"\"\n return cls(\n _df=(\n # Read FinnGen raw data.\n finngen_studies.select(\n # Select the desired columns.\n f.concat(\n f.lit(finngen_release_prefix + \"_\"), f.col(\"phenocode\")\n ).alias(\"studyId\"),\n f.col(\"phenostring\").alias(\"traitFromSource\"),\n f.col(\"num_cases\").alias(\"nCases\"),\n f.col(\"num_controls\").alias(\"nControls\"),\n # Set constant value columns.\n f.lit(finngen_release_prefix).alias(\"projectId\"),\n f.lit(\"gwas\").alias(\"studyType\"),\n f.lit(True).alias(\"hasSumstats\"),\n f.lit(\"377,277 (210,870 females and 166,407 males)\").alias(\n \"initialSampleSize\"\n ),\n )\n .withColumn(\"nSamples\", f.col(\"nCases\") + f.col(\"nControls\"))\n .withColumn(\n \"summarystatsLocation\",\n f.concat(\n f.lit(finngen_sumstat_url_prefix),\n f.col(\"studyId\"),\n f.lit(finngen_sumstat_url_suffix),\n ),\n )\n )\n )\n
This dataset captures associations between study/traits and a genetic loci as provided by finemapping methods.
Source code in src/otg/dataset/study_locus.py
@dataclass\nclass StudyLocus(Dataset):\n\"\"\"Study-Locus dataset.\n\n This dataset captures associations between study/traits and a genetic loci as provided by finemapping methods.\n \"\"\"\n\n _schema: StructType = parse_spark_schema(\"study_locus.json\")\n\n @staticmethod\n def _overlapping_peaks(credset_to_overlap: DataFrame) -> DataFrame:\n\"\"\"Calculate overlapping signals (study-locus) between GWAS-GWAS and GWAS-Molecular trait.\n\n Args:\n credset_to_overlap (DataFrame): DataFrame containing at least `studyLocusId`, `studyType`, `chromosome` and `tagVariantId` columns.\n\n Returns:\n DataFrame: containing `left_studyLocusId`, `right_studyLocusId` and `chromosome` columns.\n \"\"\"\n # Reduce columns to the minimum to reduce the size of the dataframe\n credset_to_overlap = credset_to_overlap.select(\n \"studyLocusId\", \"studyType\", \"chromosome\", \"tagVariantId\"\n )\n return (\n credset_to_overlap.alias(\"left\")\n .filter(f.col(\"studyType\") == \"gwas\")\n # Self join with complex condition. Left it's all gwas and right can be gwas or molecular trait\n .join(\n credset_to_overlap.alias(\"right\"),\n on=[\n f.col(\"left.chromosome\") == f.col(\"right.chromosome\"),\n f.col(\"left.tagVariantId\") == f.col(\"right.tagVariantId\"),\n (f.col(\"right.studyType\") != \"gwas\")\n | (f.col(\"left.studyLocusId\") > f.col(\"right.studyLocusId\")),\n ],\n how=\"inner\",\n )\n .select(\n f.col(\"left.studyLocusId\").alias(\"left_studyLocusId\"),\n f.col(\"right.studyLocusId\").alias(\"right_studyLocusId\"),\n f.col(\"left.chromosome\").alias(\"chromosome\"),\n )\n .distinct()\n .repartition(\"chromosome\")\n .persist()\n )\n\n @staticmethod\n def _align_overlapping_tags(\n credset_to_overlap: DataFrame, peak_overlaps: DataFrame\n ) -> StudyLocusOverlap:\n\"\"\"Align overlapping tags in pairs of overlapping study-locus, keeping all tags in both loci.\n\n Args:\n credset_to_overlap (DataFrame): containing `studyLocusId`, `studyType`, `chromosome`, `tagVariantId`, `logABF` and `posteriorProbability` columns.\n peak_overlaps (DataFrame): containing `left_studyLocusId`, `right_studyLocusId` and `chromosome` columns.\n\n Returns:\n StudyLocusOverlap: Pairs of overlapping study-locus with aligned tags.\n \"\"\"\n # Complete information about all tags in the left study-locus of the overlap\n overlapping_left = credset_to_overlap.select(\n f.col(\"chromosome\"),\n f.col(\"tagVariantId\"),\n f.col(\"studyLocusId\").alias(\"left_studyLocusId\"),\n f.col(\"logABF\").alias(\"left_logABF\"),\n f.col(\"posteriorProbability\").alias(\"left_posteriorProbability\"),\n ).join(peak_overlaps, on=[\"chromosome\", \"left_studyLocusId\"], how=\"inner\")\n\n # Complete information about all tags in the right study-locus of the overlap\n overlapping_right = credset_to_overlap.select(\n f.col(\"chromosome\"),\n f.col(\"tagVariantId\"),\n f.col(\"studyLocusId\").alias(\"right_studyLocusId\"),\n f.col(\"logABF\").alias(\"right_logABF\"),\n f.col(\"posteriorProbability\").alias(\"right_posteriorProbability\"),\n ).join(peak_overlaps, on=[\"chromosome\", \"right_studyLocusId\"], how=\"inner\")\n\n # Include information about all tag variants in both study-locus aligned by tag variant id\n return StudyLocusOverlap(\n _df=overlapping_left.join(\n overlapping_right,\n on=[\n \"chromosome\",\n \"right_studyLocusId\",\n \"left_studyLocusId\",\n \"tagVariantId\",\n ],\n how=\"outer\",\n )\n # ensures nullable=false for following columns\n .fillna(\n value=\"unknown\",\n subset=[\n \"chromosome\",\n \"right_studyLocusId\",\n \"left_studyLocusId\",\n \"tagVariantId\",\n ],\n )\n )\n\n @staticmethod\n def _update_quality_flag(\n qc: Column, flag_condition: Column, flag_text: StudyLocusQualityCheck\n ) -> Column:\n\"\"\"Update the provided quality control list with a new flag if condition is met.\n\n Args:\n qc (Column): Array column with the current list of qc flags.\n flag_condition (Column): This is a column of booleans, signing which row should be flagged\n flag_text (StudyLocusQualityCheck): Text for the new quality control flag\n\n Returns:\n Column: Array column with the updated list of qc flags.\n \"\"\"\n qc = f.when(qc.isNull(), f.array()).otherwise(qc)\n return f.when(\n flag_condition,\n f.array_union(qc, f.array(f.lit(flag_text.value))),\n ).otherwise(qc)\n\n @classmethod\n def from_parquet(cls: type[StudyLocus], session: Session, path: str) -> StudyLocus:\n\"\"\"Initialise StudyLocus from parquet file.\n\n Args:\n session (Session): spark session\n path (str): Path to parquet file\n\n Returns:\n StudyLocus: Study-locus dataset\n \"\"\"\n df = session.read_parquet(path=path, schema=cls._schema)\n return cls(_df=df, _schema=cls._schema)\n\n def credible_set(\n self: StudyLocus,\n credible_interval: CredibleInterval,\n ) -> StudyLocus:\n\"\"\"Filter study-locus tag variants based on given credible interval.\n\n Args:\n credible_interval (CredibleInterval): Credible interval to filter for.\n\n Returns:\n StudyLocus: Filtered study-locus dataset.\n \"\"\"\n self.df = self._df.withColumn(\n \"credibleSet\",\n f.expr(f\"filter(credibleSet, tag -> (tag.{credible_interval.value}))\"),\n )\n return self\n\n def overlaps(self: StudyLocus, study_index: StudyIndex) -> StudyLocusOverlap:\n\"\"\"Calculate overlapping study-locus.\n\n Find overlapping study-locus that share at least one tagging variant. All GWAS-GWAS and all GWAS-Molecular traits are computed with the Molecular traits always\n appearing on the right side.\n\n Args:\n study_index (StudyIndex): Study index to resolve study types.\n\n Returns:\n StudyLocusOverlap: Pairs of overlapping study-locus with aligned tags.\n \"\"\"\n credset_to_overlap = (\n self.df.join(study_index.study_type_lut(), on=\"studyId\", how=\"inner\")\n .withColumn(\"credibleSet\", f.explode(\"credibleSet\"))\n .select(\n \"studyLocusId\",\n \"studyType\",\n \"chromosome\",\n f.col(\"credibleSet.tagVariantId\").alias(\"tagVariantId\"),\n f.col(\"credibleSet.logABF\").alias(\"logABF\"),\n f.col(\"credibleSet.posteriorProbability\").alias(\"posteriorProbability\"),\n )\n .persist()\n )\n\n # overlapping study-locus\n peak_overlaps = self._overlapping_peaks(credset_to_overlap)\n\n # study-locus overlap by aligning overlapping variants\n return self._align_overlapping_tags(credset_to_overlap, peak_overlaps)\n\n def unique_lead_tag_variants(self: StudyLocus) -> DataFrame:\n\"\"\"All unique lead and tag variants contained in the `StudyLocus` dataframe.\n\n Returns:\n DataFrame: A dataframe containing `variantId` and `chromosome` columns.\n \"\"\"\n lead_tags = (\n self.df.select(\n f.col(\"variantId\"),\n f.col(\"chromosome\"),\n f.explode(\"credibleSet.tagVariantId\").alias(\"tagVariantId\"),\n )\n .repartition(\"chromosome\")\n .persist()\n )\n return (\n lead_tags.select(\"variantId\", \"chromosome\")\n .union(\n lead_tags.select(f.col(\"tagVariantId\").alias(\"variantId\"), \"chromosome\")\n )\n .distinct()\n )\n\n def unique_study_locus_ancestries(\n self: StudyLocus, studies: StudyIndexGWASCatalog\n ) -> DataFrame:\n\"\"\"All unique lead variant and ancestries contained in the `StudyLocus`.\n\n Args:\n studies (StudyIndexGWASCatalog): Metadata about studies in the `StudyLocus`.\n\n Returns:\n DataFrame: unique [\"variantId\", \"studyId\", \"gnomadPopulation\", \"chromosome\", \"relativeSampleSize\"]\n\n Note:\n This method is only available for GWAS Catalog studies.\n \"\"\"\n return (\n self.df.join(\n studies.get_gnomad_ancestry_sample_sizes(), on=\"studyId\", how=\"left\"\n )\n .filter(f.col(\"position\").isNotNull())\n .select(\n \"variantId\",\n \"chromosome\",\n \"studyId\",\n \"gnomadPopulation\",\n \"relativeSampleSize\",\n )\n .distinct()\n )\n\n def neglog_pvalue(self: StudyLocus) -> Column:\n\"\"\"Returns the negative log p-value.\n\n Returns:\n Column: Negative log p-value\n \"\"\"\n return calculate_neglog_pvalue(\n self.df.pValueMantissa,\n self.df.pValueExponent,\n )\n\n def annotate_credible_sets(self: StudyLocus) -> StudyLocus:\n\"\"\"Annotate study-locus dataset with credible set flags.\n\n Sorts the array in the `credibleSet` column elements by their `posteriorProbability` values in descending order and adds\n `is95CredibleSet` and `is99CredibleSet` fields to the elements, indicating which are the tagging variants whose cumulative sum\n of their `posteriorProbability` values is below 0.95 and 0.99, respectively.\n\n Returns:\n StudyLocus: including annotation on `is95CredibleSet` and `is99CredibleSet`.\n \"\"\"\n self.df = self.df.withColumn(\n # Sort credible set by posterior probability in descending order\n \"credibleSet\",\n f.when(\n f.size(f.col(\"credibleSet\")) > 0,\n order_array_of_structs_by_field(\"credibleSet\", \"posteriorProbability\"),\n ).when(f.size(f.col(\"credibleSet\")) == 0, f.col(\"credibleSet\")),\n ).withColumn(\n # Calculate array of cumulative sums of posterior probabilities to determine which variants are in the 95% and 99% credible sets\n # and zip the cumulative sums array with the credible set array to add the flags\n \"credibleSet\",\n f.when(\n f.size(f.col(\"credibleSet\")) > 0,\n f.zip_with(\n f.col(\"credibleSet\"),\n f.transform(\n f.sequence(f.lit(1), f.size(f.col(\"credibleSet\"))),\n lambda index: f.aggregate(\n f.slice(\n # By using `index - 1` we introduce a value of `0.0` in the cumulative sums array. to ensure that the last variant\n # that exceeds the 0.95 threshold is included in the cumulative sum, as its probability is necessary to satisfy the threshold.\n f.col(\"credibleSet.posteriorProbability\"),\n 1,\n index - 1,\n ),\n f.lit(0.0),\n lambda acc, el: acc + el,\n ),\n ),\n lambda struct_e, acc: struct_e.withField(\n CredibleInterval.IS95.value, acc < 0.95\n ).withField(CredibleInterval.IS99.value, acc < 0.99),\n ),\n ).when(f.size(f.col(\"credibleSet\")) == 0, f.col(\"credibleSet\")),\n )\n return self\n\n def clump(self: StudyLocus) -> StudyLocus:\n\"\"\"Perform LD clumping of the studyLocus.\n\n Evaluates whether a lead variant is linked to a tag (with lowest p-value) in the same studyLocus dataset.\n\n Returns:\n StudyLocus: with empty credible sets for linked variants and QC flag.\n \"\"\"\n self.df = (\n self.df.withColumn(\n \"is_lead_linked\",\n LDclumping._is_lead_linked(\n self.df.studyId,\n self.df.variantId,\n self.df.pValueExponent,\n self.df.pValueMantissa,\n self.df.credibleSet,\n ),\n )\n .withColumn(\n \"credibleSet\",\n f.when(f.col(\"is_lead_linked\"), f.array()).otherwise(\n f.col(\"credibleSet\")\n ),\n )\n .withColumn(\n \"qualityControls\",\n StudyLocus._update_quality_flag(\n f.col(\"qualityControls\"),\n f.col(\"is_lead_linked\"),\n StudyLocusQualityCheck.LD_CLUMPED,\n ),\n )\n .drop(\"is_lead_linked\")\n )\n return self\n
Annotate study-locus dataset with credible set flags.
Sorts the array in the credibleSet column elements by their posteriorProbability values in descending order and adds is95CredibleSet and is99CredibleSet fields to the elements, indicating which are the tagging variants whose cumulative sum of their posteriorProbability values is below 0.95 and 0.99, respectively.
Returns:
Name Type Description StudyLocusStudyLocus
including annotation on is95CredibleSet and is99CredibleSet.
Source code in src/otg/dataset/study_locus.py
def annotate_credible_sets(self: StudyLocus) -> StudyLocus:\n\"\"\"Annotate study-locus dataset with credible set flags.\n\n Sorts the array in the `credibleSet` column elements by their `posteriorProbability` values in descending order and adds\n `is95CredibleSet` and `is99CredibleSet` fields to the elements, indicating which are the tagging variants whose cumulative sum\n of their `posteriorProbability` values is below 0.95 and 0.99, respectively.\n\n Returns:\n StudyLocus: including annotation on `is95CredibleSet` and `is99CredibleSet`.\n \"\"\"\n self.df = self.df.withColumn(\n # Sort credible set by posterior probability in descending order\n \"credibleSet\",\n f.when(\n f.size(f.col(\"credibleSet\")) > 0,\n order_array_of_structs_by_field(\"credibleSet\", \"posteriorProbability\"),\n ).when(f.size(f.col(\"credibleSet\")) == 0, f.col(\"credibleSet\")),\n ).withColumn(\n # Calculate array of cumulative sums of posterior probabilities to determine which variants are in the 95% and 99% credible sets\n # and zip the cumulative sums array with the credible set array to add the flags\n \"credibleSet\",\n f.when(\n f.size(f.col(\"credibleSet\")) > 0,\n f.zip_with(\n f.col(\"credibleSet\"),\n f.transform(\n f.sequence(f.lit(1), f.size(f.col(\"credibleSet\"))),\n lambda index: f.aggregate(\n f.slice(\n # By using `index - 1` we introduce a value of `0.0` in the cumulative sums array. to ensure that the last variant\n # that exceeds the 0.95 threshold is included in the cumulative sum, as its probability is necessary to satisfy the threshold.\n f.col(\"credibleSet.posteriorProbability\"),\n 1,\n index - 1,\n ),\n f.lit(0.0),\n lambda acc, el: acc + el,\n ),\n ),\n lambda struct_e, acc: struct_e.withField(\n CredibleInterval.IS95.value, acc < 0.95\n ).withField(CredibleInterval.IS99.value, acc < 0.99),\n ),\n ).when(f.size(f.col(\"credibleSet\")) == 0, f.col(\"credibleSet\")),\n )\n return self\n
Find overlapping study-locus that share at least one tagging variant. All GWAS-GWAS and all GWAS-Molecular traits are computed with the Molecular traits always appearing on the right side.
Parameters:
Name Type Description Default study_indexStudyIndex
Study index to resolve study types.
required
Returns:
Name Type Description StudyLocusOverlapStudyLocusOverlap
Pairs of overlapping study-locus with aligned tags.
Source code in src/otg/dataset/study_locus.py
def overlaps(self: StudyLocus, study_index: StudyIndex) -> StudyLocusOverlap:\n\"\"\"Calculate overlapping study-locus.\n\n Find overlapping study-locus that share at least one tagging variant. All GWAS-GWAS and all GWAS-Molecular traits are computed with the Molecular traits always\n appearing on the right side.\n\n Args:\n study_index (StudyIndex): Study index to resolve study types.\n\n Returns:\n StudyLocusOverlap: Pairs of overlapping study-locus with aligned tags.\n \"\"\"\n credset_to_overlap = (\n self.df.join(study_index.study_type_lut(), on=\"studyId\", how=\"inner\")\n .withColumn(\"credibleSet\", f.explode(\"credibleSet\"))\n .select(\n \"studyLocusId\",\n \"studyType\",\n \"chromosome\",\n f.col(\"credibleSet.tagVariantId\").alias(\"tagVariantId\"),\n f.col(\"credibleSet.logABF\").alias(\"logABF\"),\n f.col(\"credibleSet.posteriorProbability\").alias(\"posteriorProbability\"),\n )\n .persist()\n )\n\n # overlapping study-locus\n peak_overlaps = self._overlapping_peaks(credset_to_overlap)\n\n # study-locus overlap by aligning overlapping variants\n return self._align_overlapping_tags(credset_to_overlap, peak_overlaps)\n
Study-Locus quality control options listing concerns on the quality of the association.
Attributes:
Name Type Description SUBSIGNIFICANT_FLAGstr
p-value below significance threshold
NO_GENOMIC_LOCATION_FLAGstr
Incomplete genomic mapping
COMPOSITE_FLAGstr
Composite association due to variant x variant interactions
VARIANT_INCONSISTENCY_FLAGstr
Inconsistencies in the reported variants
NON_MAPPED_VARIANT_FLAGstr
Variant not mapped to GnomAd
PALINDROMIC_ALLELE_FLAGstr
Alleles are palindromic - cannot harmonize
AMBIGUOUS_STUDYstr
Association with ambiguous study
UNRESOLVED_LDstr
Variant not found in LD reference
LD_CLUMPEDstr
Explained by a more significant variant in high LD (clumped)
Source code in src/otg/dataset/study_locus.py
class StudyLocusQualityCheck(Enum):\n\"\"\"Study-Locus quality control options listing concerns on the quality of the association.\n\n Attributes:\n SUBSIGNIFICANT_FLAG (str): p-value below significance threshold\n NO_GENOMIC_LOCATION_FLAG (str): Incomplete genomic mapping\n COMPOSITE_FLAG (str): Composite association due to variant x variant interactions\n VARIANT_INCONSISTENCY_FLAG (str): Inconsistencies in the reported variants\n NON_MAPPED_VARIANT_FLAG (str): Variant not mapped to GnomAd\n PALINDROMIC_ALLELE_FLAG (str): Alleles are palindromic - cannot harmonize\n AMBIGUOUS_STUDY (str): Association with ambiguous study\n UNRESOLVED_LD (str): Variant not found in LD reference\n LD_CLUMPED (str): Explained by a more significant variant in high LD (clumped)\n \"\"\"\n\n SUBSIGNIFICANT_FLAG = \"Subsignificant p-value\"\n NO_GENOMIC_LOCATION_FLAG = \"Incomplete genomic mapping\"\n COMPOSITE_FLAG = \"Composite association\"\n INCONSISTENCY_FLAG = \"Variant inconsistency\"\n NON_MAPPED_VARIANT_FLAG = \"No mapping in GnomAd\"\n PALINDROMIC_ALLELE_FLAG = \"Palindrome alleles - cannot harmonize\"\n AMBIGUOUS_STUDY = \"Association with ambiguous study\"\n UNRESOLVED_LD = \"Variant not found in LD reference\"\n LD_CLUMPED = \"Explained by a more significant variant in high LD (clumped)\"\n
Interval within which an unobserved parameter value falls with a particular probability.
Attributes:
Name Type Description IS95str
95% credible interval
IS99str
99% credible interval
Source code in src/otg/dataset/study_locus.py
class CredibleInterval(Enum):\n\"\"\"Credible interval enum.\n\n Interval within which an unobserved parameter value falls with a particular probability.\n\n Attributes:\n IS95 (str): 95% credible interval\n IS99 (str): 99% credible interval\n \"\"\"\n\n IS95 = \"is95CredibleSet\"\n IS99 = \"is99CredibleSet\"\n
"},{"location":"components/dataset/study_locus/study_locus_gwas_catalog/","title":"Study locus gwas catalog","text":"
Bases: StudyLocus
Study-locus dataset derived from GWAS Catalog.
Source code in src/otg/dataset/study_locus.py
class StudyLocusGWASCatalog(StudyLocus):\n\"\"\"Study-locus dataset derived from GWAS Catalog.\"\"\"\n\n @staticmethod\n def _parse_pvalue(pvalue: Column) -> tuple[Column, Column]:\n\"\"\"Parse p-value column.\n\n Args:\n pvalue (Column): p-value [string]\n\n Returns:\n tuple[Column, Column]: p-value mantissa and exponent\n\n Example:\n >>> import pyspark.sql.types as t\n >>> d = [(\"1.0\"), (\"0.5\"), (\"1E-20\"), (\"3E-3\"), (\"1E-1000\")]\n >>> df = spark.createDataFrame(d, t.StringType())\n >>> df.select('value',*StudyLocusGWASCatalog._parse_pvalue(f.col('value'))).show()\n +-------+--------------+--------------+\n | value|pValueMantissa|pValueExponent|\n +-------+--------------+--------------+\n | 1.0| 1.0| 1|\n | 0.5| 0.5| 1|\n | 1E-20| 1.0| -20|\n | 3E-3| 3.0| -3|\n |1E-1000| 1.0| -1000|\n +-------+--------------+--------------+\n <BLANKLINE>\n\n \"\"\"\n split = f.split(pvalue, \"E\")\n return split.getItem(0).cast(\"float\").alias(\"pValueMantissa\"), f.coalesce(\n split.getItem(1).cast(\"integer\"), f.lit(1)\n ).alias(\"pValueExponent\")\n\n @staticmethod\n def _normalise_pvaluetext(p_value_text: Column) -> Column:\n\"\"\"Normalised p-value text column to a standardised format.\n\n For cases where there is no mapping, the value is set to null.\n\n Args:\n p_value_text (Column): `pValueText` column from GWASCatalog\n\n Returns:\n Column: Array column after using GWAS Catalog mappings. There might be multiple mappings for a single p-value text.\n\n Example:\n >>> import pyspark.sql.types as t\n >>> d = [(\"European Ancestry\"), (\"African ancestry\"), (\"Alzheimer\u2019s Disease\"), (\"(progression)\"), (\"\"), (None)]\n >>> df = spark.createDataFrame(d, t.StringType())\n >>> df.withColumn('normalised', StudyLocusGWASCatalog._normalise_pvaluetext(f.col('value'))).show()\n +-------------------+----------+\n | value|normalised|\n +-------------------+----------+\n | European Ancestry| [EA]|\n | African ancestry| [AA]|\n |Alzheimer\u2019s Disease| [AD]|\n | (progression)| null|\n | | null|\n | null| null|\n +-------------------+----------+\n <BLANKLINE>\n\n \"\"\"\n # GWAS Catalog to p-value mapping\n json_dict = json.loads(\n pkg_resources.read_text(data, \"gwas_pValueText_map.json\", encoding=\"utf-8\")\n )\n map_expr = f.create_map(*[f.lit(x) for x in chain(*json_dict.items())])\n\n splitted_col = f.split(f.regexp_replace(p_value_text, r\"[\\(\\)]\", \"\"), \",\")\n mapped_col = f.transform(splitted_col, lambda x: map_expr[x])\n return f.when(f.forall(mapped_col, lambda x: x.isNull()), None).otherwise(\n mapped_col\n )\n\n @staticmethod\n def _normalise_risk_allele(risk_allele: Column) -> Column:\n\"\"\"Normalised risk allele column to a standardised format.\n\n If multiple risk alleles are present, the first one is returned.\n\n Args:\n risk_allele (Column): `riskAllele` column from GWASCatalog\n\n Returns:\n Column: mapped using GWAS Catalog mapping\n\n Example:\n >>> import pyspark.sql.types as t\n >>> d = [(\"rs1234-A-G\"), (\"rs1234-A\"), (\"rs1234-A; rs1235-G\")]\n >>> df = spark.createDataFrame(d, t.StringType())\n >>> df.withColumn('normalised', StudyLocusGWASCatalog._normalise_risk_allele(f.col('value'))).show()\n +------------------+----------+\n | value|normalised|\n +------------------+----------+\n | rs1234-A-G| A|\n | rs1234-A| A|\n |rs1234-A; rs1235-G| A|\n +------------------+----------+\n <BLANKLINE>\n\n \"\"\"\n # GWAS Catalog to risk allele mapping\n return f.split(f.split(risk_allele, \"; \").getItem(0), \"-\").getItem(1)\n\n @staticmethod\n def _collect_rsids(\n snp_id: Column, snp_id_current: Column, risk_allele: Column\n ) -> Column:\n\"\"\"It takes three columns, and returns an array of distinct values from those columns.\n\n Args:\n snp_id (Column): The original snp id from the GWAS catalog.\n snp_id_current (Column): The current snp id field is just a number at the moment (stored as a string). Adding 'rs' prefix if looks good.\n risk_allele (Column): The risk allele for the SNP.\n\n Returns:\n An array of distinct values.\n \"\"\"\n # The current snp id field is just a number at the moment (stored as a string). Adding 'rs' prefix if looks good.\n snp_id_current = f.when(\n snp_id_current.rlike(\"^[0-9]*$\"),\n f.format_string(\"rs%s\", snp_id_current),\n )\n # Cleaning risk allele:\n risk_allele = f.split(risk_allele, \"-\").getItem(0)\n\n # Collecting all values:\n return f.array_distinct(f.array(snp_id, snp_id_current, risk_allele))\n\n @staticmethod\n def _map_to_variant_annotation_variants(\n gwas_associations: DataFrame, variant_annotation: VariantAnnotation\n ) -> DataFrame:\n\"\"\"Add variant metadata in associations.\n\n Args:\n gwas_associations (DataFrame): raw GWAS Catalog associations\n variant_annotation (VariantAnnotation): variant annotation dataset\n\n Returns:\n DataFrame: GWAS Catalog associations data including `variantId`, `referenceAllele`,\n `alternateAllele`, `chromosome`, `position` with variant metadata\n \"\"\"\n # Subset of GWAS Catalog associations required for resolving variant IDs:\n gwas_associations_subset = gwas_associations.select(\n \"studyLocusId\",\n f.col(\"CHR_ID\").alias(\"chromosome\"),\n f.col(\"CHR_POS\").cast(IntegerType()).alias(\"position\"),\n # List of all SNPs associated with the variant\n StudyLocusGWASCatalog._collect_rsids(\n f.split(f.col(\"SNPS\"), \"; \").getItem(0),\n f.col(\"SNP_ID_CURRENT\"),\n f.split(f.col(\"STRONGEST SNP-RISK ALLELE\"), \"; \").getItem(0),\n ).alias(\"rsIdsGwasCatalog\"),\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ).alias(\"riskAllele\"),\n )\n\n # Subset of variant annotation required for GWAS Catalog annotations:\n va_subset = variant_annotation.df.select(\n \"variantId\",\n \"chromosome\",\n \"position\",\n f.col(\"rsIds\").alias(\"rsIdsGnomad\"),\n \"referenceAllele\",\n \"alternateAllele\",\n \"alleleFrequencies\",\n variant_annotation.max_maf().alias(\"maxMaf\"),\n ).join(\n f.broadcast(\n gwas_associations_subset.select(\"chromosome\", \"position\").distinct()\n ),\n on=[\"chromosome\", \"position\"],\n how=\"inner\",\n )\n\n # Semi-resolved ids (still contains duplicates when conclusion was not possible to make\n # based on rsIds or allele concordance)\n filtered_associations = (\n gwas_associations_subset.join(\n f.broadcast(va_subset),\n on=[\"chromosome\", \"position\"],\n how=\"left\",\n )\n .withColumn(\n \"rsIdFilter\",\n StudyLocusGWASCatalog._flag_mappings_to_retain(\n f.col(\"studyLocusId\"),\n StudyLocusGWASCatalog._compare_rsids(\n f.col(\"rsIdsGnomad\"), f.col(\"rsIdsGwasCatalog\")\n ),\n ),\n )\n .withColumn(\n \"concordanceFilter\",\n StudyLocusGWASCatalog._flag_mappings_to_retain(\n f.col(\"studyLocusId\"),\n StudyLocusGWASCatalog._check_concordance(\n f.col(\"riskAllele\"),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n ),\n ),\n )\n .filter(\n # Filter out rows where GWAS Catalog rsId does not match with GnomAD rsId,\n # but there is corresponding variant for the same association\n f.col(\"rsIdFilter\")\n # or filter out rows where GWAS Catalog alleles are not concordant with GnomAD alleles,\n # but there is corresponding variant for the same association\n | f.col(\"concordanceFilter\")\n )\n )\n\n # Keep only highest maxMaf variant per studyLocusId\n fully_mapped_associations = get_record_with_maximum_value(\n filtered_associations, grouping_col=\"studyLocusId\", sorting_col=\"maxMaf\"\n ).select(\n \"studyLocusId\",\n \"variantId\",\n \"referenceAllele\",\n \"alternateAllele\",\n \"chromosome\",\n \"position\",\n )\n\n return gwas_associations.join(\n fully_mapped_associations, on=\"studyLocusId\", how=\"left\"\n )\n\n @staticmethod\n def _compare_rsids(gnomad: Column, gwas: Column) -> Column:\n\"\"\"If the intersection of the two arrays is greater than 0, return True, otherwise return False.\n\n Args:\n gnomad (Column): rsids from gnomad\n gwas (Column): rsids from the GWAS Catalog\n\n Returns:\n A boolean column that is true if the GnomAD rsIDs can be found in the GWAS rsIDs.\n\n Examples:\n >>> d = [\n ... (1, [\"rs123\", \"rs523\"], [\"rs123\"]),\n ... (2, [], [\"rs123\"]),\n ... (3, [\"rs123\", \"rs523\"], []),\n ... (4, [], []),\n ... ]\n >>> df = spark.createDataFrame(d, ['associationId', 'gnomad', 'gwas'])\n >>> df.withColumn(\"rsid_matches\", StudyLocusGWASCatalog._compare_rsids(f.col(\"gnomad\"),f.col('gwas'))).show()\n +-------------+--------------+-------+------------+\n |associationId| gnomad| gwas|rsid_matches|\n +-------------+--------------+-------+------------+\n | 1|[rs123, rs523]|[rs123]| true|\n | 2| []|[rs123]| false|\n | 3|[rs123, rs523]| []| false|\n | 4| []| []| false|\n +-------------+--------------+-------+------------+\n <BLANKLINE>\n\n \"\"\"\n return f.when(f.size(f.array_intersect(gnomad, gwas)) > 0, True).otherwise(\n False\n )\n\n @staticmethod\n def _flag_mappings_to_retain(\n association_id: Column, filter_column: Column\n ) -> Column:\n\"\"\"Flagging mappings to drop for each association.\n\n Some associations have multiple mappings. Some has matching rsId others don't. We only\n want to drop the non-matching mappings, when a matching is available for the given association.\n This logic can be generalised for other measures eg. allele concordance.\n\n Args:\n association_id (Column): association identifier column\n filter_column (Column): boolean col indicating to keep a mapping\n\n Returns:\n A column with a boolean value.\n\n Examples:\n >>> d = [\n ... (1, False),\n ... (1, False),\n ... (2, False),\n ... (2, True),\n ... (3, True),\n ... (3, True),\n ... ]\n >>> df = spark.createDataFrame(d, ['associationId', 'filter'])\n >>> df.withColumn(\"isConcordant\", StudyLocusGWASCatalog._flag_mappings_to_retain(f.col(\"associationId\"),f.col('filter'))).show()\n +-------------+------+------------+\n |associationId|filter|isConcordant|\n +-------------+------+------------+\n | 1| false| true|\n | 1| false| true|\n | 2| false| false|\n | 2| true| true|\n | 3| true| true|\n | 3| true| true|\n +-------------+------+------------+\n <BLANKLINE>\n\n \"\"\"\n w = Window.partitionBy(association_id)\n\n # Generating a boolean column informing if the filter column contains true anywhere for the association:\n aggregated_filter = f.when(\n f.array_contains(f.collect_set(filter_column).over(w), True), True\n ).otherwise(False)\n\n # Generate a filter column:\n return f.when(aggregated_filter & (~filter_column), False).otherwise(True)\n\n @staticmethod\n def _check_concordance(\n risk_allele: Column, reference_allele: Column, alternate_allele: Column\n ) -> Column:\n\"\"\"A function to check if the risk allele is concordant with the alt or ref allele.\n\n If the risk allele is the same as the reference or alternate allele, or if the reverse complement of\n the risk allele is the same as the reference or alternate allele, then the allele is concordant.\n If no mapping is available (ref/alt is null), the function returns True.\n\n Args:\n risk_allele (Column): The allele that is associated with the risk of the disease.\n reference_allele (Column): The reference allele from the GWAS catalog\n alternate_allele (Column): The alternate allele of the variant.\n\n Returns:\n A boolean column that is True if the risk allele is the same as the reference or alternate allele,\n or if the reverse complement of the risk allele is the same as the reference or alternate allele.\n\n Examples:\n >>> d = [\n ... ('A', 'A', 'G'),\n ... ('A', 'T', 'G'),\n ... ('A', 'C', 'G'),\n ... ('A', 'A', '?'),\n ... (None, None, 'A'),\n ... ]\n >>> df = spark.createDataFrame(d, ['riskAllele', 'referenceAllele', 'alternateAllele'])\n >>> df.withColumn(\"isConcordant\", StudyLocusGWASCatalog._check_concordance(f.col(\"riskAllele\"),f.col('referenceAllele'), f.col('alternateAllele'))).show()\n +----------+---------------+---------------+------------+\n |riskAllele|referenceAllele|alternateAllele|isConcordant|\n +----------+---------------+---------------+------------+\n | A| A| G| true|\n | A| T| G| true|\n | A| C| G| false|\n | A| A| ?| true|\n | null| null| A| true|\n +----------+---------------+---------------+------------+\n <BLANKLINE>\n\n \"\"\"\n # Calculating the reverse complement of the risk allele:\n risk_allele_reverse_complement = f.when(\n risk_allele.rlike(r\"^[ACTG]+$\"),\n f.reverse(f.translate(risk_allele, \"ACTG\", \"TGAC\")),\n ).otherwise(risk_allele)\n\n # OK, is the risk allele or the reverse complent is the same as the mapped alleles:\n return (\n f.when(\n (risk_allele == reference_allele) | (risk_allele == alternate_allele),\n True,\n )\n # If risk allele is found on the negative strand:\n .when(\n (risk_allele_reverse_complement == reference_allele)\n | (risk_allele_reverse_complement == alternate_allele),\n True,\n )\n # If risk allele is ambiguous, still accepted: < This condition could be reconsidered\n .when(risk_allele == \"?\", True)\n # If the association could not be mapped we keep it:\n .when(reference_allele.isNull(), True)\n # Allele is discordant:\n .otherwise(False)\n )\n\n @staticmethod\n def _get_reverse_complement(allele_col: Column) -> Column:\n\"\"\"A function to return the reverse complement of an allele column.\n\n It takes a string and returns the reverse complement of that string if it's a DNA sequence,\n otherwise it returns the original string. Assumes alleles in upper case.\n\n Args:\n allele_col (Column): The column containing the allele to reverse complement.\n\n Returns:\n A column that is the reverse complement of the allele column.\n\n Examples:\n >>> d = [{\"allele\": 'A'}, {\"allele\": 'T'},{\"allele\": 'G'}, {\"allele\": 'C'},{\"allele\": 'AC'}, {\"allele\": 'GTaatc'},{\"allele\": '?'}, {\"allele\": None}]\n >>> df = spark.createDataFrame(d)\n >>> df.withColumn(\"revcom_allele\", StudyLocusGWASCatalog._get_reverse_complement(f.col(\"allele\"))).show()\n +------+-------------+\n |allele|revcom_allele|\n +------+-------------+\n | A| T|\n | T| A|\n | G| C|\n | C| G|\n | AC| GT|\n |GTaatc| GATTAC|\n | ?| ?|\n | null| null|\n +------+-------------+\n <BLANKLINE>\n\n \"\"\"\n allele_col = f.upper(allele_col)\n return f.when(\n allele_col.rlike(\"[ACTG]+\"),\n f.reverse(f.translate(allele_col, \"ACTG\", \"TGAC\")),\n ).otherwise(allele_col)\n\n @staticmethod\n def _effect_needs_harmonisation(\n risk_allele: Column, reference_allele: Column\n ) -> Column:\n\"\"\"A function to check if the effect allele needs to be harmonised.\n\n Args:\n risk_allele (Column): Risk allele column\n reference_allele (Column): Effect allele column\n\n Returns:\n A boolean column indicating if the effect allele needs to be harmonised.\n\n Examples:\n >>> d = [{\"risk\": 'A', \"reference\": 'A'}, {\"risk\": 'A', \"reference\": 'T'}, {\"risk\": 'AT', \"reference\": 'TA'}, {\"risk\": 'AT', \"reference\": 'AT'}]\n >>> df = spark.createDataFrame(d)\n >>> df.withColumn(\"needs_harmonisation\", StudyLocusGWASCatalog._effect_needs_harmonisation(f.col(\"risk\"), f.col(\"reference\"))).show()\n +---------+----+-------------------+\n |reference|risk|needs_harmonisation|\n +---------+----+-------------------+\n | A| A| true|\n | T| A| true|\n | TA| AT| false|\n | AT| AT| true|\n +---------+----+-------------------+\n <BLANKLINE>\n\n \"\"\"\n return (risk_allele == reference_allele) | (\n risk_allele\n == StudyLocusGWASCatalog._get_reverse_complement(reference_allele)\n )\n\n @staticmethod\n def _are_alleles_palindromic(\n reference_allele: Column, alternate_allele: Column\n ) -> Column:\n\"\"\"A function to check if the alleles are palindromic.\n\n Args:\n reference_allele (Column): Reference allele column\n alternate_allele (Column): Alternate allele column\n\n Returns:\n A boolean column indicating if the alleles are palindromic.\n\n Examples:\n >>> d = [{\"reference\": 'A', \"alternate\": 'T'}, {\"reference\": 'AT', \"alternate\": 'AG'}, {\"reference\": 'AT', \"alternate\": 'AT'}, {\"reference\": 'CATATG', \"alternate\": 'CATATG'}, {\"reference\": '-', \"alternate\": None}]\n >>> df = spark.createDataFrame(d)\n >>> df.withColumn(\"is_palindromic\", StudyLocusGWASCatalog._are_alleles_palindromic(f.col(\"reference\"), f.col(\"alternate\"))).show()\n +---------+---------+--------------+\n |alternate|reference|is_palindromic|\n +---------+---------+--------------+\n | T| A| true|\n | AG| AT| false|\n | AT| AT| true|\n | CATATG| CATATG| true|\n | null| -| false|\n +---------+---------+--------------+\n <BLANKLINE>\n\n \"\"\"\n revcomp = StudyLocusGWASCatalog._get_reverse_complement(alternate_allele)\n return (\n f.when(reference_allele == revcomp, True)\n .when(revcomp.isNull(), False)\n .otherwise(False)\n )\n\n @staticmethod\n def _harmonise_beta(\n risk_allele: Column,\n reference_allele: Column,\n alternate_allele: Column,\n effect_size: Column,\n confidence_interval: Column,\n ) -> Column:\n\"\"\"A function to extract the beta value from the effect size and confidence interval.\n\n If the confidence interval contains the word \"increase\" or \"decrease\" it indicates, we are dealing with betas.\n If it's \"increase\" and the effect size needs to be harmonized, then multiply the effect size by -1\n\n Args:\n risk_allele (Column): Risk allele column\n reference_allele (Column): Reference allele column\n alternate_allele (Column): Alternate allele column\n effect_size (Column): GWAS Catalog effect size column\n confidence_interval (Column): GWAS Catalog confidence interval column\n\n Returns:\n A column containing the beta value.\n \"\"\"\n return (\n f.when(\n StudyLocusGWASCatalog._are_alleles_palindromic(\n reference_allele, alternate_allele\n ),\n None,\n )\n .when(\n (\n StudyLocusGWASCatalog._effect_needs_harmonisation(\n risk_allele, reference_allele\n )\n & confidence_interval.contains(\"increase\")\n )\n | (\n ~StudyLocusGWASCatalog._effect_needs_harmonisation(\n risk_allele, reference_allele\n )\n & confidence_interval.contains(\"decrease\")\n ),\n -effect_size,\n )\n .otherwise(effect_size)\n .cast(DoubleType())\n )\n\n @staticmethod\n def _harmonise_beta_ci(\n risk_allele: Column,\n reference_allele: Column,\n alternate_allele: Column,\n effect_size: Column,\n confidence_interval: Column,\n p_value: Column,\n direction: str,\n ) -> Column:\n\"\"\"Calculating confidence intervals for beta values.\n\n Args:\n risk_allele (Column): Risk allele column\n reference_allele (Column): Reference allele column\n alternate_allele (Column): Alternate allele column\n effect_size (Column): GWAS Catalog effect size column\n confidence_interval (Column): GWAS Catalog confidence interval column\n p_value (Column): GWAS Catalog p-value column\n direction (str): This is the direction of the confidence interval. It can be either \"upper\" or \"lower\".\n\n Returns:\n The upper and lower bounds of the confidence interval for the beta coefficient.\n \"\"\"\n zscore_95 = f.lit(1.96)\n beta = StudyLocusGWASCatalog._harmonise_beta(\n risk_allele,\n reference_allele,\n alternate_allele,\n effect_size,\n confidence_interval,\n )\n zscore = pvalue_to_zscore(p_value)\n return (\n f.when(f.lit(direction) == \"upper\", beta + f.abs(zscore_95 * beta) / zscore)\n .when(f.lit(direction) == \"lower\", beta - f.abs(zscore_95 * beta) / zscore)\n .otherwise(None)\n )\n\n @staticmethod\n def _harmonise_odds_ratio(\n risk_allele: Column,\n reference_allele: Column,\n alternate_allele: Column,\n effect_size: Column,\n confidence_interval: Column,\n ) -> Column:\n\"\"\"Harmonizing odds ratio.\n\n Args:\n risk_allele (Column): Risk allele column\n reference_allele (Column): Reference allele column\n alternate_allele (Column): Alternate allele column\n effect_size (Column): GWAS Catalog effect size column\n confidence_interval (Column): GWAS Catalog confidence interval column\n\n Returns:\n A column with the odds ratio, or 1/odds_ratio if harmonization required.\n \"\"\"\n return (\n f.when(\n StudyLocusGWASCatalog._are_alleles_palindromic(\n reference_allele, alternate_allele\n ),\n None,\n )\n .when(\n (\n StudyLocusGWASCatalog._effect_needs_harmonisation(\n risk_allele, reference_allele\n )\n & ~confidence_interval.rlike(\"|\".join([\"decrease\", \"increase\"]))\n ),\n 1 / effect_size,\n )\n .otherwise(effect_size)\n .cast(DoubleType())\n )\n\n @staticmethod\n def _harmonise_odds_ratio_ci(\n risk_allele: Column,\n reference_allele: Column,\n alternate_allele: Column,\n effect_size: Column,\n confidence_interval: Column,\n p_value: Column,\n direction: str,\n ) -> Column:\n\"\"\"Calculating confidence intervals for beta values.\n\n Args:\n risk_allele (Column): Risk allele column\n reference_allele (Column): Reference allele column\n alternate_allele (Column): Alternate allele column\n effect_size (Column): GWAS Catalog effect size column\n confidence_interval (Column): GWAS Catalog confidence interval column\n p_value (Column): GWAS Catalog p-value column\n direction (str): This is the direction of the confidence interval. It can be either \"upper\" or \"lower\".\n\n Returns:\n The upper and lower bounds of the 95% confidence interval for the odds ratio.\n \"\"\"\n zscore_95 = f.lit(1.96)\n odds_ratio = StudyLocusGWASCatalog._harmonise_odds_ratio(\n risk_allele,\n reference_allele,\n alternate_allele,\n effect_size,\n confidence_interval,\n )\n odds_ratio_estimate = f.log(odds_ratio)\n zscore = pvalue_to_zscore(p_value)\n odds_ratio_se = odds_ratio_estimate / zscore\n return f.when(\n f.lit(direction) == \"upper\",\n f.exp(odds_ratio_estimate + f.abs(zscore_95 * odds_ratio_se)),\n ).when(\n f.lit(direction) == \"lower\",\n f.exp(odds_ratio_estimate - f.abs(zscore_95 * odds_ratio_se)),\n )\n\n @staticmethod\n def _concatenate_substudy_description(\n association_trait: Column, pvalue_text: Column, mapped_trait_uri: Column\n ) -> Column:\n\"\"\"Substudy description parsing. Complex string containing metadata about the substudy (e.g. QTL, specific EFO, etc.).\n\n Args:\n association_trait (Column): GWAS Catalog association trait column\n pvalue_text (Column): GWAS Catalog p-value text column\n mapped_trait_uri (Column): GWAS Catalog mapped trait URI column\n\n Returns:\n A column with the substudy description in the shape trait|pvaluetext1_pvaluetext2|EFO1_EFO2.\n\n Examples:\n >>> df = spark.createDataFrame([\n ... (\"Height\", \"http://www.ebi.ac.uk/efo/EFO_0000001,http://www.ebi.ac.uk/efo/EFO_0000002\", \"European Ancestry\"),\n ... (\"Schizophrenia\", \"http://www.ebi.ac.uk/efo/MONDO_0005090\", None)],\n ... [\"association_trait\", \"mapped_trait_uri\", \"pvalue_text\"]\n ... )\n >>> df.withColumn('substudy_description', StudyLocusGWASCatalog._concatenate_substudy_description(df.association_trait, df.pvalue_text, df.mapped_trait_uri)).show(truncate=False)\n +-----------------+-------------------------------------------------------------------------+-----------------+------------------------------------------+\n |association_trait|mapped_trait_uri |pvalue_text |substudy_description |\n +-----------------+-------------------------------------------------------------------------+-----------------+------------------------------------------+\n |Height |http://www.ebi.ac.uk/efo/EFO_0000001,http://www.ebi.ac.uk/efo/EFO_0000002|European Ancestry|Height|EA|EFO_0000001/EFO_0000002 |\n |Schizophrenia |http://www.ebi.ac.uk/efo/MONDO_0005090 |null |Schizophrenia|no_pvalue_text|MONDO_0005090|\n +-----------------+-------------------------------------------------------------------------+-----------------+------------------------------------------+\n <BLANKLINE>\n \"\"\"\n p_value_text = f.coalesce(\n StudyLocusGWASCatalog._normalise_pvaluetext(pvalue_text),\n f.array(f.lit(\"no_pvalue_text\")),\n )\n return f.concat_ws(\n \"|\",\n association_trait,\n f.concat_ws(\n \"/\",\n p_value_text,\n ),\n f.concat_ws(\n \"/\",\n parse_efos(mapped_trait_uri),\n ),\n )\n\n @staticmethod\n def _qc_all(\n qc: Column,\n chromosome: Column,\n position: Column,\n reference_allele: Column,\n alternate_allele: Column,\n strongest_snp_risk_allele: Column,\n p_value_mantissa: Column,\n p_value_exponent: Column,\n p_value_cutoff: float,\n ) -> Column:\n\"\"\"Flag associations that fail any QC.\n\n Args:\n qc (Column): QC column\n chromosome (Column): Chromosome column\n position (Column): Position column\n reference_allele (Column): Reference allele column\n alternate_allele (Column): Alternate allele column\n strongest_snp_risk_allele (Column): Strongest SNP risk allele column\n p_value_mantissa (Column): P-value mantissa column\n p_value_exponent (Column): P-value exponent column\n p_value_cutoff (float): P-value cutoff\n\n Returns:\n Column: Updated QC column with flag.\n \"\"\"\n qc = StudyLocusGWASCatalog._qc_variant_interactions(\n qc, strongest_snp_risk_allele\n )\n qc = StudyLocusGWASCatalog._qc_subsignificant_associations(\n qc, p_value_mantissa, p_value_exponent, p_value_cutoff\n )\n qc = StudyLocusGWASCatalog._qc_genomic_location(qc, chromosome, position)\n qc = StudyLocusGWASCatalog._qc_variant_inconsistencies(\n qc, chromosome, position, strongest_snp_risk_allele\n )\n qc = StudyLocusGWASCatalog._qc_unmapped_variants(qc, alternate_allele)\n qc = StudyLocusGWASCatalog._qc_palindromic_alleles(\n qc, reference_allele, alternate_allele\n )\n return qc\n\n @staticmethod\n def _qc_variant_interactions(\n qc: Column, strongest_snp_risk_allele: Column\n ) -> Column:\n\"\"\"Flag associations based on variant x variant interactions.\n\n Args:\n qc (Column): QC column\n strongest_snp_risk_allele (Column): Column with the strongest SNP risk allele\n\n Returns:\n Column: Updated QC column with flag.\n \"\"\"\n return StudyLocusGWASCatalog._update_quality_flag(\n qc,\n strongest_snp_risk_allele.contains(\";\"),\n StudyLocusQualityCheck.COMPOSITE_FLAG,\n )\n\n @staticmethod\n def _qc_subsignificant_associations(\n qc: Column,\n p_value_mantissa: Column,\n p_value_exponent: Column,\n pvalue_cutoff: float,\n ) -> Column:\n\"\"\"Flag associations below significant threshold.\n\n Args:\n qc (Column): QC column\n p_value_mantissa (Column): P-value mantissa column\n p_value_exponent (Column): P-value exponent column\n pvalue_cutoff (float): association p-value cut-off\n\n Returns:\n Column: Updated QC column with flag.\n\n Examples:\n >>> import pyspark.sql.types as t\n >>> d = [{'qc': None, 'p_value_mantissa': 1, 'p_value_exponent': -7}, {'qc': None, 'p_value_mantissa': 1, 'p_value_exponent': -8}, {'qc': None, 'p_value_mantissa': 5, 'p_value_exponent': -8}, {'qc': None, 'p_value_mantissa': 1, 'p_value_exponent': -9}]\n >>> df = spark.createDataFrame(d, t.StructType([t.StructField('qc', t.ArrayType(t.StringType()), True), t.StructField('p_value_mantissa', t.IntegerType()), t.StructField('p_value_exponent', t.IntegerType())]))\n >>> df.withColumn('qc', StudyLocusGWASCatalog._qc_subsignificant_associations(f.col(\"qc\"), f.col(\"p_value_mantissa\"), f.col(\"p_value_exponent\"), 5e-8)).show(truncate = False)\n +------------------------+----------------+----------------+\n |qc |p_value_mantissa|p_value_exponent|\n +------------------------+----------------+----------------+\n |[Subsignificant p-value]|1 |-7 |\n |[] |1 |-8 |\n |[] |5 |-8 |\n |[] |1 |-9 |\n +------------------------+----------------+----------------+\n <BLANKLINE>\n\n \"\"\"\n return StudyLocus._update_quality_flag(\n qc,\n calculate_neglog_pvalue(p_value_mantissa, p_value_exponent)\n < f.lit(-np.log10(pvalue_cutoff)),\n StudyLocusQualityCheck.SUBSIGNIFICANT_FLAG,\n )\n\n @staticmethod\n def _qc_genomic_location(\n qc: Column, chromosome: Column, position: Column\n ) -> Column:\n\"\"\"Flag associations without genomic location in GWAS Catalog.\n\n Args:\n qc (Column): QC column\n chromosome (Column): Chromosome column in GWAS Catalog\n position (Column): Position column in GWAS Catalog\n\n Returns:\n Column: Updated QC column with flag.\n\n Examples:\n >>> import pyspark.sql.types as t\n >>> d = [{'qc': None, 'chromosome': None, 'position': None}, {'qc': None, 'chromosome': '1', 'position': None}, {'qc': None, 'chromosome': None, 'position': 1}, {'qc': None, 'chromosome': '1', 'position': 1}]\n >>> df = spark.createDataFrame(d, schema=t.StructType([t.StructField('qc', t.ArrayType(t.StringType()), True), t.StructField('chromosome', t.StringType()), t.StructField('position', t.IntegerType())]))\n >>> df.withColumn('qc', StudyLocusGWASCatalog._qc_genomic_location(df.qc, df.chromosome, df.position)).show(truncate=False)\n +----------------------------+----------+--------+\n |qc |chromosome|position|\n +----------------------------+----------+--------+\n |[Incomplete genomic mapping]|null |null |\n |[Incomplete genomic mapping]|1 |null |\n |[Incomplete genomic mapping]|null |1 |\n |[] |1 |1 |\n +----------------------------+----------+--------+\n <BLANKLINE>\n\n \"\"\"\n return StudyLocus._update_quality_flag(\n qc,\n position.isNull() | chromosome.isNull(),\n StudyLocusQualityCheck.NO_GENOMIC_LOCATION_FLAG,\n )\n\n @staticmethod\n def _qc_variant_inconsistencies(\n qc: Column,\n chromosome: Column,\n position: Column,\n strongest_snp_risk_allele: Column,\n ) -> Column:\n\"\"\"Flag associations with inconsistencies in the variant annotation.\n\n Args:\n qc (Column): QC column\n chromosome (Column): Chromosome column in GWAS Catalog\n position (Column): Position column in GWAS Catalog\n strongest_snp_risk_allele (Column): Strongest SNP risk allele column in GWAS Catalog\n\n Returns:\n Column: Updated QC column with flag.\n \"\"\"\n return StudyLocusGWASCatalog._update_quality_flag(\n qc,\n # Number of chromosomes does not correspond to the number of positions:\n (f.size(f.split(chromosome, \";\")) != f.size(f.split(position, \";\")))\n # Number of chromosome values different from riskAllele values:\n | (\n f.size(f.split(chromosome, \";\"))\n != f.size(f.split(strongest_snp_risk_allele, \";\"))\n ),\n StudyLocusQualityCheck.INCONSISTENCY_FLAG,\n )\n\n @staticmethod\n def _qc_unmapped_variants(qc: Column, alternate_allele: Column) -> Column:\n\"\"\"Flag associations with variants not mapped to variantAnnotation.\n\n Args:\n qc (Column): QC column\n alternate_allele (Column): alternate allele\n\n Returns:\n Column: Updated QC column with flag.\n\n Example:\n >>> import pyspark.sql.types as t\n >>> d = [{'alternate_allele': 'A', 'qc': None}, {'alternate_allele': None, 'qc': None}]\n >>> schema = t.StructType([t.StructField('alternate_allele', t.StringType(), True), t.StructField('qc', t.ArrayType(t.StringType()), True)])\n >>> df = spark.createDataFrame(data=d, schema=schema)\n >>> df.withColumn(\"new_qc\", StudyLocusGWASCatalog._qc_unmapped_variants(f.col(\"qc\"), f.col(\"alternate_allele\"))).show()\n +----------------+----+--------------------+\n |alternate_allele| qc| new_qc|\n +----------------+----+--------------------+\n | A|null| []|\n | null|null|[No mapping in Gn...|\n +----------------+----+--------------------+\n <BLANKLINE>\n\n \"\"\"\n return StudyLocus._update_quality_flag(\n qc,\n alternate_allele.isNull(),\n StudyLocusQualityCheck.NON_MAPPED_VARIANT_FLAG,\n )\n\n @staticmethod\n def _qc_palindromic_alleles(\n qc: Column, reference_allele: Column, alternate_allele: Column\n ) -> Column:\n\"\"\"Flag associations with palindromic variants which effects can not be harmonised.\n\n Args:\n qc (Column): QC column\n reference_allele (Column): reference allele\n alternate_allele (Column): alternate allele\n\n Returns:\n Column: Updated QC column with flag.\n\n Example:\n >>> import pyspark.sql.types as t\n >>> schema = t.StructType([t.StructField('reference_allele', t.StringType(), True), t.StructField('alternate_allele', t.StringType(), True), t.StructField('qc', t.ArrayType(t.StringType()), True)])\n >>> d = [{'reference_allele': 'A', 'alternate_allele': 'T', 'qc': None}, {'reference_allele': 'AT', 'alternate_allele': 'TA', 'qc': None}, {'reference_allele': 'AT', 'alternate_allele': 'AT', 'qc': None}]\n >>> df = spark.createDataFrame(data=d, schema=schema)\n >>> df.withColumn(\"qc\", StudyLocusGWASCatalog._qc_palindromic_alleles(f.col(\"qc\"), f.col(\"reference_allele\"), f.col(\"alternate_allele\"))).show(truncate=False)\n +----------------+----------------+---------------------------------------+\n |reference_allele|alternate_allele|qc |\n +----------------+----------------+---------------------------------------+\n |A |T |[Palindrome alleles - cannot harmonize]|\n |AT |TA |[] |\n |AT |AT |[Palindrome alleles - cannot harmonize]|\n +----------------+----------------+---------------------------------------+\n <BLANKLINE>\n\n \"\"\"\n return StudyLocus._update_quality_flag(\n qc,\n StudyLocusGWASCatalog._are_alleles_palindromic(\n reference_allele, alternate_allele\n ),\n StudyLocusQualityCheck.PALINDROMIC_ALLELE_FLAG,\n )\n\n @classmethod\n def from_source(\n cls: type[StudyLocusGWASCatalog],\n gwas_associations: DataFrame,\n variant_annotation: VariantAnnotation,\n pvalue_threshold: float = 5e-8,\n ) -> StudyLocusGWASCatalog:\n\"\"\"Read GWASCatalog associations.\n\n It reads the GWAS Catalog association dataset, selects and renames columns, casts columns, and\n applies some pre-defined filters on the data:\n\n Args:\n gwas_associations (DataFrame): GWAS Catalog raw associations dataset\n variant_annotation (VariantAnnotation): Variant annotation dataset\n pvalue_threshold (float): P-value threshold for flagging associations\n\n Returns:\n StudyLocusGWASCatalog: StudyLocusGWASCatalog dataset\n \"\"\"\n return cls(\n _df=gwas_associations.withColumn(\n \"studyLocusId\", f.monotonically_increasing_id().cast(LongType())\n )\n .transform(\n # Map/harmonise variants to variant annotation dataset:\n # This function adds columns: variantId, referenceAllele, alternateAllele, chromosome, position\n lambda df: StudyLocusGWASCatalog._map_to_variant_annotation_variants(\n df, variant_annotation\n )\n )\n .withColumn(\n # Perform all quality control checks:\n \"qualityControls\",\n StudyLocusGWASCatalog._qc_all(\n f.array().alias(\"qualityControls\"),\n f.col(\"CHR_ID\"),\n f.col(\"CHR_POS\").cast(IntegerType()),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"STRONGEST SNP-RISK ALLELE\"),\n *StudyLocusGWASCatalog._parse_pvalue(f.col(\"P-VALUE\")),\n pvalue_threshold,\n ),\n )\n .select(\n # INSIDE STUDY-LOCUS SCHEMA:\n \"studyLocusId\",\n \"variantId\",\n # Mapped genomic location of the variant (; separated list)\n \"chromosome\",\n \"position\",\n f.col(\"STUDY ACCESSION\").alias(\"studyId\"),\n # beta value of the association\n StudyLocusGWASCatalog._harmonise_beta(\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"OR or BETA\"),\n f.col(\"95% CI (TEXT)\"),\n ).alias(\"beta\"),\n # odds ratio of the association\n StudyLocusGWASCatalog._harmonise_odds_ratio(\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"OR or BETA\"),\n f.col(\"95% CI (TEXT)\"),\n ).alias(\"oddsRatio\"),\n # CI lower of the beta value\n StudyLocusGWASCatalog._harmonise_beta_ci(\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"OR or BETA\"),\n f.col(\"95% CI (TEXT)\"),\n f.col(\"P-VALUE\"),\n \"lower\",\n ).alias(\"betaConfidenceIntervalLower\"),\n # CI upper for the beta value\n StudyLocusGWASCatalog._harmonise_beta_ci(\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"OR or BETA\"),\n f.col(\"95% CI (TEXT)\"),\n f.col(\"P-VALUE\"),\n \"upper\",\n ).alias(\"betaConfidenceIntervalUpper\"),\n # CI lower of the odds ratio value\n StudyLocusGWASCatalog._harmonise_odds_ratio_ci(\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"OR or BETA\"),\n f.col(\"95% CI (TEXT)\"),\n f.col(\"P-VALUE\"),\n \"lower\",\n ).alias(\"oddsRatioConfidenceIntervalLower\"),\n # CI upper of the odds ratio value\n StudyLocusGWASCatalog._harmonise_odds_ratio_ci(\n StudyLocusGWASCatalog._normalise_risk_allele(\n f.col(\"STRONGEST SNP-RISK ALLELE\")\n ),\n f.col(\"referenceAllele\"),\n f.col(\"alternateAllele\"),\n f.col(\"OR or BETA\"),\n f.col(\"95% CI (TEXT)\"),\n f.col(\"P-VALUE\"),\n \"upper\",\n ).alias(\"oddsRatioConfidenceIntervalUpper\"),\n # p-value of the association, string: split into exponent and mantissa.\n *StudyLocusGWASCatalog._parse_pvalue(f.col(\"P-VALUE\")),\n # Capturing phenotype granularity at the association level\n StudyLocusGWASCatalog._concatenate_substudy_description(\n f.col(\"DISEASE/TRAIT\"),\n f.col(\"P-VALUE (TEXT)\"),\n f.col(\"MAPPED_TRAIT_URI\"),\n ).alias(\"subStudyDescription\"),\n # Quality controls (array of strings)\n \"qualityControls\",\n )\n )\n\n def update_study_id(\n self: StudyLocusGWASCatalog, study_annotation: DataFrame\n ) -> StudyLocusGWASCatalog:\n\"\"\"Update studyId with a dataframe containing study.\n\n Args:\n study_annotation (DataFrame): Dataframe containing `updatedStudyId` and key columns `studyId` and `subStudyDescription`.\n\n Returns:\n StudyLocusGWASCatalog: Updated study locus.\n \"\"\"\n self.df = (\n self._df.join(\n study_annotation, on=[\"studyId\", \"subStudyDescription\"], how=\"left\"\n )\n .withColumn(\"studyId\", f.coalesce(\"updatedStudyId\", \"studyId\"))\n .drop(\"subStudyDescription\", \"updatedStudyId\")\n )\n return self\n\n def annotate_ld(\n self: StudyLocusGWASCatalog,\n session: Session,\n studies: StudyIndexGWASCatalog,\n ld_populations: list[str],\n ld_index_template: str,\n ld_matrix_template: str,\n min_r2: float,\n ) -> StudyLocus:\n\"\"\"Annotate LD set for every studyLocus using gnomAD.\n\n Args:\n session (Session): Session\n studies (StudyIndexGWASCatalog): Study index containing ancestry information\n ld_populations (list[str]): List of populations to annotate\n ld_index_template (str): Template path of the LD matrix index containing `{POP}` where the population is expected\n ld_matrix_template (str): Template path of the LD matrix containing `{POP}` where the population is expected\n min_r2 (float): Minimum r2 to include in the LD set\n\n Returns:\n StudyLocus: Study-locus with an annotated credible set.\n \"\"\"\n # TODO: call unique_study_locus_ancestries here so that it is not duplicated with ld_annotation_by_locus_ancestry\n # LD annotation for all unique lead variants in all populations (study independent).\n ld_r = LDAnnotatorGnomad.ld_annotation_by_locus_ancestry(\n session,\n self,\n studies,\n ld_populations,\n ld_index_template,\n ld_matrix_template,\n min_r2,\n ).coalesce(400)\n\n ld_set = (\n self.unique_study_locus_ancestries(studies)\n .join(ld_r, on=[\"chromosome\", \"variantId\", \"gnomadPopulation\"], how=\"left\")\n .withColumn(\"r2\", f.pow(f.col(\"r\"), f.lit(2)))\n .withColumn(\n \"r2Overall\",\n LDAnnotatorGnomad.weighted_r_overall(\n f.col(\"chromosome\"),\n f.col(\"studyId\"),\n f.col(\"variantId\"),\n f.col(\"tagVariantId\"),\n f.col(\"relativeSampleSize\"),\n f.col(\"r2\"),\n ),\n )\n .groupBy(\"chromosome\", \"studyId\", \"variantId\")\n .agg(\n f.collect_set(\n f.when(\n f.col(\"tagVariantId\").isNotNull(),\n f.struct(\"tagVariantId\", \"r2Overall\"),\n )\n ).alias(\"credibleSet\")\n )\n )\n\n self.df = self.df.join(\n ld_set, on=[\"chromosome\", \"studyId\", \"variantId\"], how=\"left\"\n )\n\n return self._qc_unresolved_ld()\n\n def _qc_ambiguous_study(self: StudyLocusGWASCatalog) -> StudyLocusGWASCatalog:\n\"\"\"Flag associations with variants that can not be unambiguously associated with one study.\n\n Returns:\n StudyLocusGWASCatalog: Updated study locus.\n \"\"\"\n assoc_ambiguity_window = Window.partitionBy(\n f.col(\"studyId\"), f.col(\"variantId\")\n )\n\n self._df.withColumn(\n \"qualityControls\",\n StudyLocus._update_quality_flag(\n f.col(\"qualityControls\"),\n f.count(f.col(\"variantId\")).over(assoc_ambiguity_window) > 1,\n StudyLocusQualityCheck.AMBIGUOUS_STUDY,\n ),\n )\n return self\n\n def _qc_unresolved_ld(self: StudyLocusGWASCatalog) -> StudyLocusGWASCatalog:\n\"\"\"Flag associations with variants that are not found in the LD reference.\n\n Returns:\n StudyLocusGWASCatalog: Updated study locus.\n \"\"\"\n self._df.withColumn(\n \"qualityControls\",\n StudyLocus._update_quality_flag(\n f.col(\"qualityControls\"),\n f.col(\"credibleSet\").isNull(),\n StudyLocusQualityCheck.UNRESOLVED_LD,\n ),\n )\n return self\n
Annotate LD set for every studyLocus using gnomAD.
Parameters:
Name Type Description Default sessionSession
Session
required studiesStudyIndexGWASCatalog
Study index containing ancestry information
required ld_populationslist[str]
List of populations to annotate
required ld_index_templatestr
Template path of the LD matrix index containing {POP} where the population is expected
required ld_matrix_templatestr
Template path of the LD matrix containing {POP} where the population is expected
required min_r2float
Minimum r2 to include in the LD set
required
Returns:
Name Type Description StudyLocusStudyLocus
Study-locus with an annotated credible set.
Source code in src/otg/dataset/study_locus.py
def annotate_ld(\n self: StudyLocusGWASCatalog,\n session: Session,\n studies: StudyIndexGWASCatalog,\n ld_populations: list[str],\n ld_index_template: str,\n ld_matrix_template: str,\n min_r2: float,\n) -> StudyLocus:\n\"\"\"Annotate LD set for every studyLocus using gnomAD.\n\n Args:\n session (Session): Session\n studies (StudyIndexGWASCatalog): Study index containing ancestry information\n ld_populations (list[str]): List of populations to annotate\n ld_index_template (str): Template path of the LD matrix index containing `{POP}` where the population is expected\n ld_matrix_template (str): Template path of the LD matrix containing `{POP}` where the population is expected\n min_r2 (float): Minimum r2 to include in the LD set\n\n Returns:\n StudyLocus: Study-locus with an annotated credible set.\n \"\"\"\n # TODO: call unique_study_locus_ancestries here so that it is not duplicated with ld_annotation_by_locus_ancestry\n # LD annotation for all unique lead variants in all populations (study independent).\n ld_r = LDAnnotatorGnomad.ld_annotation_by_locus_ancestry(\n session,\n self,\n studies,\n ld_populations,\n ld_index_template,\n ld_matrix_template,\n min_r2,\n ).coalesce(400)\n\n ld_set = (\n self.unique_study_locus_ancestries(studies)\n .join(ld_r, on=[\"chromosome\", \"variantId\", \"gnomadPopulation\"], how=\"left\")\n .withColumn(\"r2\", f.pow(f.col(\"r\"), f.lit(2)))\n .withColumn(\n \"r2Overall\",\n LDAnnotatorGnomad.weighted_r_overall(\n f.col(\"chromosome\"),\n f.col(\"studyId\"),\n f.col(\"variantId\"),\n f.col(\"tagVariantId\"),\n f.col(\"relativeSampleSize\"),\n f.col(\"r2\"),\n ),\n )\n .groupBy(\"chromosome\", \"studyId\", \"variantId\")\n .agg(\n f.collect_set(\n f.when(\n f.col(\"tagVariantId\").isNotNull(),\n f.struct(\"tagVariantId\", \"r2Overall\"),\n )\n ).alias(\"credibleSet\")\n )\n )\n\n self.df = self.df.join(\n ld_set, on=[\"chromosome\", \"studyId\", \"variantId\"], how=\"left\"\n )\n\n return self._qc_unresolved_ld()\n
Clumping is a commonly used post-processing method that allows for identification of independent association signals from GWAS summary statistics and curated associations. This process is critical because of the complex linkage disequilibrium (LD) structure in human populations, which can result in multiple statistically significant associations within the same genomic region. Clumping methods help reduce redundancy in GWAS results and ensure that each reported association represents an independent signal.
We have implemented 2 clumping methods:
"},{"location":"components/method/clumping/#clumping-based-on-linkage-disequilibrium-ld","title":"Clumping based on Linkage Disequilibrium (LD)","text":"
LD clumping reports the most significant genetic associations in a region in terms of a smaller number of \u201cclumps\u201d of genetically linked SNPs.
Source code in src/otg/method/clump.py
class LDclumping:\n\"\"\"LD clumping reports the most significant genetic associations in a region in terms of a smaller number of \u201cclumps\u201d of genetically linked SNPs.\"\"\"\n\n @staticmethod\n def _is_lead_linked(\n study_id: Column,\n variant_id: Column,\n p_value_exponent: Column,\n p_value_mantissa: Column,\n credible_set: Column,\n ) -> Column:\n\"\"\"Evaluates whether a lead variant is linked to a tag (with lowest p-value) in the same studyLocus dataset.\n\n Args:\n study_id (Column): studyId\n variant_id (Column): Lead variant id\n p_value_exponent (Column): p-value exponent\n p_value_mantissa (Column): p-value mantissa\n credible_set (Column): Credible set <array of structs>\n\n Returns:\n Column: Boolean in which True indicates that the lead is linked to another tag in the same dataset.\n \"\"\"\n leads_in_study = f.collect_set(variant_id).over(Window.partitionBy(study_id))\n tags_in_studylocus = f.array_union(\n # Get all tag variants from the credible set per studyLocusId\n f.transform(credible_set, lambda x: x.tagVariantId),\n # And append the lead variant so that the intersection is the same for all studyLocusIds in a study\n f.array(f.col(\"variantId\")),\n )\n intersect_lead_tags = f.array_sort(\n f.array_intersect(leads_in_study, tags_in_studylocus)\n )\n return (\n # If the lead is in the credible set, we rank the peaks by p-value\n f.when(\n f.size(intersect_lead_tags) > 0,\n f.row_number().over(\n Window.partitionBy(study_id, intersect_lead_tags).orderBy(\n p_value_exponent, p_value_mantissa\n )\n )\n > 1,\n )\n # If the intersection is empty (lead is not in the credible set or cred set is empty), the association is not linked\n .otherwise(f.lit(False))\n )\n\n @classmethod\n def clump(cls: type[LDclumping], associations: StudyLocus) -> StudyLocus:\n\"\"\"Perform clumping on studyLocus dataset.\n\n Args:\n associations (StudyLocus): StudyLocus dataset\n\n Returns:\n StudyLocus: including flag and removing credibleSet information for LD clumped loci.\n \"\"\"\n return associations.clump()\n
Calculate bayesian colocalisation based on overlapping signals from credible sets.
Based on the R COLOC package, which uses the Bayes factors from the credible set to estimate the posterior probability of colocalisation. This method makes the simplifying assumption that only one single causal variant exists for any given trait in any genomic region.
Hypothesis Description H0 no association with either trait in the region H1 association with trait 1 only H2 association with trait 2 only H3 both traits are associated, but have different single causal variants H4 both traits are associated and share the same single causal variant
Approximate Bayes factors required
Coloc requires the availability of approximate Bayes factors (ABF) for each variant in the credible set (logABF column).
Source code in src/otg/method/colocalisation.py
class Coloc:\n\"\"\"Calculate bayesian colocalisation based on overlapping signals from credible sets.\n\n Based on the [R COLOC package](https://github.com/chr1swallace/coloc/blob/main/R/claudia.R), which uses the Bayes factors from the credible set to estimate the posterior probability of colocalisation. This method makes the simplifying assumption that **only one single causal variant** exists for any given trait in any genomic region.\n\n | Hypothesis | Description |\n | ------------- | --------------------------------------------------------------------- |\n | H<sub>0</sub> | no association with either trait in the region |\n | H<sub>1</sub> | association with trait 1 only |\n | H<sub>2</sub> | association with trait 2 only |\n | H<sub>3</sub> | both traits are associated, but have different single causal variants |\n | H<sub>4</sub> | both traits are associated and share the same single causal variant |\n\n !!! warning \"Approximate Bayes factors required\"\n Coloc requires the availability of approximate Bayes factors (ABF) for each variant in the credible set (`logABF` column).\n\n \"\"\"\n\n @staticmethod\n def _get_logsum(log_abf: ndarray) -> float:\n\"\"\"Calculates logsum of vector.\n\n This function calculates the log of the sum of the exponentiated\n logs taking out the max, i.e. insuring that the sum is not Inf\n\n Args:\n log_abf (ndarray): log approximate bayes factor\n\n Returns:\n float: logsum\n\n Example:\n >>> l = [0.2, 0.1, 0.05, 0]\n >>> round(Coloc._get_logsum(l), 6)\n 1.476557\n \"\"\"\n themax = np.max(log_abf)\n result = themax + np.log(np.sum(np.exp(log_abf - themax)))\n return float(result)\n\n @staticmethod\n def _get_posteriors(all_abfs: ndarray) -> DenseVector:\n\"\"\"Calculate posterior probabilities for each hypothesis.\n\n Args:\n all_abfs (ndarray): h0-h4 bayes factors\n\n Returns:\n DenseVector: Posterior\n\n Example:\n >>> l = np.array([0.2, 0.1, 0.05, 0])\n >>> Coloc._get_posteriors(l)\n DenseVector([0.279, 0.2524, 0.2401, 0.2284])\n \"\"\"\n diff = all_abfs - Coloc._get_logsum(all_abfs)\n abfs_posteriors = np.exp(diff)\n return Vectors.dense(abfs_posteriors)\n\n @classmethod\n def colocalise(\n cls: type[Coloc],\n overlapping_signals: StudyLocusOverlap,\n priorc1: float = 1e-4,\n priorc2: float = 1e-4,\n priorc12: float = 1e-5,\n ) -> Colocalisation:\n\"\"\"Calculate bayesian colocalisation based on overlapping signals.\n\n Args:\n overlapping_signals (StudyLocusOverlap): overlapping peaks\n priorc1 (float): Prior on variant being causal for trait 1. Defaults to 1e-4.\n priorc2 (float): Prior on variant being causal for trait 2. Defaults to 1e-4.\n priorc12 (float): Prior on variant being causal for traits 1 and 2. Defaults to 1e-5.\n\n Returns:\n Colocalisation: Colocalisation results\n \"\"\"\n # register udfs\n logsum = f.udf(Coloc._get_logsum, DoubleType())\n posteriors = f.udf(Coloc._get_posteriors, VectorUDT())\n return Colocalisation(\n _df=(\n overlapping_signals.df\n # Before summing log_abf columns nulls need to be filled with 0:\n .fillna(0, subset=[\"left_logABF\", \"right_logABF\"])\n # Sum of log_abfs for each pair of signals\n .withColumn(\"sum_log_abf\", f.col(\"left_logABF\") + f.col(\"right_logABF\"))\n # Group by overlapping peak and generating dense vectors of log_abf:\n .groupBy(\"chromosome\", \"left_studyLocusId\", \"right_studyLocusId\")\n .agg(\n f.count(\"*\").alias(\"coloc_n_vars\"),\n fml.array_to_vector(f.collect_list(f.col(\"left_logABF\"))).alias(\n \"left_logABF\"\n ),\n fml.array_to_vector(f.collect_list(f.col(\"right_logABF\"))).alias(\n \"right_logABF\"\n ),\n fml.array_to_vector(f.collect_list(f.col(\"sum_log_abf\"))).alias(\n \"sum_log_abf\"\n ),\n )\n .withColumn(\"logsum1\", logsum(f.col(\"left_logABF\")))\n .withColumn(\"logsum2\", logsum(f.col(\"right_logABF\")))\n .withColumn(\"logsum12\", logsum(f.col(\"sum_log_abf\")))\n .drop(\"left_logABF\", \"right_logABF\", \"sum_log_abf\")\n # Add priors\n # priorc1 Prior on variant being causal for trait 1\n .withColumn(\"priorc1\", f.lit(priorc1))\n # priorc2 Prior on variant being causal for trait 2\n .withColumn(\"priorc2\", f.lit(priorc2))\n # priorc12 Prior on variant being causal for traits 1 and 2\n .withColumn(\"priorc12\", f.lit(priorc12))\n # h0-h2\n .withColumn(\"lH0abf\", f.lit(0))\n .withColumn(\"lH1abf\", f.log(f.col(\"priorc1\")) + f.col(\"logsum1\"))\n .withColumn(\"lH2abf\", f.log(f.col(\"priorc2\")) + f.col(\"logsum2\"))\n # h3\n .withColumn(\"sumlogsum\", f.col(\"logsum1\") + f.col(\"logsum2\"))\n # exclude null H3/H4s: due to sumlogsum == logsum12\n .filter(f.col(\"sumlogsum\") != f.col(\"logsum12\"))\n .withColumn(\"max\", f.greatest(\"sumlogsum\", \"logsum12\"))\n .withColumn(\n \"logdiff\",\n (\n f.col(\"max\")\n + f.log(\n f.exp(f.col(\"sumlogsum\") - f.col(\"max\"))\n - f.exp(f.col(\"logsum12\") - f.col(\"max\"))\n )\n ),\n )\n .withColumn(\n \"lH3abf\",\n f.log(f.col(\"priorc1\"))\n + f.log(f.col(\"priorc2\"))\n + f.col(\"logdiff\"),\n )\n .drop(\"right_logsum\", \"left_logsum\", \"sumlogsum\", \"max\", \"logdiff\")\n # h4\n .withColumn(\"lH4abf\", f.log(f.col(\"priorc12\")) + f.col(\"logsum12\"))\n # cleaning\n .drop(\n \"priorc1\", \"priorc2\", \"priorc12\", \"logsum1\", \"logsum2\", \"logsum12\"\n )\n # posteriors\n .withColumn(\n \"allABF\",\n fml.array_to_vector(\n f.array(\n f.col(\"lH0abf\"),\n f.col(\"lH1abf\"),\n f.col(\"lH2abf\"),\n f.col(\"lH3abf\"),\n f.col(\"lH4abf\"),\n )\n ),\n )\n .withColumn(\n \"posteriors\", fml.vector_to_array(posteriors(f.col(\"allABF\")))\n )\n .withColumn(\"coloc_h0\", f.col(\"posteriors\").getItem(0))\n .withColumn(\"coloc_h1\", f.col(\"posteriors\").getItem(1))\n .withColumn(\"coloc_h2\", f.col(\"posteriors\").getItem(2))\n .withColumn(\"coloc_h3\", f.col(\"posteriors\").getItem(3))\n .withColumn(\"coloc_h4\", f.col(\"posteriors\").getItem(4))\n .withColumn(\"coloc_h4_h3\", f.col(\"coloc_h4\") / f.col(\"coloc_h3\"))\n .withColumn(\"coloc_log2_h4_h3\", f.log2(f.col(\"coloc_h4_h3\")))\n # clean up\n .drop(\n \"posteriors\",\n \"allABF\",\n \"coloc_h4_h3\",\n \"lH0abf\",\n \"lH1abf\",\n \"lH2abf\",\n \"lH3abf\",\n \"lH4abf\",\n )\n .withColumn(\"colocalisationMethod\", f.lit(\"COLOC\"))\n )\n )\n
It extends CAVIAR\u00a0framework to explicitly estimate the posterior probability that the same variant is causal in 2 studies while accounting for the uncertainty of LD. eCAVIAR computes the colocalization posterior probability (CLPP) by utilizing the marginal posterior probabilities. This framework allows for multiple variants to be causal in a single locus.
Source code in src/otg/method/colocalisation.py
class ECaviar:\n\"\"\"ECaviar-based colocalisation analysis.\n\n It extends [CAVIAR](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5142122/#bib18)\u00a0framework to explicitly estimate the posterior probability that the same variant is causal in 2 studies while accounting for the uncertainty of LD. eCAVIAR computes the colocalization posterior probability (**CLPP**) by utilizing the marginal posterior probabilities. This framework allows for **multiple variants to be causal** in a single locus.\n \"\"\"\n\n @staticmethod\n def _get_clpp(left_pp: Column, right_pp: Column) -> Column:\n\"\"\"Calculate the colocalisation posterior probability (CLPP).\n\n If the fact that the same variant is found causal for two studies are independent events,\n CLPP is defined as the product of posterior porbabilities that a variant is causal in both studies.\n\n Args:\n left_pp (Column): left posterior probability\n right_pp (Column): right posterior probability\n\n Returns:\n Column: CLPP\n\n Examples:\n >>> d = [{\"left_pp\": 0.5, \"right_pp\": 0.5}, {\"left_pp\": 0.25, \"right_pp\": 0.75}]\n >>> df = spark.createDataFrame(d)\n >>> df.withColumn(\"clpp\", ECaviar._get_clpp(f.col(\"left_pp\"), f.col(\"right_pp\"))).show()\n +-------+--------+------+\n |left_pp|right_pp| clpp|\n +-------+--------+------+\n | 0.5| 0.5| 0.25|\n | 0.25| 0.75|0.1875|\n +-------+--------+------+\n <BLANKLINE>\n\n \"\"\"\n return left_pp * right_pp\n\n @classmethod\n def colocalise(\n cls: type[ECaviar], overlapping_signals: StudyLocusOverlap\n ) -> Colocalisation:\n\"\"\"Calculate bayesian colocalisation based on overlapping signals.\n\n Args:\n overlapping_signals (StudyLocusOverlap): overlapping signals.\n\n Returns:\n Colocalisation: colocalisation results based on eCAVIAR.\n \"\"\"\n return Colocalisation(\n _df=(\n overlapping_signals.df.withColumn(\n \"clpp\",\n ECaviar._get_clpp(\n f.col(\"left_posteriorProbability\"),\n f.col(\"right_posteriorProbability\"),\n ),\n )\n .groupBy(\"left_studyLocusId\", \"right_studyLocusId\", \"chromosome\")\n .agg(\n f.count(\"*\").alias(\"coloc_n_vars\"),\n f.sum(f.col(\"clpp\")).alias(\"clpp\"),\n )\n .withColumn(\"colocalisationMethod\", f.lit(\"eCAVIAR\"))\n )\n )\n
@classmethod\ndef ld_annotation_by_locus_ancestry(\n cls: type[LDAnnotatorGnomad],\n session: Session,\n associations: StudyLocusGWASCatalog,\n studies: StudyIndexGWASCatalog,\n ld_populations: list[str],\n ld_index_template: str,\n ld_matrix_template: str,\n min_r2: float,\n) -> DataFrame:\n\"\"\"LD information for all locus and ancestries.\n\n Args:\n session (Session): Session\n associations (StudyLocusGWASCatalog): GWAS associations\n studies (StudyIndexGWASCatalog): study metadata of the associations\n ld_populations (list[str]): List of populations to annotate\n ld_index_template (str): Template path of the LD matrix index containing `{POP}` where the population is expected\n ld_matrix_template (str): Template path of the LD matrix containing `{POP}` where the population is expected\n min_r2 (float): minimum r2 to keep\n\n Returns:\n DataFrame: LD annotation [\"variantId\", \"chromosome\", \"gnomadPopulation\", \"tagVariantId\", \"r\"]\n \"\"\"\n # Unique lead - population pairs:\n locus_ancestry = (\n associations.unique_study_locus_ancestries(studies)\n # Ignoring study information / relativeSampleSize to get unique lead-ancestry pairs\n .drop(\"studyId\", \"relativeSampleSize\")\n .distinct()\n .persist()\n )\n\n # All gnomad populations captured in associations:\n assoc_populations = locus_ancestry.rdd.map(\n lambda x: x.gnomadPopulation\n ).collect()\n\n # Retrieve LD information from gnomAD\n ld_annotated_assocs = []\n for population in ld_populations:\n if population in assoc_populations:\n pop_parsed_ldindex_path = ld_index_template.format(POP=population)\n pop_matrix_path = ld_matrix_template.format(POP=population)\n ld_index = LDIndex.from_parquet(session, pop_parsed_ldindex_path)\n ld_matrix = BlockMatrix.read(pop_matrix_path)\n ld_annotated_assocs.append(\n LDAnnotatorGnomad.get_ld_annotated_assocs_for_population(\n population,\n ld_index,\n ld_matrix,\n locus_ancestry,\n min_r2,\n ).coalesce(400)\n )\n return reduce(DataFrame.unionByName, ld_annotated_assocs)\n
Probabilistic Identification of Causal SNPs (PICS), an algorithm estimating the probability that an individual variant is causal considering the haplotype structure and observed pattern of association at the genetic locus.
Source code in src/otg/method/pics.py
class PICS:\n\"\"\"Probabilistic Identification of Causal SNPs (PICS), an algorithm estimating the probability that an individual variant is causal considering the haplotype structure and observed pattern of association at the genetic locus.\"\"\"\n\n @staticmethod\n def _pics_relative_posterior_probability(\n neglog_p: float, pics_snp_mu: float, pics_snp_std: float\n ) -> float:\n\"\"\"Compute the PICS posterior probability for a given SNP.\n\n !!! info \"This probability needs to be scaled to take into account the probabilities of the other variants in the locus.\"\n\n Args:\n neglog_p (float): Negative log p-value of the lead variant\n pics_snp_mu (float): Mean P value of the association between a SNP and a trait\n pics_snp_std (float): Standard deviation for the P value of the association between a SNP and a trait\n\n Returns:\n Relative posterior probability of a SNP being causal in a locus\n\n Examples:\n >>> rel_prob = PICS._pics_relative_posterior_probability(neglog_p=10.0, pics_snp_mu=1.0, pics_snp_std=10.0)\n >>> round(rel_prob, 3)\n 0.368\n \"\"\"\n return float(norm(pics_snp_mu, pics_snp_std).sf(neglog_p) * 2)\n\n @staticmethod\n def _pics_standard_deviation(neglog_p: float, r2: float, k: float) -> float | None:\n\"\"\"Compute the PICS standard deviation.\n\n This distribution is obtained after a series of permutation tests described in the PICS method, and it is only\n valid when the SNP is highly linked with the lead (r2 > 0.5).\n\n Args:\n neglog_p (float): Negative log p-value of the lead variant\n r2 (float): LD score between a given SNP and the lead variant\n k (float): Empiric constant that can be adjusted to fit the curve, 6.4 recommended.\n\n Returns:\n Standard deviation for the P value of the association between a SNP and a trait\n\n Examples:\n >>> PICS._pics_standard_deviation(neglog_p=1.0, r2=1.0, k=6.4)\n 0.0\n >>> round(PICS._pics_standard_deviation(neglog_p=10.0, r2=0.5, k=6.4), 3)\n 0.143\n >>> print(PICS._pics_standard_deviation(neglog_p=1.0, r2=0.0, k=6.4))\n None\n \"\"\"\n return (\n (1 - abs(r2) ** 0.5**k) ** 0.5 * (neglog_p) ** 0.5 / 2\n if r2 >= 0.5\n else None\n )\n\n @staticmethod\n def _pics_mu(neglog_p: float, r2: float) -> float | None:\n\"\"\"Compute the PICS mu that estimates the probability of association between a given SNP and the trait.\n\n This distribution is obtained after a series of permutation tests described in the PICS method, and it is only\n valid when the SNP is highly linked with the lead (r2 > 0.5).\n\n Args:\n neglog_p (float): Negative log p-value of the lead variant\n r2 (float): LD score between a given SNP and the lead variant\n\n Returns:\n Mean P value of the association between a SNP and a trait\n\n Examples:\n >>> PICS._pics_mu(neglog_p=1.0, r2=1.0)\n 1.0\n >>> PICS._pics_mu(neglog_p=10.0, r2=0.5)\n 5.0\n >>> print(PICS._pics_mu(neglog_p=10.0, r2=0.3))\n None\n \"\"\"\n return neglog_p * r2 if r2 >= 0.5 else None\n\n @staticmethod\n def _finemap(\n credible_set: list[Row], lead_neglog_p: float, k: float\n ) -> list | None:\n\"\"\"Calculates the probability of a variant being causal in a study-locus context by applying the PICS method.\n\n It is intended to be applied as an UDF in `PICS.finemap`, where each row is a StudyLocus association.\n The function iterates over every SNP in the `credibleSet` array, and it returns an updated credibleSet with\n its association signal and causality probability as of PICS.\n\n Args:\n credible_set (list): list of tagging variants after expanding the locus\n lead_neglog_p (float): P value of the association signal between the lead variant and the study in the form of -log10.\n k (float): Empiric constant that can be adjusted to fit the curve, 6.4 recommended.\n\n Returns:\n List of tagging variants with an estimation of the association signal and their posterior probability as of PICS.\n \"\"\"\n if credible_set is None:\n return None\n elif not credible_set:\n return []\n\n tmp_credible_set = []\n new_credible_set = []\n # First iteration: calculation of mu, standard deviation, and the relative posterior probability\n for tag_struct in credible_set:\n tag_dict = (\n tag_struct.asDict()\n ) # tag_struct is of type pyspark.Row, we'll represent it as a dict\n if (\n not tag_dict[\"r2Overall\"]\n or tag_dict[\"r2Overall\"] < 0.5\n or not lead_neglog_p\n ):\n # If PICS cannot be calculated, we'll return the original credible set\n new_credible_set.append(tag_dict)\n continue\n pics_snp_mu = PICS._pics_mu(lead_neglog_p, tag_dict[\"r2Overall\"])\n pics_snp_std = PICS._pics_standard_deviation(\n lead_neglog_p, tag_dict[\"r2Overall\"], k\n )\n pics_snp_std = 0.001 if pics_snp_std == 0 else pics_snp_std\n if pics_snp_mu is not None and pics_snp_std is not None:\n posterior_probability = PICS._pics_relative_posterior_probability(\n lead_neglog_p, pics_snp_mu, pics_snp_std\n )\n tag_dict[\"tagPValue\"] = 10**-pics_snp_mu\n tag_dict[\"tagStandardError\"] = 10**-pics_snp_std\n tag_dict[\"relativePosteriorProbability\"] = posterior_probability\n\n tmp_credible_set.append(tag_dict)\n\n # Second iteration: calculation of the sum of all the posteriors in each study-locus, so that we scale them between 0-1\n total_posteriors = sum(\n tag_dict.get(\"relativePosteriorProbability\", 0)\n for tag_dict in tmp_credible_set\n )\n\n # Third iteration: calculation of the final posteriorProbability\n for tag_dict in tmp_credible_set:\n if total_posteriors != 0:\n tag_dict[\"posteriorProbability\"] = float(\n tag_dict.get(\"relativePosteriorProbability\", 0) / total_posteriors\n )\n tag_dict.pop(\"relativePosteriorProbability\")\n new_credible_set.append(tag_dict)\n return new_credible_set\n\n @classmethod\n def finemap(\n cls: type[PICS], associations: StudyLocus, k: float = 6.4\n ) -> StudyLocus:\n\"\"\"Run PICS on a study locus.\n\n !!! info \"Study locus needs to be LD annotated\"\n The study locus needs to be LD annotated before PICS can be calculated.\n\n Args:\n associations (StudyLocus): Study locus to finemap using PICS\n k (float): Empiric constant that can be adjusted to fit the curve, 6.4 recommended.\n\n Returns:\n StudyLocus: Study locus with PICS results\n \"\"\"\n # Register UDF by defining the structure of the output credibleSet array of structs\n credset_schema = t.ArrayType(\n [field.dataType.elementType for field in associations.schema if field.name == \"credibleSet\"][0] # type: ignore\n )\n _finemap_udf = f.udf(\n lambda credible_set, neglog_p: PICS._finemap(credible_set, neglog_p, k),\n credset_schema,\n )\n\n associations.df = (\n associations.df.withColumn(\"neglog_pvalue\", associations.neglog_pvalue())\n .withColumn(\n \"credibleSet\",\n f.when(\n f.col(\"credibleSet\").isNotNull(),\n _finemap_udf(f.col(\"credibleSet\"), f.col(\"neglog_pvalue\")),\n ),\n )\n .drop(\"neglog_pvalue\")\n )\n return associations\n
The study locus needs to be LD annotated before PICS can be calculated.
Parameters:
Name Type Description Default associationsStudyLocus
Study locus to finemap using PICS
required kfloat
Empiric constant that can be adjusted to fit the curve, 6.4 recommended.
6.4
Returns:
Name Type Description StudyLocusStudyLocus
Study locus with PICS results
Source code in src/otg/method/pics.py
@classmethod\ndef finemap(\n cls: type[PICS], associations: StudyLocus, k: float = 6.4\n) -> StudyLocus:\n\"\"\"Run PICS on a study locus.\n\n !!! info \"Study locus needs to be LD annotated\"\n The study locus needs to be LD annotated before PICS can be calculated.\n\n Args:\n associations (StudyLocus): Study locus to finemap using PICS\n k (float): Empiric constant that can be adjusted to fit the curve, 6.4 recommended.\n\n Returns:\n StudyLocus: Study locus with PICS results\n \"\"\"\n # Register UDF by defining the structure of the output credibleSet array of structs\n credset_schema = t.ArrayType(\n [field.dataType.elementType for field in associations.schema if field.name == \"credibleSet\"][0] # type: ignore\n )\n _finemap_udf = f.udf(\n lambda credible_set, neglog_p: PICS._finemap(credible_set, neglog_p, k),\n credset_schema,\n )\n\n associations.df = (\n associations.df.withColumn(\"neglog_pvalue\", associations.neglog_pvalue())\n .withColumn(\n \"credibleSet\",\n f.when(\n f.col(\"credibleSet\").isNotNull(),\n _finemap_udf(f.col(\"credibleSet\"), f.col(\"neglog_pvalue\")),\n ),\n )\n .drop(\"neglog_pvalue\")\n )\n return associations\n
This workflow runs colocalization analyses that assess the degree to which independent signals of the association share the same causal variant in a region of the genome, typically limited by linkage disequilibrium (LD).
Source code in src/otg/colocalisation.py
@dataclass\nclass ColocalisationStep(ColocalisationStepConfig):\n\"\"\"Colocalisation step.\n\n This workflow runs colocalization analyses that assess the degree to which independent signals of the association share the same causal variant in a region of the genome, typically limited by linkage disequilibrium (LD).\n \"\"\"\n\n session: Session = Session()\n\n def run(self: ColocalisationStep) -> None:\n\"\"\"Run colocalisation step.\"\"\"\n # Study-locus information\n sl = StudyLocus.from_parquet(self.session, self.study_locus_path)\n si = StudyIndex.from_parquet(self.session, self.study_index_path)\n\n # Study-locus overlaps for 95% credible sets\n sl_overlaps = sl.credible_set(CredibleInterval.IS95).overlaps(si)\n\n coloc_results = Coloc.colocalise(\n sl_overlaps, self.priorc1, self.priorc2, self.priorc12\n )\n ecaviar_results = ECaviar.colocalise(sl_overlaps)\n\n coloc_results.df.unionByName(ecaviar_results.df, allowMissingColumns=True)\n\n coloc_results.df.write.mode(self.session.write_mode).parquet(self.coloc_path)\n
Colocalisation step requirements.
Attributes:
Name Type Description study_locus_pathDictConfig
Input Study-locus path.
coloc_pathDictConfig
Output Colocalisation path.
priorc1float
Prior on variant being causal for trait 1.
priorc2float
Prior on variant being causal for trait 2.
priorc12float
Prior on variant being causal for traits 1 and 2.
Source code in src/otg/config.py
@dataclass\nclass ColocalisationStepConfig:\n\"\"\"Colocalisation step requirements.\n\n Attributes:\n study_locus_path (DictConfig): Input Study-locus path.\n coloc_path (DictConfig): Output Colocalisation path.\n priorc1 (float): Prior on variant being causal for trait 1.\n priorc2 (float): Prior on variant being causal for trait 2.\n priorc12 (float): Prior on variant being causal for traits 1 and 2.\n \"\"\"\n\n _target_: str = \"otg.colocalisation.ColocalisationStep\"\n study_locus_path: str = MISSING\n study_index_path: str = MISSING\n coloc_path: str = MISSING\n priorc1: float = 1e-4\n priorc2: float = 1e-4\n priorc12: float = 1e-5\n
Variant annotation step produces a dataset of the type VariantAnnotation derived from gnomADs gnomad.genomes.vX.X.X.sites.ht Hail's table. This dataset is used to validate variants and as a source of annotation.
Source code in src/otg/variant_annotation.py
@dataclass\nclass VariantAnnotationStep(VariantAnnotationStepConfig):\n\"\"\"Variant annotation step.\n\n Variant annotation step produces a dataset of the type `VariantAnnotation` derived from gnomADs `gnomad.genomes.vX.X.X.sites.ht` Hail's table. This dataset is used to validate variants and as a source of annotation.\n \"\"\"\n\n session: Session = Session()\n\n def run(self: VariantAnnotationStep) -> None:\n\"\"\"Run variant annotation step.\"\"\"\n # init hail session\n hl.init(sc=self.session.spark.sparkContext, log=\"/dev/null\")\n\n\"\"\"Run variant annotation step.\"\"\"\n variant_annotation = VariantAnnotation.from_gnomad(\n self.gnomad_genomes,\n self.chain_38_to_37,\n self.populations,\n )\n # Writing data partitioned by chromosome and position:\n (\n variant_annotation.df.repartition(400, \"chromosome\")\n .sortWithinPartitions(\"chromosome\", \"position\")\n .write.partitionBy(\"chromosome\")\n .mode(self.session.write_mode)\n .parquet(self.variant_annotation_path)\n )\n
Using a VariantAnnotation dataset as a reference, this step creates and writes a dataset of the type VariantIndex that includes only variants that have disease-association data with a reduced set of annotations.
Source code in src/otg/variant_index.py
@dataclass\nclass VariantIndexStep(VariantIndexStepConfig):\n\"\"\"Variant index step.\n\n Using a `VariantAnnotation` dataset as a reference, this step creates and writes a dataset of the type `VariantIndex` that includes only variants that have disease-association data with a reduced set of annotations.\n \"\"\"\n\n session: Session = Session()\n\n def run(self: VariantIndexStep) -> None:\n\"\"\"Run variant index step.\"\"\"\n # Variant annotation dataset\n va = VariantAnnotation.from_parquet(self.session, self.variant_annotation_path)\n\n # Study-locus dataset\n study_locus = StudyLocus.from_parquet(self.session, self.study_locus_path)\n\n # Reduce scope of variant annotation dataset to only variants in study-locus sets:\n va_slimmed = va.filter_by_variant_df(\n study_locus.unique_lead_tag_variants(), [\"id\", \"chromosome\"]\n )\n\n # Generate variant index ussing a subset of the variant annotation dataset\n vi = VariantIndex.from_variant_annotation(va_slimmed)\n\n # Write data:\n # self.etl.logger.info(\n # f\"Writing invalid variants from the credible set to: {self.variant_invalid}\"\n # )\n # vi.invalid_variants.write.mode(self.etl.write_mode).parquet(\n # self.variant_invalid\n # )\n\n self.session.logger.info(f\"Writing variant index to: {self.variant_index_path}\")\n (\n vi.df.write.partitionBy(\"chromosome\")\n .mode(self.session.write_mode)\n .parquet(self.variant_index_path)\n )\n
This step aims to generate a dataset that contains multiple pieces of evidence supporting the functional association of specific variants with genes. Some of the evidence types include:
Chromatin interaction experiments, e.g. Promoter Capture Hi-C (PCHi-C).
In silico functional predictions, e.g. Variant Effect Predictor (VEP) from Ensembl.
Distance between the variant and each gene's canonical transcription start site (TSS).
Source code in src/otg/v2g.py
@dataclass\nclass V2GStep(V2GStepConfig):\n\"\"\"Variant-to-gene (V2G) step.\n\n This step aims to generate a dataset that contains multiple pieces of evidence supporting the functional association of specific variants with genes. Some of the evidence types include:\n\n 1. Chromatin interaction experiments, e.g. Promoter Capture Hi-C (PCHi-C).\n 2. In silico functional predictions, e.g. Variant Effect Predictor (VEP) from Ensembl.\n 3. Distance between the variant and each gene's canonical transcription start site (TSS).\n\n \"\"\"\n\n session: Session = Session()\n\n def run(self: V2GStep) -> None:\n\"\"\"Run V2G dataset generation.\"\"\"\n # Filter gene index by approved biotypes to define V2G gene universe\n gene_index_filtered = GeneIndex.from_parquet(\n self.session, self.gene_index_path\n ).filter_by_biotypes(self.approved_biotypes)\n\n vi = VariantIndex.from_parquet(self.session, self.variant_index_path).persist()\n va = VariantAnnotation.from_parquet(self.session, self.variant_annotation_path)\n vep_consequences = self.session.spark.read.csv(\n self.vep_consequences_path, sep=\"\\t\", header=True\n )\n\n # Variant annotation reduced to the variant index to define V2G variant universe\n va_slimmed = va.filter_by_variant_df(vi.df, [\"id\", \"chromosome\"]).persist()\n\n # lift over variants to hg38\n lift = LiftOverSpark(\n self.liftover_chain_file_path, self.liftover_max_length_difference\n )\n\n v2g_datasets = [\n va_slimmed.get_distance_to_tss(gene_index_filtered, self.max_distance),\n # variant effects\n va_slimmed.get_most_severe_vep_v2g(vep_consequences, gene_index_filtered),\n va_slimmed.get_polyphen_v2g(gene_index_filtered),\n va_slimmed.get_sift_v2g(gene_index_filtered),\n va_slimmed.get_plof_v2g(gene_index_filtered),\n # intervals\n Intervals.parse_andersson(\n self.session, self.anderson_path, gene_index_filtered, lift\n ).v2g(vi),\n Intervals.parse_javierre(\n self.session, self.javierre_path, gene_index_filtered, lift\n ).v2g(vi),\n Intervals.parse_jung(\n self.session, self.jung_path, gene_index_filtered, lift\n ).v2g(vi),\n Intervals.parse_thurman(\n self.session, self.thurnman_path, gene_index_filtered, lift\n ).v2g(vi),\n ]\n\n # merge all V2G datasets\n v2g = V2G(\n _df=reduce(\n lambda x, y: x.unionByName(y, allowMissingColumns=True),\n [dataset.df for dataset in v2g_datasets],\n ).repartition(\"chromosome\")\n )\n # write V2G dataset\n (\n v2g.df.write.partitionBy(\"chromosome\")\n .mode(self.session.write_mode)\n .parquet(self.v2g_path)\n )\n
"}]}
\ No newline at end of file
diff --git a/sitemap.xml.gz b/sitemap.xml.gz
index 1057ad0e2..e9db2912a 100644
Binary files a/sitemap.xml.gz and b/sitemap.xml.gz differ