Pfam × AlphaFold

This project leverages UniProt, Pfam, and AlphaFold data to provide a comprehensive overview of Pfam matches against UniProt proteins and utilizes AlphaFold's pLDDT scores to calculate an average domain pLDDT score for each Pfam family.

Getting started

Install the package:

$ pip install git+https://github.com/matthiasblum/pfam-alphafold.git

Preparing data

The following sections explain how to obtain and pre-process data in order to build the Pfam×AlphaFold database. Steps are independent and can be run in parallel.

AlphaFold

Prerequisite: directory of predictions grouped in one tar archive per proteome. See Bulk download from the AlphaFold Protein Structure Database GitHub repository to learn how to download these archives.

Extract and index pLDDT scores:

$ pfafindex alphafold [-p N] INDIR OUTPUT

Arguments:

• -p N: Use up to N processors (default: 1).
• INDIR: Directory of individual tar archives.
• OUTPUT: Output file of indexed AlphaFold pLDDT scores.

With -p 16, it took about 50 hours and 23 GB of memory to complete. The output file is about 600 GB big.

UniProtKB

Prerequisites:

• UniProtKB/Swiss-Prot flat file (uniprot_sprot.dat.gz)
• UniProtKB/TrEMBL flat file (uniprot_trembl.dat.gz)

You can obtain the most recent version of these files from UniProt's FTP site:

$ wget https://ftp.ebi.ac.uk/pub/databases/uniprot/current_release/knowledgebase/complete/uniprot_sprot.dat.gz
$ wget https://ftp.ebi.ac.uk/pub/databases/uniprot/current_release/knowledgebase/complete/uniprot_trembl.dat.gz

Index UniProtKB entries:

$ pfafindex uniprot uniprot_sprot.dat.gz uniprot_trembl.dat.gz OUTPUT

Arguments:

• uniprot_sprot.dat.gz: UniProtKB/Swiss-Prot flat file.
• uniprot_trembl.dat.gz: UniProtKB/TrEMBL flat file.
• OUTPUT: Output file of indexed UniProtKB entries.

Runtime: ~14 hours; memory: ~2 GB; output file: ~60 GB.

Pfam

Prerequisites:

• Pfam summary flat file (Pfam-A.hmm.dat.gz)
• InterPro matches XML file (match_complete.xml.gz)

You can obtain the most recent version of these files from the InterPro's and Pfam's FTP sites:

$ wget https://ftp.ebi.ac.uk/pub/databases/Pfam/current_release/Pfam-A.hmm.dat.gz
$ wget https://ftp.ebi.ac.uk/pub/databases/interpro/current_release/match_complete.xml.gz

Index Pfam entries:

$ pfafindex pfam Pfam-A.hmm.dat.gz match_complete.xml.gz OUTPUT

Arguments:

• Pfam-A.hmm.dat.gz: Pfam summary flat file.
• match_complete.xml.gz: InterPro matches XML file.
• OUTPUT: Output file of indexed UniProtKB entries with Pfam matches.

Runtime: ~16 hours; memory: ~2 GB; output file: ~50 GB.

Building the database

$ pfafbuild alphafold.dat pfam.dat uniprot.dat pfam-alphafold.db

Arguments:

• alphafold.dat: File of indexed AlphaFold pLDDT scores.
• pfam.dat: File of indexed UniProtKB entries with Pfam matches.
• uniprot.dat: File of indexed UniProtKB entries.
• pfam-alphafold.db: Output SQLite3 database.

Runtime: ~11 hours; memory: ~2 GB; output file: ~50 GB.

Web application

Set the FLASK_DATABASE environment variable to the path of the SQLite3 database built with pfafbuild, then start the web server:

$ export FLASK_DATABASE="/path/to/pfam-alphafold.db"
$ gunicorn -b 0.0.0.0:8000 -w 4 pfam_alphafold.web:app

Now head over http://localhost:8000/ to access the web application.

Demo database

This project includes data for a demo database (~800k proteins from key species and model organisms).

Re-create the database with the following command:

$ gunzip -c data/demo.sql.gz | sqlite3 data/demo.db

Docker image

Alternatively, the web application with the demo database can be run using Docker.

To build the image, run:

$ docker build -t pfam-alphafold .

Once the image is built, you can run the container:

$ docker run --rm -p 8000:8000 pfam-alphafold

Now, the web application should be accessible at http://localhost:8000/.