variational_lsrc

data:

Input data brain_snp_covars_meancentered_scaled.h5 is an hdf5 file, see the file save_h5.py for more information. Cross validation splits were pre-made and saved to a csv file, see save_cv_splits.py for more information.

computing

• this is intended for HCP use, if you only need to run this a single time, you can bypass all of the singularity, and nipype and just use the R files(you will have to change them accordingly).

docker url:

https://hub.docker.com/r/ysa6/genus/

command to build using singularity:

singularity build genus_img.sqsh docker://ysa6/genus:latest

command to run the container for an interactive session

singularity shell --bind /storage:/storage genus_img.sqsh

General logic of execution:

1. gtoi_cv.py
   • a) write_varbvs function creates directory structure and writes a bash file that calls the gtoi_cv_template.r file when inside the singularity image
   • b) run_sing function writes batch files for slurm that execute a singularity image with all dependencies and environmental variables to run the analysis.

• example command to run the first part of the analysis

python gtoi_cv.py -dp /storage/gablab001/data/genus/GIT/genus/bayes/data_sets/files_for_edward/brain_snp_covars_meancentered_scaled.h5 \
-sn /storage/gablab001/data/genus/GIT/genus/bayes/data_sets/files_for_edward/dev_for_container/bf_out_cv \
-nr 170 \
-ip /storage/gablab001/data/genus/GIT/genus/genus_img.sqsh \
-bp /storage:/storage \
-cn I_cols \
-cv /storage/gablab001/data/genus/GIT/genus/bayes/data_sets/files_for_edward/dev_for_container/shuffle_split_cv.csv \
-gd /storage/gablab001/data/genus/GIT/genus/bayes/data_sets/files_for_edward/dev_for_container/gtoi_res \

that first part should have created a directory where the bayes factor scores are outputed, using the path to that directory you run the second part with this example command, using the same logic execution as above:

2. itod_cv.py

python itod_cv.py -dp /storage/gablab001/data/genus/GIT/genus/bayes/data_sets/files_for_edward/brain_snp_covars_meancentered_scaled.h5 \
-sn /storage/gablab001/data/genus/GIT/genus/bayes/data_sets/files_for_edward/dev_for_container/fxvb_out_cv \
-ip /storage/gablab001/data/genus/GIT/genus/genus_img.sqsh \
-bp /storage:/storage \
-bf /storage/gablab001/data/genus/GIT/genus/bayes/data_sets/files_for_edward/dev_for_container/bf_out_cv \
-nc 100 \
-cp /storage/gablab001/data/genus/GIT/genus/bayes/data_sets/files_for_edward/dev_for_container/shuffle_split_cv.csv

• take a look at itody_cv.py and gtoi_cv.py to see what the arguments e.g. -bf mean.

Please cite:

@article{batmanghelich2016probabilistic,
  title={Probabilistic modeling of imaging, genetics and diagnosis},
  author={Batmanghelich, Nematollah K and Dalca, Adrian and Quon, Gerald and Sabuncu, Mert and Golland, Polina},
  journal={IEEE transactions on medical imaging},
  volume={35},
  number={7},
  pages={1765--1779},
  year={2016},
  publisher={IEEE}
}


@article{carbonetto2012scalable,
  title={Scalable variational inference for Bayesian variable selection in regression, and its accuracy in genetic association studies},
  author={Carbonetto, Peter and Stephens, Matthew and others},
  journal={Bayesian analysis},
  volume={7},
  number={1},
  pages={73--108},
  year={2012},
  publisher={International Society for Bayesian Analysis}
}