PRS (polygenic risk score)

This repository is to run PRS models across various phenotypes in the UK Biobank.

Where can you find the data?

Where is the data directory?

We store the output from this repo into two places:

• private_output
  • /oak/stanford/groups/mrivas/projects/PRS/private_output/
  • This contains individual-level data, such as the PRS score computed for each individual.
  • I would recommend to make a symbolic link so that you can access this directory with <repository_root>/private_output
• public_output
  • https://github.com/rivas-lab/PRS/tree/master/public_output
  • This contains non-private outputs from the repo, such as evaluation metric (R/AUC).

Directory structure

• We have multiple PRS models supported in this repo. This includes:
  • vanilla_PRS: The canonical PRS with p-value cut-off and LD-clumping
  • snpnet_PRS: Junyang's snpnet (https://github.com/junyangq/snpnet)
• In general we structure the repository so that the results (and pipeline) for each model is kept separate:
  • <repository_root>/[src | public_output | private_output | jobs]/<PRS_model>/
  • The only exception is in helper directory because we share the same scripts across multiple models (such as the one for evaluation)
• src
  • This directory contains pipeline (bash script) that can run particular PRS model.
  • If you run $bash src/script.sh, it should dump usage.
• helpr
• jobs
• public_output

example

• The score file is stored here:
  • $OAK/projects/PRS/private_output/snpnet_PRS/<dataset_name>/4_score/<GBE_ID>.sscore
  • For example, IOP score is in: $OAK/projects/PRS/private_output/snpnet_PRS/test/4_score/INI5255.sscore

Getting started

0. Install packages

$ ml load snpnet # the module (/home/groups/mrivas/.modules/snpnet.lua) loads dependencies
$ R
> # If you don't have devtools on R, please install it.
> # install.packages("devtools") 
> install.packages("BGData", dependencies=TRUE)
> library(devtools)
> install_github("junyangq/snpnet")
> # snpnet depends on glmnetPlus package. 
> # This package is in Junyang's private repo.
> # You may use a copy on $OAK
> install_local('/oak/stanford/groups/mrivas/software/glmnetPlus')
> # Plase make sure you can load the new packages
> library(glmnetPlus)
> library(snpnet)

You may want to install libssh and/or libssh2 from Anaconda.

• https://anaconda.org/conda-forge/libssh
• https://anaconda.org/conda-forge/libssh2

1. Clone the repo and create a symbolic link to private_output

$ cd $OAK/users/$USER/repos/rivas-lab
$ git clone git@github.com:rivas-lab/PRS.git
$ # There is a dependency to some scripts in ukbb-tools
$ git clone git@github.com:rivas-lab/ukbb-tools.git
$ cd PRS
$ ln -s $OAK/projects/PRS/private_output .

2. load the relevant modules

$ ml load snpnet

3. run snpnet_PRS or vanilla_PRS

$ cd $OAK/users/$USER/repos/rivas-lab/PRS/src
$ PRS_model="snpnet_PRS"
$ dataset_name="test_$USER"
$ memory=120000 # 120 GB
$ threads=10 # 10 CPUs
$ app_id=24983
$ n_PCs=10
$ bash snpnet_PRS.sh [ GBE_ID | phe_file.phe ] [ qt | bin ] keep.phe $OAK/projects/PRS/private_output/${PRS_model}/${dataset_name} ${memory} ${threads} ${app_id} ${nPCs}

You sould be able to run $OAK/users/$USER/repos/rivas-lab/PRS/src/vanilla_PRS.sh in a very similar way.