Tool to calculate PRS from PGSCatalog using your genetic data

TL;DR

ssh to hulk, download PGSCatalog's PRS weight matrix and run:

PYTHON="/home/mvn373/fast/tools/python36/bin/python3"
SCRIPT="/home/mvn373/fast/prs-workshop/apply-pgscatalog-prs-stable.py"
PLINK_PREFIX="path to plink data, with common prefix, see example below"
PGSCATALOG_FILE="file downloaded from pgscatalog, may be .txt or gzipped"
OUT_PATH="path and prefix for the output files, see example below"
$PYTHON -u $SCRIPT --genetic $PLINK_PREFIX --prs-wm $PGSCATALOG_FILE --out $OUT_PATH

Detailed description

NB. This version is meant to be run on hulk server of CBMR. More general-use tool is to come.

Results

It creates a file with the PRS values for the individuals. To do so, it aligns the files and runs the plink in the proper way.

Side-effects

It will:

• unpack the file from PGSCatalog, if it was gzipped.
• create an output .prs file,
• produce an intermediate file next to the initial PGSCatalog file,

Input:

• PGSCatalog file
  File must provide beta or log(OR) (but not just OR).
• plink data

How to start using it:

Right now it works interactively (I suggest you run it on hulk. Use your terminal full-screen, and don't resize it (I will make it dynamically adopt to the screen size later)).

1. Log on the hulk
2. Download the file from PGSCatalog and locate your plink files
3. Run the code and follow the prompts

How to run the tool

Here is a general command to run the code:

PYTHON="/home/mvn373/fast/tools/python36/bin/python3"
SCRIPT="/home/mvn373/fast/prs-workshop/apply-pgscatalog-prs-stable.py"
PLINK_PREFIX="path to plink data, with common prefix, see example below"
PGSCATALOG_FILE="file downloaded from pgscatalog, may be .txt or gzipped"
OUT_PATH="path and prefix for the output files, see example below"
$PYTHON -u $SCRIPT --genetic $PLINK_PREFIX --prs-wm $PGSCATALOG_FILE --out $OUT_PATH

It is using my specific version of python, so use the python and the script paths provided, and supply the rest three of the paths. As a matter of nice style, please use the absolute path for the paths - this will make everything run nicer.

You need to supply three arguments:

• --genetic - prefix for your plink data. For example, for a plink trio of files like "/folder/file.bed", "/folder/file.bim", and "/folder/file.fam" it would be "--genetic /folder/file".
• --prs-wm - file from PGSCatalog.
• --out - prefix for the output. For example, if you want an output file "/outfolder/out.prs" it would be "--out /outfolder/out".

Running

Standard mode

Once you start, it will prompt you at the bottom of the screen to provide the information. You will need to: supply the names of the columns for beta and effect allele in the PGSCatalog data choose if your plink data are using rsid or chrom:pos as identifiers (more options and better alignment will be done in the future) supply the names of the columns for rsid OR chromosome+position in the PGSCatalog data make the final confirmation

Automated mode

Experienced users might want to run the tool from their code.
To make it possible the tool also accepts --config argument, which must point to a YAML file, defining the following scalar variables:

# first, define beta and effect allele columns:
beta column: NAME OF THE BETA COLUMN
effect allele column: NAME OF THE EFFECT ALLELE COLUMN
# next, choose how to match your files -
# using rsid [rsid] or chromosome+position [pos]:
matching on: rsid OR pos
# finally, define relevant columns for the chosen matching approach
# - for "rsid" define rsID column:
rsID column: NAME OF THE RSID COLUMN
# - for "pos" define chromosome and position columns:
chromosome column: NAME OF THE CHROMOSOME COLUMN
position column: NAME OF THE POSITION COLUMN

Other arguments (--genetic, --prs-wm, and --out) must still be supplied in the command line.
Be aware that tool won't check the correctness of the config and will just fail if a required argument is undefined.

Example of a valid config for http://www.pgscatalog.org/score/PGS000255/:

beta column: effect_weight
effect allele column: effect_allele
matching on: rsid
rsID column: rsID

Conscious limitation

If your plink data does not have at least 50% of the variants from the PGSCatalog file, the tool won't allow you to run the PRS calculation, as the resulting scores might be too unreliable.