Snakemake workflow: prs_pipeline

This pipeline performs the computation of Polygenic Risck Score using different methods. It integrates different algorithm and allows an easy comparison between the performance of each PRS.

Installation

An installation of snakemake is needed to run this workflow. We suggest to use a conda/mamba environments to handle the installation, in breif, once conda or mamba is installed:

mamba create -n snakemake -c conda-forge -c bioconda 'snakemake>=8'

Activate the conda environment and clone the repository. All the installation and requirements will be handled by snakemake.

mamba activate snakemake
git clone https://github.com/EuracBiomedicalResearch/prs_pipeline.git

Requirements

In order to compute a Polygenic Risk Score the following files are needed:

1. Genotype files in plink (ped) format separated by chromosome. For file specification please have a look (here)[https://www.cog-genomics.org/plink/1.9/formats#bed]
2. GWAS summary statistic. The results of a GWAS analysis for a specific phenotype
3. Linkage Disequilibrium panel (optional - only for specific methods). The choice of the panel depends on the GWAS analysis. The populations included in the LD panel should be genetically as close as possible to the population used in the GWAS analysis.

Dataset details can be defined through the configuration file in YAML format and json files.

Additional information on the configuration can be found in the config.

Genotype dataset

Genotype dataset can be described through a json file, see the example config/genotype.json.

The json file should contain the following entries:

• build: containing the genome build the samples are genotyped. It could on of hg19 or hg38 (hg18 not supported) as string character.
• nchrom: number of chromosomes the genotype files are divided.
• plinkfiles: it will be a python dictionary with keys starting from 1 to nchrom.

GWAS manifest file

GWAS analysis can be done using a variety of methods, and can be obtained from custom analysis or from databases such as GWAS catalog. To support different file specification we use a manifest file in json format. Multiple gwases can be included in the same manifest file, then the PRS will be computed on each GWAS provided. You can find and example of this file in the config directory: config/gwas.json and in the schema file here

For each trait, you have to specify the trait type if it is a quantitative or binary trait, and the columns specification as a python dictionary mapping.

Available algorithms:

• PRScs, see https://github.com/getian107/PRScs
• LDPred2 (through R implementation in bigsnpr package)
• lassosum (implemented in bigsnpr package)
• CT method (implemented in bigsnpr package)
• sBayesR (implemented in gctb) - under development

Running

Once edited the config.yaml file you can run the pipeline with the following command. Needed softwares, packages and files are handled by snakemake.

cd prs_pipeline
snakemake --cores all --use-conda

Contacts

• Michele Filosi