PRS Admixture Simulation

A repository for exploring PRS weighting and snp selection strategies for generalizing polygenic risk scores in admixed individuals.

These functions simulate European, African, and Admixed individuals using msprime and rfmix's simulate function. Then a true polygenic risk score is constructed as described by Martin et. al. 2017, AJHG. GWAS estimated polygenic risk scores are constructed by P+T with SNPs selected based on summary statistics from a European or African population of a Fixed-Effects meta of both. SNPs are weighted using the same stragegies or by using weights specific to the local ancestry at a given PRS locus.

See our recent publication for a detailed application and description of methods:

Cavazos and Witte, Inclusion of Variants Discovered from Diverse Populations Improves Polygenic Risk Score Transferability, HGG Advances 2020

Getting Started

To run the functions described above follow the below instructions:

Package installation

# Main package installation
git clone https://github.com/taylorcavazos/PRS_Admixture_Simulation.git
cd PRS_Admixture_Simulation

# Rfmix simulate installation and set-up
git clone https://github.com/slowkoni/rfmix.git
cd rfmix
autoreconf --force --install
./configure
make
cd ..
mv rfmix/simulate simulation/simulate-admixed
rm -rf rfmix

# Set up virtual environment (recommended: conda) 
conda create --name prs_sim python=3.7
conda activate prs_sim
conda install -c conda-forge --yes --file requirements.txt
conda install -c conda-forge r=3.6.0
conda deactivate

Example run

A possible simulation run is shown below:

conda activate prs_sim
python run_simulation.py --sim 1 --snp_selection ceu --snp_weighting ceu
conda deactivate

Simulation parameters

All parameters from the simulation can be obtained by python run_simulation.py --help. Output provided below:

Functions for simulating European, African, and Admixed populations and
testing PRS building strategies for better generalization to diverse
populations. Additional parameters can be adjusted in the config.py file

optional arguments:
  -h, --help            show this help message and exit
  --sim SIM             Population identifier. Must give unique value if you
                        want to run the simulation multiple times and retain
                        outputs.
  --m M                 # of causal variants to assume
  --h2 H2               heritability to assume
  --snp_weighting {la,yri,ceu,meta}
                        Weighting strategy for PRS building. Can use weights
                        from European (ceu) or African (yri) GWAS as well as
                        weights from a fixed-effects meta analysis of both
                        GWAS (meta) or local-ancestry specific weights (la).
  --snp_selection {yri,ceu,meta}
                        SNP selection strategy for PRS building. Choice
                        between using significant SNPs from a European (ceu)
                        or African (yri) GWAS.
  --pvalue PVALUE       pvalue cutoff to be used for LD clumping
  --ld_r2 LD_R2         r2 cutoff to be used for LD clumping
  --decrease_samples_yri DECREASE_SAMPLES_YRI
                        # of cases used in YRI analysis to represent the lack
                        of non-European data
  --output_dir OUTPUT_DIR
                        location for output data to be written
  --threads THREADS     # of threads to use for parallel processing