Estimation of trans-ancestry PRS BridgePRS is an R and bash based package that integrates GWAS summary statistics from two populations. It was designed to improve prediction in a population for which GWASs are relatively under-powered (population 2) but there exist powerful GWAS data in another population (population 1). In addition to GWAS summary statistics, BridgePRS also requires genotype and phenotype data from the two populations for parameter optimisation (test data) and estimation of LD. The LD and test data files can be different or identical. Measures of model fit and SNP weights for the best model for population 2 are returned. In addition, validation data from population 2, can be supplied to make out-of-sample prediction and measures of model fit.
Paper descfribing BridgePRS is available here: https://www.nature.com/articles/s41588-023-01583-9
Clone this repository from your home directory using the following git command:
git clone https://github.com/clivehoggart/BridgePRS.git
Alternatively, download the source files from the github website (https://github.com/clivehoggart/BridgePRS.git)
Required R packages: parallel, glmnet, boot, data.table, doMC, BEDMatrix v2.0.3, optparse and MASS. \ Assumes Plink v1.9 is in your path and BridgePRS directory is in your home directory.
Make BridgePRS shell script and example shell script executable; \ chmod +x bin/BridgePRS.sh \ chmod +x BridgePRS_example.sh
Example script in the installation directory runs the example data by typing \ nohup ./BridgePRS_example.sh > log.txt \ Script utilises 8 processors for paralel computation, this can be changed from the --n_cores argument, see below.
BridgePRS is implemented by clumping GWAS summary statistics from populations 1 and 2 in plink using the following clumping arguments
--clump-p1 1e-1 --clump-p2 1e-1 --clump-kb 1000 --clump-r2 0.01
and subsequently running a series of R scripts to estimate PRS and tune parameters by fitting the PRS to test data. The individual steps can be run separately, in particular code will run faster if user parallelises clumping step by running by chromosomes.
~/BridgePRS/bin/BridgePRS.sh
--outdir. Output directory
--pop1. Label for population 1 result files (default pop1)
--pop2. Label for population 2 result files (default pop2)
--pop1_sumstats. Population 1 sumstats file, can be gzipped (required)
--pop2_sumstats. Population 2 sumstats file, can be gzipped (required)
--by_chr_sumstats. If set assumes sumstats are split by chromosome and are
labelled pop1_sumstatsXby_chr_sumstats, where X indexes the chromosomes
--pop1_bfile. Population 1 plink binary file to calculate PRS in test and
validation sets (required)
--pop2_bfile. As above for population 2 (required)
--pop1_qc_snplist. List of QCed variants for pop1_bfile. Clumping of population 1
sumstats will use only these variants (required)
--pop2_qc_snplist. As above for population 2 (required). Intersection of variants
in pop1_qc_snplist and pop2_qc_snplist will be used for pop2 PRS.
--pop1_test_data. Population 1 phenotype and covariate test file. Requires a
column labelled IID of IDs to use (required)
--pop1_valid_data. As above but for validation data (optional)
--pop1_ld_bfile. Population 1 plink binary file for LD estimation (required)
--pop1_ld_ids. Population 1 IDs of individuals to use for LD estimation, matches on
2nd column, i.e. IID (required)
--pop2_test_data. As above for population 2 (required)
--pop2_valid_data. As above for population 2 (optional)
--pop2_ld_bfile. As above for population 2 (required)
--pop2_ld_ids. As above for population 2 (required)
--cov_names. Column names of covariates to control for in optimising PRS and
calculating R2 (optional)
--pheno_name. Column name of phenotype used in test and validation files (required)
--sumstats_snpID. Sumstat column label of SNP ID (required)
--sumstats_beta. Sumstat column label of SNP effect (required)
--sumstats_allele1. Sumstat column label SNP effect allele (required)
--sumstats_allele0. Sumstat column label of SNP ref allele (required)
--sumstats_n. Sumstat column label for sample size (required)
--sumstats_se. Sumstat column label of SNP effect standard error (required)
--sumstats_frq. Sumstat column label of SNP allele frequency (required)
--strand_check. Logical to check whether summary statistics and reference data
input files are on the same strand. If true/1 only unambiguous SNPs are
used in analyses. If false/0 all variants with matching alleles between
studies are used. With both options effect and reference alleles are
checked and flipped where necessary. If a matching pair of alleles is not
found variants are removed (default 0)
--n_cores. Nummber of processors to use by R scripts
--do_clump_pop1. Logical to run plink to clump population 1 summary statistis
(default 1)
--do_est_beta_pop1. Logical to run R script to estimate PRS using population 1
summary statistics (default 1)
--do_predict_pop1. Logical to run R script to make predictions in population 1
in test and if supplied validation data (default 1)
--do_est_beta_pop1_precision. Logical to run R script to output precision
matrices of best fitting model in population 1 (default 1)
--do_est_beta_InformPrior. Logical to run R script to estimate
PRS in population 2 given prior from population 1 (default 1)
--do_predict_pop2_stage2. Logical to run R script to make prediction in
population 2 in test and if supplied validation data using popultion 1
informed PRS (default 1)
--do_clump_pop2. Logical to run plink to clump population 2 summary statistis
(default 1)
--do_est_beta_pop2. Logical to run R script to estimate PRS using population 2
summary statistics (default 1)
--do_predict_pop2. Logical to run R script to to make predictions in population 2
in test and if supplied validation data using PRS informed only by
population 2 summary statistics (default 1)
--do_combine. Logical to run R script to combine PRS in population 2. Outputs R2
in test and validation data and SNP weights of best model (default 1)pop_stageX_all_preds_test.dat -- PRS predictions for test individuals for prior parameters
pop_stageX_all_preds_valid.dat -- PRS predictions for validation individuals for prior parameters
pop_stageX_best_model_params.dat -- best fitting model parameters in test set
pop_stageX_best_pred_test.dat -- PRS predictions for test individuals for best prior parameters
pop_stageX_best_pred_valid.dat -- PRS predictions for valid individuals for best prior parameters
X=1 -- PRS estimated using sumstats of target population
X=2 -- PRS estimated using using sumstats from both populations
AFR_weighted_combined_preds.dat -- PRS predictions in validation samples of best weighted PRS across all estimated PRS
AFR_weighted_combined_snp_weights.dat -- PRS SNP weights of best weighted PRS across all estimated PRS with columns SNP ID, effect allele, ref allele and effect.
AFR_weighted_combined_var_explained.txt -- Variance explained of best weighted PRS in validation data