Input for --score in PLINK 1.9 when generationg PRS using GWAS summary statistics

Question

I'm conducting a simulation and I need to obtain polygenic risk scores (PRS) using genome wide association studies (GWAS) summary statistics. It is known that the GWAS will give BETA and OR for quantitative and binary phenotypes respectively. When conducting the simulation for quantitative traits I used the following code to obtain PRS from GWAS performed:

./plink1.9 --bfile binary_file1_name --linear interaction --pheno pheno_file_name --covar cov_file_name --parameters 1 --allow-no-sex
./plink1.9 --bfile binary_file2_name --score plink.assoc.linear 2 4 7 header --out prs_quantitative

This is fine, where plink.assoc.linear file contains BETA.

When conducting simulation for binary traits I used the following code to obtain PRS from GWAS performed:

./plink1.9 --bfile binary_file1_name --logistic interaction --pheno pheno_file_name --covar cov_file_name --parameters 1 --allow-no-sex
./plink1.9 --bfile binary_file2_name --score plink.assoc.logistic 2 4 7 header --out prs_binary

I'm not sure whether we need to transform OR to BETA before inputting the file to --score in performing this for binary traits or, simply use the original plink.assoc.logistic which has OR (instead of BETA) to go ahead.

Simply, do we need to input file containing BETA values "only" for --score or either a file containing OR values or BETA values is fine?

Answer 1

If the GWAS summary statistics file provide SNP effects (BETA), use them directly in PLINK --score to compute PRS. For binary traits, GWAS summary statistics file usually outputs odds ratio (OR), and hence, when constructing PRS using OR, we need to first trasnform OR to BETA (where BETA = log(OR)) and then use the same --score command to generate PRS.

Additional informaion: Alternatively, this R package (GxEprs) can be used to construct PRSs using GWAS summary statistics for both quantitative and binary traits. If you need to opt out of additional OR to BETA transformation step, this R package is useful, as it implements this transformation within its in-built function PRS_binary() for generating PRSs for binary traits.