After running GWAS, we selected approximately independent SNPs associated with the trait (referred to as GWS SNPs or PGS-SNPs), using the LD clumping algorithm implemented in PLINK1.9052. We used the following command: --clump-p1 5e-8 --clump-p2 5e-8 --clump-kb 2000 --clump-r2 0.01. The genotypes of the training population were used as LD reference for clumping. We used here a more stringent LD threshold than classically used (e.g. 0.1 or 0.2) because SNPs with LD r2 as large as 0.1 can still reflect the same signal when GWAS sample size is large (e.g. N > 300,000).
