The posterior effect sizes of variants were estimated using PRS-CS (Ge et al., 2019) through a Bayesian regression framework using continuous shrinkage priors. This method models local LD patterns and variants with small effects are excluded from analysis, therefore, neither LD pruning nor P-value thresholding are needed. PRS-CS requires an external LD reference panel and European samples from the 1000 Genomes project (phase 3, NCBI GRCh37) were used. Variants that were present in the discovery datasets, LD reference panel, and the COGA dataset were included (N=2,077,165). Posterior effect sizes were estimated for 326,000 variants by PRS-CS, and these variants were used to calculate PRS using PLINK (Chang et al., 2015, Purcell et al., 2007).
