We quantified the accuracy of a PGS based on GWS SNPs as well as the variance explained by SNPs within GWS loci, in eight different datasets independent of our discovery GWAS meta-analyses. These datasets include two samples of EUR from the UKB (n = 14,587) and the Lifelines study (n = 14,058), two samples of AFR from the UKB (n = 6,911) and the PAGE study (n = 8,238), two samples of EAS (n = 2,246) from the UKB and the China Kadoorie Biobank (CKB; n = 47,693), one sample of SAS from the UKB (n = 9,257) and one sample of HIS from the PAGE study (n = 4,939). Analyses were adjusted for age, sex, 20 genotypic principal components and study-specific covariates (for example, recruitment centres). Genotypes of EUR UKB participants were imputed to the Haplotype Reference Consortium (HRC) and to a combined reference panel including haplotypes from the 1KG Project and the UK10K Project. To improve variant coverage in non-EUR participants of UKB, we re-imputed their genotypes to the 1KG reference panel, as described previously38. Lifelines samples were
