Fig. 5: Variance in PSA explained by PRSs.

PRSs for PSA trained on the discovery GWAS meta-analysis was evaluated in GERA, in addition to the three main validation cohorts (PCPT, SELECT and AOU). Joint PRSs were trained on summary statistics from the meta-analysis of the discovery GWAS and replication GWAS. A PRS was first constructed based on independent variants identified using mJAM that reached genome-wide significance. Then a multiancestry genome-wide score was developed using PRS-CSx. a,b, The variance explained by genome-wide PRSs was up to 16.9% in EUR, 18.6% in HIS/LAT, 9.5% in AFR and 15.3% in ASN (a) and decreased as age increased (b). Estimates and full details are found in Supplementary Table 9. Error bars indicate 95% CIs. Variance explained was estimated from partial r2 estimates from linear regression models adjusted for age and genetic ancestry PCs. Sample sizes of each population are specified in a. EUR, European; LAT, Hispanic/Latino; AFR, African; EAS, East Asian; ASN, Asian; OTH, other; AMR, Admixed American; Disc., discovery.