In UK Biobank Europeans, a GWAS significant European-derived PRS (with European variants and weights) had a partial-R2 of 1.6%, 1.2%, and 1.5%, respectively, for HbA1c levels, asthma, and prostate cancer; the same PRS applied to African ancestry individuals with approximately 13.1% European ancestry8 only explained 0.07%, 0.38%, and 0.19% (Figure S6). Although the proportion of variation explained by the PRS (partial-R2) was consistently lower in UK Biobank African ancestry individuals compared to Europeans, prediction was improved through the inclusion of variants or weights discovered in an independent African ancestry cohort across all traits (Figure S6). Interestingly, we found that a linear combination of the best-performing PRS with European-discovered variants and African ancestry-discovered variants improved accuracy substantially (Table S2), supporting our simulation finding that a combined PRS that includes variants from the target population, even at smaller sample sizes, is the optimal approach for constructing PRS in admixed and non-European individuals.
