We found that including in the PRS variants discovered in African ancestry GWASs with population-specific weights results in less disparity in PRS accuracy across ancestries compared to European selected variants, confirming that GWASs in non-bottlenecked populations may yield a more unbiased set of disease variants.25 For example, applying to individuals of African ancestry a PRS derived from GWAS variants and weights discovered in training data from the target population results in a 15.7% higher accuracy compared to using a PRS comprised of variants discovered in a European GWAS (also with African weights). In contrast, the gains in accuracy achieved by sourcing variants from ancestry-matched studies were much lower in European ancestry individuals. Compared to a PRS with variants from an African ancestry GWAS (with European weights), a PRS derived from a European GWAS (also with European weights) only gave a 3.9% higher accuracy. We also observed better generalization of PRSs based on African selected variants across all admixed groups (Figure 2).
