We sought to understand what factors impacted PRS generalizability across the different variant selection approaches. GWASs performed in European and African ancestry populations (for SNPs with MAF ≥ 0.01) were both more likely to identify significant variants that were more common in their own population than in the other. Approximately 60% of variants identified in European ancestry populations had MAFs less than 1% in African ancestry populations and vice versa; however, given the underlying assumption of homogeneity, the smaller number of variants selected by a meta-analysis of the two populations tended to have more similar MAFs (Figure 4A). Although European and African ancestry GWASs were both better powered to detect variants at intermediate frequencies within the same study population, GWASs in European ancestry populations may be unable to capture derived risk variants that have remained in Africa, which could be the result of genetic drift, whereas GWASs in African ancestry populations are not subject to this bias.25
