Our finding that the predictive power of polygenic risk scores is poorer in non-European populations, particularly among African ancestry individuals, is almost certainly due to the use of European ancestry training data, combined with the greater genetic diversity within many African populations, the phenotypic consequences of which are not well-captured by European ancestry GWAS. (Differences in heritability across populations for the same trait might also exist and could also impact the performance of polygenic scores in particular populations. Ultimately, the performance of a particular polygenic score, as applied in a particular population, should be evaluated with respect to the population-specific heritability for that phenotype). Indeed, the relative reduction of genetic diversity among European ancestry and other non-African populations (due to past population bottlenecks) is a natural limitation of European ancestry GWAS. In addition to the important goal of collecting more samples from more populations, there can also be improvements in the manner in which variant frequencies and linkage disequilibrium are handled in polygenic scoring studies. The results presented here provide benchmarks for the performance of polygenic scores in diverse populations,
