of training data. We also evaluated a meta-analysis PRS (EUR-LAT-meta) and determined that EUR+LAT attained a 19% relative improvement vs. EUR-LAT-meta (Table 3; also see S2 Fig), highlighting the advantages of optimizing mixing weights distinct from meta-analysis weights. Although adding an ancestry predictor to LAT produced a substantial improvement (LAT+ANC vs. LAT), adding an ancestry predictor to EUR+LAT produced an insignificant change in accuracy for EUR+LAT+ANC compared to EUR+LAT; this can be explained by the large negative correlation between the European PRS (EUR) and the proportion of European ancestry within Latino samples (R = -0.75; S9 Table), such that any predictor that includes EUR already includes effects of genetic ancestry. This correlation is far larger than analogous correlations due to random genetic drift in our simulations (S3 Table), suggesting that this systematically lower load of T2D risk alleles in Latino individuals with more European ancestry could be due to polygenic selection (Robinson et al., 2015; Turchin et al., 2012) in ancestral European and/or Native American populations; previous studies using top GWAS-associated SNPs have also reported continental differences in genetic risk for T2D (R. Chen et al., 2012; Corona et al., 2013). We observed a similar correlation (R=−0.77) when using British
