again highlighting the advantages of optimizing mixing weights distinct from meta-analysis weights. Adding an ancestry predictor to EUR+SAS produced an insignificant change in accuracy for EUR+ SAS +ANC compared to EUR+SAS; we note a modest correlation between each prediction method and the proportion of European-related ancestry (Reich, Thangaraj, Patterson, Price, & Singh, 2009) within South Asian samples (see S16 Table). We repeated our analyses using stratified 10-fold cross-validation to estimate mixing weights (see Methods). We observed that methods that use two training populations continued to substantially outperform PRS using a single training population despite a decrease in prediction adjusted R2 (vs. Table 4) for each method, consistent with the limited sample size for estimating mixing weights (S17 Table).
