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 UK Biobank type 2 diabetes samples as European training data (row 4 of Table 1; see Methods), confirming that this negative correlation is not caused by population stratification in DIAGRAM. As in our simulations, predictions using Latino effect sizes that were not adjusted for genetic ancestry (LATunadj, EUR+LATunadj, EUR+LATunadj+ANC, as compared to LAT, EUR+LAT, EUR+LAT+ANC) were much less accurate (S10 Table), consistent with the fact that these predictions were dominated by genetic ancestry (S9 Table). We also computed predictions for each method using imputed SNPs from the SIGMA T2D data set; this did not improve prediction accuracy, but predicting using two training populations still achieved the highest accuracy (S11 Table).
