In this study, we use the P+T PRS framework. We constructed PRS from either EUR summary statistics or EAS summary statistics and evaluated their predictive performance on individual EAS phenotypes. For PRS-EUR, we utilized genome-wide summary statistics from EUR as reported in their publicly available version. For PRS-EAS, we held out 5,000 individuals for PRS analysis and conducted GWAS using the remaining individuals to avoid overfitting. For each trait separately, we restricted our analysis to variants that exist in both GWAS summary statistics and post-imputation genotype data of EAS individuals used for PRS analysis (imputation quality of r2 > 0.3 in minimac3). A detailed description related to the genotyping platform and imputation strategy is provided in a previous report38. We excluded the MHC region in this analysis.
