Hence, we tested this hypothesis before assessing PRS performance. We selected 21 of 29 summary statistics shared between EUR and EAS with an identified lead IMPACT association in both populations. Then, using EUR lead IMPACT annotations for each trait, we partitioned the genome in three ways: (1) the SNPs within the top 5% of the IMPACT annotation, (2) the SNPs within the bottom 95% of the IMPACT annotation and (3) the set of all SNPs genome wide (with no IMPACT prioritization). We then performed stringent LD pruning (r2 < 0.1 from EUR individuals of phase 3 of 1000 Genomes53), guided by the EUR GWAS P value, to acquire sets of independent SNPs to compute a EUR–EAS marginal effect size estimate correlation (Methods).
