Functionally informed PRS might, to some extent, compensate for population-specific LD differences between populations. Hence, we hypothesized that IMPACT-informed PRS would improve standard PRS, more so in the trans-ancestry prediction framework, in which EUR PRS models predict EAS phenotypes, than in a within-population framework, in which EAS PRS models predict EAS phenotypes. Here, we define within-population PRS as PRS-EAS and trans-ancestry PRS as PRS-EUR to avoid confusion. To compare PRS model improvements directly between PRS-EAS and PRS-EUR, we evaluated prediction accuracy on the same individuals. Briefly, we partitioned the BBJ cohort to reserve 5,000 individuals for PRS testing, derived GWAS summary statistics from the remaining individuals, and performed P+T PRS modeling and prediction as done above (Fig. 5d, Extended Data Fig. 9, Supplementary Figs. 18 and 19, Supplementary Tables 21 and 22 and Methods). For functionally informed PRS-EAS, we selected lead IMPACT annotations from S-LDSC results using GWAS summary statistics, as done above, on the partition of the BBJ cohort excluding the 5,000 PRS test individuals. We defined improvement as the percent increase in R2 from standard to functionally informed
