We conventionally define PRS as the product of marginal SNP effect size estimates and imputed allelic dosage (ranging from 0 to 2), summed over M SNPs in the model. Conventional P+T utilizes marginal effect size estimates and selects variants with the lowest P value in a locus; therefore, P+T is susceptible to selecting tagging variants. Therefore, we hypothesized that improvement due to leveraging IMPACT annotations could result from prioritizing variants with higher marginal trans-ancestry effect size correlation (Fig. 1d(ii)), suggesting these SNPs are less likely to be associated solely by linkage.
