P+T rather than LDpred2,22 or AnnoPred21, which compute a posterior effect size estimate for all SNPs genome-wide based on membership to functional categories. With P+T, we can partition the genome by IMPACT-prioritized and deprioritized SNPs, whereas the assumptions of the LDpred and AnnoPred models do not support the removal of variants, making it difficult to assess improvement directly due to IMPACT prioritization. Moreover, these models have not been designed or tested explicitly for the trans-ancestry application of PRS and thus are beyond the scope of our work.
