In research settings, trans-ancestry PRS are often derived from multi-ethnic meta-GWAS [7, 44] or trained in each target population separately [45, 46]. However, the former approach does not model population-specific allele frequency and LD patterns, which limits the performance of PRS, while the latter approach requires assigning individuals to discrete ancestral groups to optimize PRS estimation, which is challenging in clinical applications because self-reported race/ethnicity may be inconsistent with genetic ancestry. The patient group for returning PRS results may also contain admixed individuals who cannot fit into ancestry categories easily. Communication of PRS results in clinical settings would thus be facilitated by development of a trans-ancestry PRS without stratifying patients into individual ancestral groups.
