We first evaluated the trans-ancestry PRS constructed by PRS-CSx among the European, African and Hispanic/Latino individuals in the eMERGE study [17, 19, 26]. To define T2D cases and controls in eMERGE, we validated an EMR-based phenotyping algorithm of T2D to apply across the eMERGE samples (see below). We benchmarked the prediction accuracy of the PRS-CSx-derived trans-ancestry PRS against three alternative scores: (i) a European-specific score derived by applying PRS-CS-auto [24] to the European T2D GWAS summary statistics (PRS-CS EUR); (ii) a trans-ancestry score derived by applying PRS-CS-auto to the meta-analysis of the European [8], MEDIA [22], and BBJ [23] T2D GWAS (PRS-CS Meta); (iii) a trans-ancestry score derived by applying LDpred2-auto [27] to the T2D meta-GWAS (LDpred2 Meta). Given that individuals of African descent were underrepresented in the discovery GWAS, the predictive performance of the PRS was expected to be lower in African individuals. We next evaluated the trans-ancestry PRS in four independent self-reported Black cohorts—REGARDS [28], GenHAT [29], HyperGEN [30], and WPC [31]—collected by the University of Alabama at Birmingham (UAB). In all four UAB cohorts, T2D cases were
