We compared the trans-ancestry PRS constructed by PRS-CSx with a European-specific score derived by applying PRS-CS-auto to the European T2D GWAS summary statistics (PRS-CS EUR), and two alternative trans-ancestry scores derived by applying PRS-CS-auto and LDpred2-auto to the T2D meta-GWAS, denoted as PRS-CS Meta and LDpred2 Meta, respectively. PRS-CSx showed better overall prediction accuracy and tail discrimination (Additional File 2: Table S3) and identified more cases in the top percentiles of the PRS distribution than alternative methods in the African and Hispanic/Latino samples of the eMERGE dataset (Fig. 2). The improvement of PRS-CSx relative to PRS-CS EUR highlights the importance of integrating GWAS from diverse ancestries to increase the portability of PRS to non-European populations, while the improvement of PRS-CSx relative to PRS-CS Meta and LDpred2 Meta demonstrates the benefits of explicitly modeling population-specific allele frequencies and LD patterns.
