but the improvement is limited even with sophisticated statistical methods due to the differences between the discovery GWAS and the target datasets [23]. Our gene-based PRS framework leverages the concordant variants across different populations and discriminates variants unrelated to the disease of interest leading to the improved performance of PRS. Using concordant variants also reduces the chance of selecting the wrong independent index variants due to a mismatch of LD patterns among the discovery and target datasets, as well as the external LD reference panels. Moreover, as PRSintergenic were not associated with AUD in our analyses, the performance of PRSgene was further improved by focusing on concordant variants within gene boundaries. RRSgene had superior performance in all our AA target datasets, thus, we conclude that this strategy can be used to improve the performance of PRS when the discovery GWAS sample sizes are not sufficiently large, notable in admixture populations, and other groups that have been underrepresented in GWAS studies to date.
