While PRSgene outperformed PRSall in AA, the opposite was observed in EA. This was expected for the following reasons. First, many GWAS findings, such as variants in KLB and GCKR, which reached genome-wide significance in EA, had P-values > 0.05 in AA (i.e., these genes may not be AUD-related in AA for some unknown mechanisms, or variants acting on these genes in AA have not been identified), therefore, they were not included in calculating PRSgene but were used in calculating PRSall in EA. Second, even within genes that have shown associations with AUD in both AA and EA, different causal variants may have been important in each ancestral group. One example is rs2066702 in the ADH1B gene. While relatively common in AA individuals (MAF = 0.18), the variant is rare in EA individuals (MAF = 0.002) (https://www.ncbi.nlm.nih.gov/snp/rs2066702?vertical_tab=true#frequency_tab). This was the only variant selected in ADH1B in calculating PRSgene, resulting in no contribution of ADH1B when calculating PRSgene in EA individuals from the Indiana Biobank. However, for PRSall, multiple common EA variants in ADH1B (e.g., rs2066701, rs1042026, and rs2075633) were included,
