We compared five approaches: (1) no weighting, (2) weighting xQTL SNPs found for any of the molecular phenotypes, (3) weighting SNPs within predefined windows from the molecular features (1Mb, 5Kb, and 1Mb for eQTL, mQTL, and haQTL analyses) to account for distance bias, (4) weighting generic functional SNP in the LDSR baseline model39, and (5) weighting xQTL SNPs that are shared across any of the molecular phenotypes. Over the 19 GWAS datasets (Supplementary Information), weighting xQTL SNPs resulted in equal or more GWAS hits than no weighting, except for inflammatory bowel disease (Table S8). For 8 of the 19 studies, the xQTL-weighted GWAS approach found at least 2 new independent loci (Table S8). By contrast, weighting SNPs within predefined windows from the molecular features as well as weighting SNPs in the LDSR baseline model resulted in little change in detection sensitivity. Interestingly, the gain in sensitivity was not always the highest when we weighted the shared xQTL SNPs. Also, compared to weighting the DGN eQTL SNPs, weighting the union of all xQTL SNPs found in this study identified more additional
