We identified putative causal variants for MD using PolyFun v1.0.047 and SuSiE v0.11.92.48 We restricted the meta-analyzed summary statistics to variants lying outside of the HLA region, with imputation INFO score > 0.6 and MAF > 0.001. We computed prior causal probabilities based on 187 functional enrichments from the baselineLF2.2 model,47 using an L2-regularized extension of stratified LD Score regression (S-LDSC) implemented in PolyFun, using LD scores derived from the UK Biobank and provided with PolyFun. We performed fine-mapping of genome-wide significant loci with windows defined by clumping (as described above). We used SuSiE to perform the fine-mapping, assuming a single causal variant in each case (as single-variant fine-mapping does not require population-accurate estimation of LD47). We identified variants of interest as having a posterior inclusion probability in the causal set (PIP) > 0.95. We ranked all variants by PIP and defined 95% credible causal sets of variants as the minimum set of variants whose PIPs summed to >=0.95. We mapped credible causal sets to 19,878 protein coding genes. We defined high-confidence genes as those containing all variants within the
