We selected the most representative GWAS which contains maximum causal blocks for each MeSH term. Then we inspected the causal blocks in GWAS pairs and calculated the PP of a variant influencing both traits (potential pleiotropy) using gwas-pw (model 3) (29). Since the overlapping samples or comorbid samples were largely unreported accompanying released GWAS summary statistics, we empirically set the expected correlation (-cor in gwas-pw) between two traits as described in a recent simulation study (30). For the UKBB cohort studies we were certain that they contain overlapping samples and incorporate similar categories of traits, we varied the expected correlation of 0.09, 0.18, 0.27, 0.36, 0.45 according to the hierarchy of MeSH tree. For example, the tree numbers of Heart Failure and Atrial Fibrillation are C14.280.434 and C14.280.067.198 respectively which all belong to C14.280 parent node of Heart Diseases. We hence classified the relationship between these two diseases into Level 2, so the expected correlation in summary statistics between Heart Failure and Atrial Fibrillation was set to 0.18. For non-UKBB cohort studies which we cannot ascertain overlapping samples, the expected correlation was uniformly set to 0.
