variants if the non-effect allele could not be clearly determined. Third, MAF is required by certain fine-mapping tools, but it was at times unavailable in the raw data. In such cases, we converted other allele frequencies (such as reference allele frequency or effect allele frequency) to MAF or estimated it from matched 1KGP populations. Fourth, we discarded summary statistics that did not have P-value and effect size [BETA or odds ratio (OR)] for test variants. In some cases for which standard error (SE) of BETA or confidence intervals of OR was missing but effect size was available, we calculated SE using effect size, P, and sample size using the quantile function. In addition, if INFO metric of imputation was available in the raw data, variants with INFO <0.9 were filtered out. Thus, for all summary statistics in the Neale Lab UKBB cohort, we excluded variants with INFO <0.9.
