Given that not all GWAS have summary statistics available, two fine-mapping methods are applied, one using full summary statistics, and another using LD information only. For studies with full summary statistics, we use GCTA-COJO to identify independent signals, and then perform per-signal conditional analysis adjusting for other independent signals in a ±2 Mb region from the lead variant (18). For every conditionally independent signal, fine-mapping using the Approximate Bayes Factor approach is performed (19). For GWAS without summary statistics, we use the PICS method with an LD reference from the most closely matched 1000 genomes project superpopulation (20). This enables us to estimate the probability that each variant is causal across 133,441 study-lead variant associated loci. Output from both methods provides a posterior probability (PP) for each variant being causal for a given association. In order to identify complex traits and diseases that share common molecular mechanisms, we also perform cross-trait colocalization analyses for 3,621 GWAS studies with summary statistics.
