Next, we focused on evaluating the power of the TWAS approach to detect significant expression-trait associations using GWAS summary data from complex traits (equivalent to TWAS from individual level data; Methods, Supplementary Figure 5). For comparison, we also measured power to detect significant SNP-trait associations through standard GWAS (testing each SNP individually) and eQTL-based GWAS (eGWAS, where the best eQTL in each gene is the only variant tested for association to trait), with all three tests corrected for their genome-wide testing burdens. Using real genotype data, we simulated a causal SNP-expression-trait model with realistic effect-sizes and measured the power of each strategy to identify genome-wide significant variants (accounting for 1 million SNPs for GWAS and 15,000 expressed genes using family-wise error rate control). Over many diverse disease architectures TWAS substantially increased power when the expression-causing variants were un-typed or poorly tagged by an individual SNP (Figure 5, Supplementary Figures 6–11). The greatest power gains were observed in the case of multiple causal variants: 92% power for TWAS compared to 18% and 25% for GWAS and eGWAS. This scenario would correspond
