COLOC uses summary data from eQTL and GWAS studies and a Bayesian framework to identify the subset of GWAS signals that co-localize with eQTLs. We sought to compare TWAS to the COLOC-estimated posterior probability of association (PPA) being shared for both phenotypes (PP4 in the COLOC implementation). COLOC additionally evaluates the hypothesis of multiple independent associations (PP3), but this is more general than the proposed TWAS model and was not tested. Because COLOC relies on priors of association to produce posterior probabilities of co-localization, we sought to identify a significance threshold that would make a fair comparison to the TWAS p-value-based threshold. Specifically, we ran both methods on a realistic null expression simulation (with the generative model described previously): the expression was sampled from a null standard normal for 1,000 individuals and eQTLs computed; the trait associations were derived from a simulated 300,000 GWAS with a single typed causal variant that explained 0.001 variance of the trait (high effect). We believe this scenario is both realistic and consistent with the GWAS assumptions of COLOC. We then empirically identified the statistical
