TWAS is also conceptually similar a test for co-localization of signal between expression and complex trait38,39, and we compared to a recently proposed method, COLOC38, that evaluates co-localization of expression at known GWAS risk loci. After matching the false-discovery rate of the two methods in simulations (Methods), TWAS and COLOC had similar power under the single typed causal variant scenario (with slightly lower COLOC power at small GWAS sizes), but TWAS has superior performance when the causal variant was un-typed or in the presence of allelic heterogeneity (Supplementary Figure 10). This is likely due the fact that TWAS explicitly models LD to better capture the un-typed variants.
