Taken together, our study showed substantial differences in power between the methods, dependent on the simulation scenario. For some of the simulation scenarios, a large increase in power of multivariate compared to univariate analyses was observed, which suggests that the multivariate methods might be able to identify genetic variants that are currently not identifiable by standard univariate analysis. Overall, MV-PLINK, MV-SNPTEST, MultiPhen and MV-BIMBAM performed best for the majority of the tested scenarios, with a remarkable increase in power for scenarios with an opposite sign of genetic and residual correlation. As a consequence, results of these methods will be biased towards QTLs that cause a genetic correlation that is opposite in sign to the residual correlation. PCHAT and TATES showed a robust performance over all simulation scenarios and are therefore recommended to use if one aims to obtain a reflection of the underlying genetic architecture of the traits.
