A number of methods for simultaneous analysis of multiple traits in population-based GWAS have been published (e.g. [4], [6]–[19]). Although a few of the methods have been compared to newly proposed methods [12], [15] and some of the methods have been compared to univariate analysis [4], [7], [12], little is known about their relative performances. Here, we performed the first direct comparison of several multivariate (MV) GWAS methods using simulated data. We included six methods, with a focus on methods already implemented in freely available software: the multivariate test of association MQFAM implemented in the genetic association analysis software PLINK (MV-PLINK) [7], a Bayesian multiple phenotype test implemented in SNPTEST (MV-SNPTEST) [20], the R package MultiPhen (MultiPhen) [12], a Bayesian model comparison and model averaging for multivariate regression in BIMBAM (MV-BIMBAM) [21], [22], the Principal Component of Heritability Association Test (PCHAT) [4], and a Trait-based Association Test that uses Extended Simes procedure (TATES) [15]. These can be classified into direct, indirect and univariate-based methods (Figure 1). MV-SNPTEST, MultiPhen and MV-BIMBAM are direct MV methods, in which the effects of the
