Multi-Trait Genome Wide Association Tests Result in a Reduction in False Positives

In plant and animal genetic research and breeding, typical experiments measure multiple traits. However, standard quantiative trait locus and genome-wide association tests only consider one trait at a time. Multivariate tests are of interest because they could provide an increase in power by borrowing information from correlated phenotypes. Using a Bayesian framework, we implemented three tests that use information from multiple phenotypes. We compared their performance with a standard single-trait test using an extensive set of simulated data. We find that the Bayesian analog of the Hoteling test, while slightly less powerful than other approaches, incurs far fewer false positives. A model that identifies associations for a trait conditional on all other phenotypes is the least powerful, but identifies markers missed by other tests.