Detecting Bias in Phylogenetic Inference: Empirical Tests of Posterior Predictive Model Assessment

Abstract

Inferring the ‘Tree of Life’ depends heavily on statistical models of sequence evolution. As is the case with all statistical inference, these models are only approximations of the actual evolutionary processes they are meant to describe. When a model poorly describes a given dataset, the resulting phylogeny can be inaccurate. Methods that directly assess goodness of fit are increasingly being recognized as the means to circumvent this problem, although this framework is still in its infancy within phylogenetics. Here we use phylogenies inferred from several hundred mitochondrial genomes to assess the performance of these new approaches at detecting poor absolute model fit and related problems. This study provides clear examples of when these new methods prove useful. We also detect some unforeseen behaviors for larger, more complex datasets where these methods are most critically needed. These issues point the way forward for future development of this emerging framework in phylogenetics.