SVDquintets: a New Tool for Species Tree Inference

Abstract

1Species tree inference is complicated by the fact that different segments have the genome can have evolutionary histories that differ from each other and from the evolutionary history of the species as a whole. One source of this mismatch is incomplete lineage sorting (ILS), which is commonly modeled by the multispecies coalescent process. Here we derive site pattern probabilities under the multispecies coalescent model, the Jukes-Cantor substitution model, and a relaxed molecular clock for five species at a time. As a result, we can demonstrate that the rank results that form the theoretical basis for SVDQuartets also appear to hold for sets of five species. Based on this, we have developed a new species tree inference algorithm: SVDquintets. Comparison with SVDQuartets shows improved species tree inference under a variety of simulated data settings.