Impacts of Misspecifying the Evolutionary Model in Phylogenetic Tree Estimation

Abstract We consider phylogenetic tree estimation with emphasis on estimating the number of groups (clades). We rarely know the full evolutionary model, so we want to understand the impact of model estimation errors. Sensitivity to misspecifying the model or model parameters depends on how distinct the clades are, so it is important to consider differing degrees of clade resolution. We do this by varying the macroscopic growth rate and microscopic mutation rate of the taxa. We simulate DNA sequence data using coalescent theory to simulate the sample genealogy, and one of several mutation models. For each case, we compute all pairwise distances between sequences using the true and several alternate models. The within-group variance (with distance data represented in principal coordinates) is used to choose the number of clades. We conclude that the estimated number of clades can be sensitive to model estimation errors, to an extent determined by clade resolution