Consistency of Bayesian inference of resolved phylogenetic trees

Abstract

Bayesian inference is now a leading technique for reconstructing phylogenetic trees from aligned sequence data. In this short note, we formally show that the maximum posterior tree topology provides a statistically consistent estimate of a fully-resolved evolutionary tree under a wide variety of conditions. This includes the inference of gene trees from aligned sequence data across the entire parameter range of branch lengths, and under general conditions on priors in models where the usual `identifiability' conditions hold. We extend this to the inference of species trees from sequence data, where the gene trees constitute `nuisance parameters', as in the program *BEAST. This note also addresses earlier concerns raised in the literature questioning the extent to which statistical consistency for Bayesian methods might hold in general.