The State of Bayesian Phylogenetics: Bayes for the Uninitiated

The bulk of this paper is aimed at an audience that has been exposed, however tangentially, to Bayesian inference of phylogeny; who feel the excitement and intuit its promise, but are intimidated by both the mathematics and the Bayesian/MCMC jargon. We are in a unique position in that, as Paul O. Lewis (2002) states, the field is young enough that “the literature on Bayesian phylogenetics is still small enough that you can have some hope of actually reading all the papers on the topic!” This will undoubtedly become much more difficult within the decade. The purpose of this paper in part is to summarize the research to date; to sort out and explain the jargon, to point out some of the advantages of Bayesian inference, and to elaborate on some possible pitfalls.

Mathematical equations are given, but not overly dwelled on; instead, it is hoped that the reader will get a sense of the process of navigating through tree space and assessing reliability rather than dissecting high-dimensional functions. I have written this review in a specific mindset: to write a paper that I myself would liked to have read when first being exposed to the field of Bayesian phylogenetics. As such, particular attention has been paid to making it as readable as possible, in parts relying on figures to expand on points. In this respect I would like to thank those authors that contributed (unknowingly) many of the figures contained herein.

The reader is expected to have knowledge of rudimentary classical statistics and a working understanding of model-based phylogenetic inference. Numerous papers (Kuhner and Felsenstein, 1994; Huelsenbeck and Hillis, 1993; Huelsenbeck, 1995; Gaut and Lewis, 1995; etc.) have explained in detail the superiority of maximum likelihood in most situations when compared to such inference methods as maximum parsimony, minimum evolution, and various distance methods; I will not, therefore, repeat those arguments here. Nor will I compare Bayesian inference with those proven inferior methods. Rather, what is of interest is whether Bayesian inference of phylogeny can perform at least as well as maximum likelihood.