Model Building and Selection in Bayesian Phylogenetic Reconstruction

• Activity Number: 282
• Type: Topic Contributed
• Date/Time: Tuesday, August 5, 2003 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Stat. Sciences
• Author(s): Marc A. Suchard*+
• Companies: University of California, Los Angeles
• Keywords: Markov chain Monte Carlo ; phylogenetics ; model selection ; HIV recombination ; evolution

Abstract

Bayesian phylogenetic reconstruction provides a flexible framework in which to explore model building and selection of nested, non-nested and varying dimensional models using reversible jump Markov chain Monte Carlo. I develop the tools necessary to compare both common and novel models of reconstruction using Bayes factors. I propose new priors to test modeling restrictions without assuming an outgroup and allow for the computationally efficient identification of unknown evolutionary relationships. I introduce evolutionary similarity between multiple genes in the same biological pathways and describe methods to test for it when data are sparse. I develop a new class of multiple changepoint models that allow evolutionary relationships and pressures to vary along the sequences used in recombinant analysis. Finally, I propose a new statistical model to improve estimation of Bayes factors by incorporating results from multiple posterior simulations under differing priors. I apply these methods to datasets of evolutionary and medical importance, including mitochondrial and nuclear DNA from primates, ribosomal RNA from across the four kingdoms of life and entire HIV genomes.