Computer Science

Graduate Research Forum Details

Finite mixture models for phylogenetic reconstruction

Abstract

Reconstructing evolutionary relationships among species is fundamental to all of comparative biology. Incorrect inference can have dramatic impacts across the biological sciences, leading to erroneous conclusions with serious consequences for our understanding of, for example, the evolutionary process, the spread of infectious disease, and the impact of human activities on biodiversity. Current parametric reconstruction methods - primarily maximum likelihood (ML) and Bayesian Markov Chain Monte Carlo (BMCMC) analysis - rely on the assumption that the processes governing the evolution of species are largely homogeneous at the molecular level, and assumption known to be strongly violated in numerous cases. I will discuss the algorithmic process of phylogenetic reconstruction and present evidence from extensive computer simulations that heterogeneity in the evolutionary process can cause existing methods to become strongly biased and statistically inconsistent under a range of conditions. I will then present a prototype finite mixture model for incorporating evolutionary heterogeneity into existing ML and BMCMC techniques and show preliminary data that mixture models can dramatically improve accuracy when the evolutionary process is heterogeneous.