TIC Seminar - Desirable Properties of Models for Phylogenetic Analysis

Barbara Holland, University of Tasmania

Joint work with: Jeremy Sumner, David Liberles, Michael Woodhams, Michael Charleston

These days the most widely used methods of phylogenetic inference are based on explicit models of how sequences (DNA nucleotides, amino acids, or codons) evolve. Recently our research group in Hobart has been interested in a property of sequence evolution models that seems fundamental, but surprisingly it is not attained by many popular models. This property is that of closure under multiplication. The closure property ensures that if I take a transition matrix from a particular class of models (e.g. F81) and multiply it by another matrix from the same class, the resulting process is still in the same class. Such a property becomes important if we believe that evolution is likely to be a heterogeneous process that might act differently in different branches of the tree.

In this talk I will discuss which models have this property, and which don't. I will give some results from simulation studies that explore the consequences of lack of closure for phylogenetic inference in terms of both branch lengths and topological accuracy. I will also discuss the consequences of lack of closure for codon models when they are used to estimate the ratio of synonymous to non-synonymous changes.