morePhyML: improving the phylogenetic tree space exploration with PhyML 3

Mol Phylogenet Evol. 2011 Dec;61(3):944-8. doi: 10.1016/j.ympev.2011.08.029. Epub 2011 Sep 8.

Abstract

PhyML is a widely used Maximum Likelihood (ML) phylogenetic tree inference software based on a standard hill-climbing method. Starting from an initial tree, the version 3 of PhyML explores the tree space by using "Nearest Neighbor Interchange" (NNI) or "Subtree Pruning and Regrafting" (SPR) tree swapping techniques in order to find the ML phylogenetic tree. NNI-based local searches are fast but can often get trapped in local optima, whereas it is expected that the larger (but slower to cover) SPR-based neighborhoods will lead to trees with higher likelihood. Here, I verify that PhyML infers more likely trees with SPRs than with NNIs in almost all cases. However, I also show that the SPR-based local search of PhyML often does not succeed at locating the ML tree. To improve the tree space exploration, I deliver a script, named morePhyML, which allows escaping from local optima by performing character reweighting. This ML tree search strategy, named ratchet, often leads to higher likelihood estimates. Based on the analysis of a large number of amino acid and nucleotide data, I show that morePhyML allows inferring more accurate phylogenetic trees than several other recently developed ML tree inference softwares in many cases.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Computational Biology / methods*
  • Likelihood Functions
  • Phylogeny*
  • Software*
  • Time Factors