Phylogenetic reconstruction with disk-covering and Bayesian approaches

The DCM approach is commonly used to divide the dataset into smaller subproblems, analyze each subproblem using a base method to obtain subtrees, then recombine these subtrees to build the final phylogeny over the whole dataset. In recent years, the new and improved method MrBayes, a Bayesian Markov Chain Monte Carlo (MCMC) approach is widely used for phylogeny analysis. In this paper, a new method for large scale Bayesian phylogeny analysis is proposed. This new method (DCM3-MrBayes) is an improved version of Rec-I-DCM3 (recursive iterative disk-covering method), which uses a divide-and-conquer approach and is designed for large dataset analysis. To integrate MrBayes with Rec-I-DCM3, we have to deal with some unique problems and proposed several methods to tackle these problems. Our improvements include a cache system that can avoid unnecessary computations and a method to eliminate weak branches indicated by the Bayesian analysis to filter out potential bad branches. Our experiments on simulated datasets shows promising improvement over the original DCM. One of the most important advantages of using Bayesian method for phylogeny reconstruction is being able to calculate the posterior probabilities. A divide-and-conquer Bayesian method looses its ability to calculate the posterior probabilities due to the fact that each subproblem generates its own posterior probabilities, which posts some difficulties for obtaining the posterior probability for the whole problem. In order to preserve the advantage of Bayesian approach, we also introduce an algorithm that calculates the posterior probabilities of the whole phylogeny from the subproblemspsila posterior probabilities.