Using Parsimony-Guided Tree Proposals to Accelerate Convergence in Bayesian Phylogenetic Inference.

Area of Science:

• Computational Biology
• Evolutionary Biology
• Phylogenetics

Background:

• Bayesian phylogenetic inference using Markov chain Monte Carlo (MCMC) faces challenges in efficiently exploring tree space.
• Standard MCMC tree moves can lead to slow convergence and poor mixing, hindering accurate phylogenetic analysis.

Purpose of the Study:

• Introduce a novel class of tree proposal moves for MCMC algorithms in Bayesian phylogenetics.
• Enhance the efficiency of sampling tree space by utilizing parsimony scores.

Main Methods:

• Developed parsimony-guided tree proposal moves for MCMC.
• Evaluated performance against standard moves using simulations and six empirical data sets (357-934 taxa, 1740-5681 sites).
• Compared convergence and mixing rates against reference estimates obtained through long MCMC runs and Metropolis coupling.

Main Results:

• Parsimony-guided moves correctly sample the uniform distribution of topologies from the prior in simulations.
• Single MCMC chains using parsimony-guided moves demonstrated an order of magnitude faster convergence compared to standard moves.
• Observed improved mixing, allowing quicker exploration of the most probable trees.

Conclusions:

• Tree moves guided by parsimony scores offer a significant performance improvement over standard MCMC tree proposals.
• Quick and dirty estimates of posterior probability can substantially enhance phylogenetic inference efficiency.
• Future research should explore optimizing posterior probability approximations for further gains in speed and accuracy.