Adaptive Tree Proposals for Bayesian Phylogenetic Inference

Area of Science:

• Evolutionary Biology
• Computational Biology
• Phylogenetics

Background:

• Bayesian phylogenetic inference using Markov chain Monte Carlo (MCMC) is crucial for evolutionary studies.
• A long-standing challenge is the design of efficient tree topology proposals for sampling tree spaces.
• Existing methods struggle with efficient exploration of the posterior probability distribution of trees.

Purpose of the Study:

• To introduce and evaluate adaptive tree proposals for Bayesian phylogenetic inference.
• To develop novel adaptive proposals that adjust to the estimated posterior distribution.
• To compare the performance of adaptive proposals against standard and parsimony-guided methods.

Main Methods:

• Development of two adaptive variants of existing tree topology proposals.
• Creation of a novel adaptive proposal informed by the posterior distribution of trees.
• Introduction of a performance metric to evaluate proposals within a mixture model.
• Testing on 11 empirical data sets using Bayesian phylogenetic inference.

Main Results:

• Adaptive proposals consistently improved performance across all tested data sets.
• Observed up to 18-fold increases in mixing efficiency compared to standard proposals.
• Achieved up to 6-fold increases in convergence rate without increasing computational cost.

Conclusions:

• Adaptive tree proposals offer significant performance gains in Bayesian phylogenetic inference.
• These novel methods efficiently sample tree spaces, enhancing the reliability of evolutionary studies.
• Adaptive proposals represent a practical solution to a long-standing challenge in phylogenetics.