Parallel Metropolis coupled Markov chain Monte Carlo for Bayesian phylogenetic inference.

Area of Science:

• Computational Biology
• Phylogenetics
• Statistical Modeling

Background:

• Bayesian phylogenetic inference relies on approximating posterior probability distributions of trees.
• Markov chain Monte Carlo (MCMC) is the primary numerical method for this approximation.
• Standard MCMC can be trapped in local optima, while Metropolis coupled MCMC (MC3) explores better but is slower.

Purpose of the Study:

• To develop a parallel algorithm for MC3 to enhance phylogenetic tree estimation.
• To maintain the ability of MC3 to explore multiple peaks in tree posterior distributions.
• To achieve faster execution times compared to standard MC3 implementations.

Main Methods:

• Implementation of a parallel algorithm for Metropolis coupled MCMC (MC3).
• Utilized two parallel programming models: message passing and shared memory.
• Tested the algorithm on both small and large datasets.

Main Results:

• The parallel MC3 algorithm effectively explores multiple peaks in the posterior distribution of trees.
• Achieved nearly linear speed improvements across both message passing and shared memory models.
• Demonstrated efficiency for both small and large datasets.

Conclusions:

• The proposed parallel MC3 algorithm significantly enhances the speed and efficiency of Bayesian phylogenetic inference.
• This method overcomes the limitations of standard MCMC and slower MC3 implementations.
• Offers a scalable solution for complex phylogenetic analyses.