Parallel power posterior analyses for fast computation of marginal likelihoods in phylogenetics

Area of Science:

• Computational Biology
• Phylogenetics
• Statistical Modeling

Background:

• Bayesian phylogenetic inference relies on marginal likelihood estimation.
• Current methods like path-sampling and stepping-stone-sampling are computationally intensive due to extensive Markov chain Monte Carlo (MCMC) simulations.
• These demanding computations limit the feasibility of complex phylogenetic analyses.

Purpose of the Study:

• To develop and present a general parallelization strategy for accelerating marginal likelihood estimation in Bayesian phylogenetics.
• To demonstrate the effectiveness of this strategy across various statistical models, with a focus on molecular substitution models.
• To reduce the substantial computational time required for marginal likelihood calculations.

Main Methods:

• Implemented a parallelization strategy distributing power posterior MCMC simulations and likelihood computations across available CPUs.
• Applied the strategy to molecular substitution models within the Bayesian phylogenetic inference framework.
• Tested the approach on two empirical phylogenetic datasets.

Main Results:

• Achieved significant reductions in runtime for marginal likelihood estimation.
• Demonstrated an average performance increase of 1.96x with just two CPUs.
• Observed a near-linear performance increase with more CPUs, reaching 13.3x with 16 CPUs.
• Made the methods available in the open-source software RevBayes.

Conclusions:

• The proposed parallelization strategy substantially accelerates marginal likelihood estimation in Bayesian phylogenetics.
• This approach makes previously time-prohibitive analyses (days, weeks, months) computationally feasible.
• The strategy's general applicability and implementation in RevBayes facilitate broader adoption in phylogenetic research.