Finding high posterior density phylogenies by systematically extending a directed acyclic graph.

Area of Science:

• Computational Biology
• Evolutionary Biology
• Statistical Methods

Background:

• Bayesian phylogenetics commonly uses Markov chain Monte Carlo (MCMC) for posterior distribution estimation.
• MCMC methods explore tree space via random walks using local rearrangements.
• Previous attempts at systematic search were hindered by the vast number of probable trees.

Purpose of the Study:

• To develop novel methods for efficient exploration of tree space in Bayesian phylogenetics.
• To leverage subsplit directed acyclic graphs (sDAGs) for representing multiple trees simultaneously.
• To identify collections of trees with high posterior density more effectively.

Main Methods:

• Utilizing subsplit directed acyclic graphs (sDAGs) to represent a collection of trees.
• Developing methods to introduce, rank, and select local rearrangements on sDAGs.
• Comparing a novel sDAG-based systematic search with traditional MCMC random walks.

Main Results:

• One proposed method successfully recovered high posterior density trees across diverse datasets.
• A simpler strategy of aggregating trees into an sDAG proved computationally faster.
• The sDAG aggregation approach yielded a higher fraction of probable trees.

Conclusions:

• sDAGs offer a powerful framework for enhancing Bayesian phylogenetic inference.
• Aggregating trees into an sDAG is a computationally efficient strategy.
• This approach improves the identification of probable trees in phylogenetic analyses.