A topology-marginal composite likelihood via a generalized phylogenetic pruning algorithm.

Area of Science:

• Computational Biology
• Phylogenetics
• Statistical Inference

Background:

• Bayesian phylogenetics presents significant computational challenges.
• Traditional Markov chain Monte Carlo (MCMC) methods involve computationally intensive random-walk proposals.
• Variational phylogenetics offers an alternative by approximating the posterior distribution, often using stochastic gradient descent.

Purpose of the Study:

• To develop a novel, efficient algorithm for Bayesian phylogenetic inference.
• To leverage the unique structure of phylogenetic trees for computational gains.
• To provide a faster method for estimating phylogenetic parameters, particularly branch lengths.

Main Methods:

• Generalization of the Felsenstein pruning algorithm (sum-product algorithm).
• Simultaneous marginalization of ancestral states and subtrees to compute a composite-like likelihood.
• Application of the algorithm for rapid point estimation of branch lengths in multi-tree models.

Main Results:

• The generalized pruning algorithm significantly accelerates the estimation of phylogenetic branch lengths.
• Point estimates obtained are consistent with results from lengthy MCMC runs and existing variational methods.
• Demonstrated rapid and accurate parameter estimation for complex phylogenetic models.

Conclusions:

• The generalized pruning algorithm offers a computationally efficient alternative to MCMC and standard variational methods for phylogenetic inference.
• This approach provides a valuable foundation for developing future variational inference algorithms in phylogenetics.
• The method enables faster, yet accurate, estimation of key phylogenetic parameters.