Bayesian and maximum likelihood phylogenetic analyses of protein sequence data under relative branch-length...

Area of Science:

• Computational Biology
• Bioinformatics
• Evolutionary Biology

Background:

• Bayesian phylogenetic inference is explored as an alternative to maximum likelihood, especially for large molecular sequence datasets.
• The study investigates Bayesian inference performance using empirical and simulated protein sequence data.
• Conditions examined include relative branch-length differences and model violation.

Purpose of the Study:

• To evaluate the performance and robustness of Bayesian phylogenetic inference compared to maximum likelihood.
• To assess the impact of branch-length variation and model misspecification on phylogenetic accuracy.
• To determine the reliability of Bayesian posterior probabilities versus bootstrap proportions.

Main Methods:

• Phylogenetic tree reconstruction using Bayesian inference and maximum likelihood methods.
• Analysis of both empirical and simulated protein sequence datasets.
• Inclusion of conditions with varying relative branch lengths and model misspecification, incorporating gamma distribution for rate variation.

Main Results:

• Bayesian posterior probabilities offer more optimistic estimates of subtree reliability than bootstrap proportions on empirical data.
• Bayesian inference demonstrates robustness to branch-length differences and model violation, particularly with gamma-corrected models.
• Accuracy of tree reconstruction degrades with increasing branch-length ratios, with a more rapid decline when two branches are elongated.

Conclusions:

• Bayesian phylogenetic inference is robust to biologically relevant levels of branch-length variation and model misspecification.
• It presents a viable and promising alternative to maximum likelihood for phylogenetic tree inference from protein sequence data.
• Gamma-corrected Bayesian inference outperforms maximum likelihood and uncorrected Bayesian inference under model violation.