A reversible jump method for Bayesian phylogenetic inference with a nonhomogeneous substitution model.

Area of Science:

• Computational Biology
• Phylogenetics
• Molecular Evolution

Background:

• Nonhomogeneous substitution models are crucial for phylogenetic inference when evolutionary processes vary across lineages (nonstationarity).
• Existing models often suffer from high parameterization, leading to computational challenges in parameter learning and model selection.
• Accurate phylogenetic models are essential for understanding evolutionary history, especially for ancient sequences like rRNA.

Purpose of the Study:

• To develop an efficient Bayesian inference method for nonhomogeneous substitution models, including model order selection.
• To introduce a novel hierarchical prior to improve inference with limited lineage-specific substitution processes.
• To apply these advancements to RNA genes, considering conserved secondary structures, and re-evaluate the Last Universal Common Ancestor (LUCA) conditions.

Main Methods:

• Implemented a reversible jump Markov chain Monte Carlo (MCMC) method for Bayesian inference of model order and phylogenetic parameters.
• Developed specialized nonhomogeneous substitution models for RNA genes incorporating conserved secondary structure.
• Applied an RNA-specific nonhomogeneous model to a structure-based alignment of rRNA sequences from diverse life forms.

Main Results:

• The new method enables efficient Bayesian inference and model selection for complex nonhomogeneous models.
• A hierarchical prior improves results when few lineages share specific substitution processes.
• Analysis of rRNA sequences revealed that the inferred G+C composition at the root of the tree is highly dependent on root placement, challenging previous conclusions about LUCA's optimal growth temperature.

Conclusions:

• The developed Bayesian approach enhances phylogenetic inference with nonhomogeneous models, particularly for RNA genes.
• Rooting the tree with bacterial sequences weakens support for a specific temperature optimum for LUCA, indicating greater uncertainty.
• Discrepancies between analyses using only RNA helices versus all aligned sites highlight the importance of model choices in nonhomogeneous RNA gene phylogenetics.