Scalable Bayesian Divergence Time Estimation With Ratio Transformations.

Area of Science:

• Phylogenetics and Evolutionary Biology
• Computational Biology
• Bioinformatics

Background:

• Estimating divergence times is critical for understanding evolutionary events, from species divergence to viral transmission.
• Large-scale phylogenetic analyses with thousands of sequences present computational challenges due to correlated internal node heights.

Purpose of the Study:

• To develop a scalable computational method for divergence time estimation in phylogenetics.
• To address the computational infeasibility of analyzing large sequence datasets in Bayesian phylogenetic studies.

Main Methods:

• A ratio transformation technique was employed to remap internal node heights into a more computationally tractable parameter space.
• Linear-time algorithms were developed to compute gradient and Jacobian-associated terms for efficient log-likelihood calculations.
• Hamiltonian Monte Carlo sampling was utilized within a Bayesian framework to infer divergence times.

Main Results:

• The ratio transform method improved inference efficiency by at least 5-fold for Lassa virus, rabies virus, and coralline red algae datasets.
• Computational feasibility was achieved for incorporating mixed-effects molecular clock models in Ebola virus divergence time estimation.
• A mixing issue in West Nile virus divergence time estimation was resolved, and clearer multimodal distributions were revealed for certain clades.

Conclusions:

• The proposed ratio transformation and associated algorithms provide a scalable and efficient solution for divergence time estimation in large-scale phylogenetic analyses.
• This method enhances the computational feasibility of complex evolutionary models, leading to more robust inferences of evolutionary history.
• The approach has broad applicability in evolutionary biology, particularly for pathogen evolution and species diversification studies.