Random-Effects Substitution Models for Phylogenetics via Scalable Gradient Approximations.

Area of Science:

• Computational Biology
• Evolutionary Biology
• Phylogenetics

Background:

• Phylogenetic and discrete-trait evolutionary inference rely on accurate characterization of evolutionary processes.
• Common continuous-time Markov chain models have limitations in capturing diverse substitution dynamics.

Purpose of the Study:

• To present random-effects substitution models that extend existing models for richer evolutionary process characterization.
• To develop an efficient computational approach for parameter inference in these complex models.

Main Methods:

• Development of random-effects substitution models.
• Proposal of an efficient gradient approximation for likelihood computation.
• Application of Hamiltonian Monte Carlo for Bayesian inference.
• Analysis of SARS-CoV-2, Influenza A virus (H3N2), and tree frog (Hylinae) datasets.

Main Results:

• Random-effects models capture a wider range of substitution dynamics, showing improved model adequacy (e.g., SARS-CoV-2 non-reversibility).
• Efficient gradient computation enables scalable Bayesian inference for large datasets and state-spaces.
• Phylogeographic analysis of Influenza A virus (H3N2) suggests air travel volume predicts dispersal rates.
• State-dependent models found no evidence for arboreality affecting swimming mode in Hylinae.

Conclusions:

• Random-effects substitution models offer a more flexible and accurate framework for evolutionary inference.
• The proposed gradient-based inference method significantly improves computational efficiency.
• These models provide valuable insights into various evolutionary processes, from viral spread to trait evolution.