Shrinkage-based Random Local Clocks with Scalable Inference.

Area of Science:

• Computational Biology
• Phylogenetics
• Evolutionary Biology

Background:

• Molecular clock models are crucial for estimating divergence times in evolutionary biology.
• Existing local clock models face limitations in scalability, model misspecification, and require prior knowledge of clock locations.
• Efficient and accurate molecular clock inference is essential for understanding evolutionary history.

Purpose of the Study:

• To present a new autocorrelated, Bayesian model for heritable clock rate evolution that overcomes limitations of current methods.
• To develop an efficient computational method for scaling molecular clock inference to large phylogenetic trees.
• To apply the new model to infer evolutionary rates in mammalian and influenza virus phylogenies.

Main Methods:

• Developed an autocorrelated Bayesian model with heavy-tailed priors for heritable clock rate evolution.
• Implemented an efficient Hamiltonian Monte Carlo sampler with closed-form gradient computations for scalability.
• Applied the 'shrinkage clock' model to simulated datasets, mammalian phylogenies, and influenza A virus surface glycoproteins.

Main Results:

• The shrinkage clock model demonstrates significant speed-up compared to random local clock methods, especially for large datasets.
• The model successfully recovers known local clock structures in rodent and mammalian phylogenies.
• Enabled computationally intensive analysis of influenza A virus surface glycoprotein evolution without prior clock placement assumptions.

Conclusions:

• The proposed shrinkage clock model offers a scalable and accurate approach to molecular clock inference.
• This method advances the estimation of divergence times and the understanding of evolutionary rate variation.
• The publicly available implementation in BEAST facilitates broader application in evolutionary and phylogenetic studies.