A general comparison of relaxed molecular clock models

Area of Science:

• Evolutionary Biology
• Computational Biology
• Phylogenetics

Background:

• Estimating divergence times using molecular data relies on relaxed clock models.
• The optimal model choice, considering rate autocorrelation and prior distributions, remains unclear for molecular dating.

Purpose of the Study:

• To benchmark and compare various relaxed clock models and prior choices for molecular dating.
• To provide guidelines for selecting appropriate models for divergence time estimation.

Main Methods:

• Reimplementation of existing relaxed clock models within a Bayesian framework.
• Introduction of a new autocorrelated model, the CIR process.
• Assessment of model and prior fit using Bayes factors and thermodynamic integration on protein datasets.

Main Results:

• Autocorrelated models (CIR and lognormal) showed superior performance over uncorrelated models across all tested datasets.
• Model fit was consistently better with autocorrelated models for eukaryotes, vertebrates, and mammals.
• The choice of prior distribution for divergence times was dataset-dependent.

Conclusions:

• Autocorrelated relaxed clock models are recommended for molecular dating.
• Guidelines are provided for selecting models and priors, improving the accuracy of divergence time estimates.
• This work facilitates more comprehensive model comparisons in phylogenetics.