Bayesian Selection of Relaxed-Clock Models: Distinguishing between Independent and Autocorrelated Rates.

Area of Science:

• Evolutionary biology
• Computational phylogenetics
• Molecular evolution

Background:

• Bayesian molecular-clock dating relies on rate models for inferring species divergence times.
• Commonly used independent and autocorrelated rate models can yield divergent time estimates.
• Understanding Bayesian rate model selection is vital, especially with large datasets and potential calibration misspecification.

Purpose of the Study:

• To investigate the statistical behavior of Bayesian rate model selection.
• To explore methods for improving computational efficiency in phylogenomic analyses.
• To assess the impact of calibration misspecification on rate model selection.

Main Methods:

• Computer simulations and real data analysis were employed.
• Marginal likelihoods were calculated using Markov Chain Monte Carlo sampling.
• Approximations of the phylogenetic likelihood using an arcsine branch-length transform were explored.

Main Results:

• Posterior probability for the correct rate model converges to one with more sequence partitions and no calibrations.
• Model selection is robust to slight calibration misspecification but fails with serious errors.
• Arcsine branch-length transform significantly reduces computational cost without compromising accuracy.

Conclusions:

• Bayesian rate model selection is reliable under ideal conditions and slight calibration inaccuracies.
• Serious calibration misspecification leads to erroneous model selection.
• Approximate likelihood methods offer a computationally efficient and accurate alternative for large phylogenomic datasets.