Calibrated tree priors for relaxed phylogenetics and divergence time estimation

Area of Science:

• Evolutionary biology
• Computational phylogenetics
• Statistical modeling

Background:

• Fossil evidence has historically calibrated divergence time estimation.
• Bayesian Markov chain Monte Carlo (MCMC) is now dominant, using fossils as soft calibrations (prior distributions).
• The statistical properties of these calibrated tree priors in Bayesian settings are not well understood.

Purpose of the Study:

• To clarify the statistical properties of soft calibrations in Bayesian divergence time estimation.
• To investigate whether calibration densities accurately represent marginal prior distributions.
• To propose an alternative method for calibrated Yule priors.

Main Methods:

• Analytical derivations on small phylogenetic trees.
• Comparison of calibration densities with marginal prior distributions.
• Development of an alternative calibrated Yule prior construction.

Main Results:

• Calibration densities (e.g., in BEAST 1.5) do not represent the marginal prior distribution of the calibration node.
• An alternative Yule prior construction allows direct specification of marginal prior distributions for divergence times.
• A practical solution for multiple calibrations remains challenging.

Conclusions:

• Direct estimation of the prior induced by multiple calibration densities is crucial for divergence time dating.
• Current soft calibration methods may require re-evaluation.
• Further research is needed for robust multiple calibration strategies.