Bayesian phylogenetic estimation of fossil ages

TL;DR

This paper demonstrates that Bayesian phylogenetic models can accurately estimate fossil ages using morphological and molecular data, with high precision and internal consistency across well-characterized fossil and extant species datasets.

Contribution

It introduces a framework combining fossilized birth-death models and morphological evolution to reliably estimate fossil ages and assess model fit using real data.

Findings

High estimation accuracy with median relative error of 5.7% and 13.2%.

Median relative standard error was 9.2% and 7.2%.

Estimated fossil ages closely match geological strata ages.

Abstract

Recent advances have allowed for both morphological fossil evidence and molecular sequences to be integrated into a single combined inference of divergence dates under the rule of Bayesian probability. In particular the fossilized birth-death tree prior and the Lewis-Mk model of discrete morphological evolution allow for the estimation of both divergence times and phylogenetic relationships between fossil and extant taxa. We exploit this statistical framework to investigate the internal consistency of these models by producing phylogenetic estimates of the age of each fossil in turn, within two rich and well-characterized data sets of fossil and extant species (penguins and canids). We find that the estimation accuracy of fossil ages is generally high with credible intervals seldom excluding the true age and median relative error in the two data sets of 5.7% and 13.2% respectively. The median relative standard error (RSD) was 9.2% and 7.2% respectively, suggesting good precision, although with some outliers. In fact in the two data sets we analyze the phylogenetic estimates of fossil age is on average < 2 My from the midpoint age of the geological strata from which it was excavated. The high level of internal consistency found in our analyses suggests that the Bayesian statistical model employed is an adequate fit for both the geological and morphological data, and provides evidence from real data that the framework used can accurately model the evolution of discrete morphological traits coded from fossil and extant taxa. We anticipate that this approach will have diverse applications beyond divergence time dating, including dating fossils that are temporally unconstrained, testing of the "morphological clock", and for uncovering potential model misspecification and/or data errors when controversial phylogenetic hypotheses are obtained based on combined divergence dating analyses.