Probabilistic modelling improves relative dating from gene phylogenies

Establishing the timing of past evolutionary events is a fundamental task in the reconstruction of the history of life. State-of-the-art molecular dating methods generally involve the reconstruction of a species tree from conserved, vertically evolving genes, and the assumption of a molecular clock calibrated with the fossil record. Although this approach is extremely useful, its use is limited to speciation events and does not account for genes following different evolutionary paths. Recently, an alternative methodology for the relative dating of evolutionary events has been proposed that considers the distribution of branch lengths across sets of gene trees. Here, we validate this methodology using a fossil-calibrated phylogeny and propose a model-based formalisation using a Bayesian framework. Our analyses revealed that the normalisation of the compared branch lengths with branch lengths of a shared reference clade results in narrower distributions, allowing the correct inference of the relative ordering of evolutionary events. Moreover, we show that distributions of normalised lengths can be modelled using gamma or lognormal distributions. Finally, we demonstrate that inference of the posterior distribution of the mode allows accurate relative age estimation, as assessed by a strong correlation with the molecular clock-dated tree. Overall, we provide a novel, model-based approach to infer relative ages from sets of gene phylogenies.