Fit of Fossils and Mammalian Molecular Trees: Dating Inconsistencies Revisited

Divergence time estimation requires the reconciliation of two major sources of data. These are fossil and/or biogeographic evidence that give estimates of the absolute age of nodes (ancestors) and molecular estimates that give us estimates of the relative ages of nodes in a molecular evolutionary tree. Both forms of data are often best characterized as yielding continuous probability distributions on nodes. Here, the distributions modeling older fossil calibrations within the tree of placental (eutherian) mammals are reconsidered. In particular the Horse/Rhino, Human/Tarsier, Whale/ Hippo, Rabbit/Pika and Rodentia calibrations are reexamined and adjusted. Inferring the relative ages of nodes in a phylogeny also requires the assumption of a model of evolutionary rate change across the tree. Here nine models of evolutionary rate change, are combined with various continuous distributions modeling fossil calibrations. Fit of model is measured both relative to a normalized fit, which assumes that all models fit well in the absence of multiple fossil calibrations, and also by the linearity of their residuals. The normalized fit used attempts to track twice the log likelihood difference from the best expected model. The results suggest there is a very large difference in the age of the root proposed by calibrations in Supraprimates (informally Euarchontoglires) versus Laurasiatheria. Combining both sets of calibrations results in the penalty function vastly increasing in all cases. These issues remain irrespective of the model used or whether the newer calibrations are used.