Bayesian time-tree priors
Seminar Room 1, Newton Institute
Bayesian phylogenetic inference requires specification of a prior distribution on the space of trees. In BEAST, the phylogenetic history is explicitly separated into a time-tree (by definition ultrametric for contemporaneous taxa) and the rates of evolution, which may vary from lineage to lineage. The combination can account for phylogenetic trees with non-ultrametric branches in units of substitutions per site. Hierarchical Bayesian priors on the time-tree lead to inference of population and dynamical priors through application of the coalescent, Yule or Birth-death tree priors, and calibration of the individual nodes in the time-tree leads to estimation of the absolute age of other divergence times of interest. Here I discuss difficulties and recent advances in the development of calibrated time-tree priors for Bayesian phylogenetic inference, including calculation of the marginal distribution of divergence times under various time-tree priors and a class of new time-tree priors based on a "Birth-death skyline" model of branching.