B0891
Title: Bayesian inference of mixed Gaussian phylogenetic models
Authors: Bayu Brahmantio - Linköping University (Sweden) [presenting]
Abstract: The evolution of continuous traits is often modelled using stochastic differential equations that combine deterministic change of a trait through time with noise that represents different unobservable evolutionary pressures. Two of the most popular choices are Brownian motion and Ornstein-Uhlenbeck processes, which belong to the GLInv family of models, i.e., models with a Gaussian transition probability whose expectation is linear with respect to ancestral value and variance is invariant with respect to it. Using this framework, it is possible to set different GLInv models into different parts of a phylogenetic tree to do parameter inferences and model comparisons. A Bayesian scheme is implemented as an extension to the maximum likelihood framework to include uncertainties in the parameter estimate and prior knowledge that are more biologically relevant. The method is written as an R package that applies Monte Carlo inference to retrieve posterior quantities. The package also features custom user-defined priors and Bayesian model selection.