IF YOU CAN SIMULATE IT, YOU CAN FIT IT: DO-IT-YOURSELF PHYLOGENETIC MODELS OF TRAIT EVOLUTION FOR THE FOSSIL RECORD WITH THE R PACKAGE TREEVO

Modern phylogenetic comparative methods (PCMs) are often used to compare the fit of alternative models of trait evolution, usually limited to the small set of models for which we can calculate likelihoods for. Many of our qualitative explanations of trait evolution for a particular scenario are much more complicated than the simple, mathematically tractable models. Model selection approaches in PCMs can be a red herring, however, often preferring more complex models with outlandish or imperceptibly small parameter values. A better, truly Bayesian approach would be to fit complex, generalized models tailored for the specific evolutionary scenario we are interested in, and then assess parameter estimates for whether they support a significant role for evolutionary processes of interest. Approximate Bayesian Computation (ABC) assesses the fit of models via summary statistics, allowing us to use any model that we can simulate data from. Thus, the only real limits to ABC are the computational cost of repeating those simulations, and whether we our data and summary statistics are sufficient to obtain precise and accurate parameter values. We have implemented an ABC approach in the R package TreEvo, which allows users to write their own functions describing trait change at a given time-step, allowing for models that incorporate competitive exclusion, heterogeneous rates of trait change, multi-optima models of macroevolutionary landscapes, directional trends, and combinations thereof. We will present the application of TreEvo to fossil datasets with dated phylogenies, as well as analyses of simulated datasets to demonstrate the utility and limits of the ABC approach.