CAST Seminar: Benjamin Bolker

Join us at the CANSSI Ontario STatistics Seminars (CAST) with

Benjamin Bolker

Professor
Departments of Mathematics & Statistics and of Biology
McMaster University

Talk Title

Bayesian Modeling of Trait Evolution in Fishes

Abstract

Biologists often want to understand why certain traits of organisms evolved. One way of translating this question into a statistical context is to set up a continuous-time, discrete-state Markov process along the branches of a phylogenetic tree and estimate how the evolutionary rates (log-hazards of changes in particular states) vary depending on the current values of other traits. Using an example from fish biology I will explain how to estimate the log-likelihood of a set of tip states given a transition matrix; based on this computation we can use standard Bayesian (Markov chain Monte Carlo) techniques to estimate linear contrasts that describe the effects of one biological trait on the evolutionary rate of another trait. I will also (1) rant briefly about some philosophical issues, in particular why single, scalar metrics describing model performance (in this case, p-values, but applicable to other metrics from statistics and machine learning) are less useful than interval measures of uncertainty (2) discuss ongoing technical improvements using automatic differentiation to enable Hamilton Monte Carlo sampling.

Speaker Profile

TBD