Complex birth-death models for Bayesian phylodynamic inferences

Abstract

Phylogenetic trees show the evolutionary relationships between individuals, populations or species and are generally built from genetic sequences. Phylodynamic inference focuses on reconstructing the underlying evolutionary processes from a phylogenetic tree, and can infer biologically meaningful parameters such as the rate of transmission of a pathogen or the rate of extinction of certain species. Its applications thus range from tracking the spread of epidemics to evaluating the impact of environmental conditions on the diversification process. Birth-death models are one of the main categories of models used for phylodynamic inference. This thesis presents work realized on two important types of birth-death models, the multi-state model for structured populations and the fossilized birth-death process. Chapter 1 presents an overview of Bayesian phylodynamic inference and its applications as well as birth-death models. In Chapter 2, I introduce a new multi-state birth-death (MSBD) model which can be used to study variations in birth and death rates across a phylogenetic tree. I show that this model can reliably infer these rates on both simulated and empirical datasets. Chapter 3 shows an application of the MSBD model to the detection of transmission clusters in HIV transmission networks, for which I show that it performs better than existing cutpoint-based methods. Chapter 4 presents an R package for simulating fossil and taxonomy datasets, which can be used to test and validate existing or future birth-death models integrating fossils. An application of this package is shown in Chapter 5, where I compare several different methods of handling fossil age uncertainty and evaluate their impact on the accuracy of the estimates. In particular, I show that commonly used methods of simplifying the data by disregarding the age uncertainty lead to strong biases in the resulting inference. In Chapter 6, I present a series of workshops and an online knowledge repository I have contributed to, which are designed to help users of Bayesian phylodynamic inference via the software BEAST2 make the best choices for their own datasets. Indeed, as more complex models are developed, communication between users and developers is increasingly crucial. Finally, in Chapter 7, I discuss the methods developed in this thesis and suggest directions for future research.