Abstract: The phylogenetic birth-death model is widely used to study evolutionary processes like speciation and extinction, as well as the spread of pathogens. In this talk, I present some new theoretical and applied results concerning this model. Our first contribution is a scalable variational Bayesian method for inferring birth and death rates from very large quantities of serially-sampled genetic data. Our method produces results that are comparable to, or better than, existing MCMC-based approaches while being several orders of magnitude faster. We study the utility of our method for inferring present and historical epidemiological parameters for the COVID-19 pandemic. Our second contribution is theoretical, and consists of proving that the models we are inferring in the first part of the talk are in fact statistically identifiable. This result contrasts with (though does not contradict) recent nonidentifability theorems for phylogenetic birth-death models that have received significant attention. Finally, I will discuss some complementary hardness-of-estimation results which establish that, even in identifiable model classes, obtaining reliable inferences from finite amounts of data may be extremely challenging.

This is joint work with Brandon Legried and Caleb Ki.
https://jthlab.github.io/