Hamiltonian Monte Carlo Sampling for the Skygrid Coalescent-Based Model with Covariates

Phylodynamic inference has emerged as a powerful tool for research in evolutionary biology.Coalescent-based models are widely used in phylodynamics, furnishing prior distributions for phylogenetic trees in Bayesian models and enabling inference of the effective population size, an abstract parameter that characterizes genetic diversity and is of fundamental importance in evolutionary biology. The Skygrid model enables the integration of external covariates into a coalescent-based model and provides a framework to study the relationship between past population dynamics and potential driving factors. However, the Skygrid’s complexity makes posterior approximation challenging, and there is a need for algorithms that can allow it to scale efficiently to large genomic data sets. Here, we evaluate the effectiveness of a promising Markov chain Monte Carlo method, Hamiltonian Monte Carlo, for the Skygrid model with covariates. Through an analysis of three data sets, we show that Hamiltonian Monte Carlo generally outperforms earlier approaches.