Locally-Adaptive Bayesian Nonparametric Models for Multivariate Processes in Phylodynamics

• Activity Number: 19 - Advances in Statistical Modeling and Inference for Phylodynamics and Molecular Epidemiology
• Type: Topic Contributed
• Date/Time: Sunday, July 30, 2017 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): James Faulkner* and Vladimir Minin
• Companies: University of Washington and University of Washington
• Keywords: coalescent ; shrinkage prior ; smoothing ; effective population size ; Hamiltonian Monte Carlo

Abstract

The goal of phylodynamics is to estimate effective population size trajectories from genetic sequence data. We extend to a multivariate setting a locally-adaptive nonparametric method for phylodynamic inference based on shrinkage priors and Markov random fields. This method is designed to accurately recover population trajectories that exhibit behaviors traditionally difficult to capture, such as abrupt changes or varying levels of smoothness. Our extension allows joint estimation of correlated population processes. We build on recent phylodynamic models that account for preferential sampling by allowing both the population size trajectory and the functional relationship between sampling intensity and population size to co-vary with time via our locally-adaptive smoothing method. We assess model performance with simulated data and show that our method results in reduced bias and increased precision compared to other contemporary methods. We also apply our models to real data from seasonal human influenza outbreaks.