Miguel Lacerda will present the Department of Statistical Science's seminar with a talk entitled, "A Population Genetics Approach to Inferring Selection from Longitudinally-Sampled HIV-1 Haplotypes".

Miguel Lacerda is a senior lecturer in the Department of Statistical Sciences.

Abstract: With high-throughput sequencing technologies, we are now able to accurately track changes in the genetic composition of an intra-host viral population over time. Such a time series of haplotype frequencies may be used to infer the strength of natural selection acting on viral variants. To achieve this, we modelled the sampled viral haplotypes as observations from a hidden Markov model, with the hidden states representing unobserved population frequencies that are assumed to have evolved according to a multivariate Wright-Fisher process. For computational reasons, the discrete state space is usually approximated by the standard simplex and the transition density between compositional vectors is obtained in a diffusion limit that assumes negligible selection. This assumption is not appropriate for rapidly evolving pathogens such as HIV and influenza. We therefore derived alternative approximations to the transition density by matching the moments of the Dirichlet and logistic normal distributions to those obtained for the Wright-Fisher process using the delta method without making any assumptions about the strength of selection. To avoid numerical integration over the hidden state vectors, we implemented our model with constant selection in a Bayesian framework and obtained posterior distributions for the selection coefficients using MCMC. We then extended this model to infer time-varying selection by performing a multiple change-point analysis with reversible jump MCMC. We applied our models to infer the selection coefficients of HIV-1 envelope haplotypes that are associated with immune escape.