Bayesian methods for source attribution using HIV deep sequence data

Abstract

The advent of pathogen deep-sequencing technology provides new opportunities for infec- tious disease surveillance, especially for fast-evolving viruses like human immunodeficiency virus (HIV). In particular, multiple reads per host contain detailed information on viral within- host diversity. This information allows the reconstruction of partial directed transmission networks, where estimates of who is source and who is recipient are directly available from the phylogenetic ordering of the viruses of any two individuals. This is a new approach for phylodynamics, and the topic of my thesis. In this thesis, I present updates to the bioinformatics pipeline used by the Phylogenetics And Networks for Generalised Epidemics in Africa consortium for processing HIV deep sequence data and running the phyloscanner program. I then present a semi-parametric Bayesian Poisson model for inferring infectious disease transmission flows and the sources of infection at the population level. The framework is computationally scalable in high- dimensional flow spaces thanks to Hilbert Space Gaussian process approximations, allows for sampling bias adjustments, and estimation of gender- and age-specific transmission flows at a finer resolution than previously possible. In this sense, the methods that I developed enable us to overcome some problems which have been unable to be solved by conventional phylodynamic approaches. We apply the approach to densely sampled, population-based HIV deep-sequence data from Rakai, Uganda. I focus on characterising age-specific transmission dynamics, and examining the sources of HIV infections in adolescent and young women in particular.