bayroot: Bayesian sampling of HIV-1 integration dates by root-to-tip regression

Abstract

The composition of the latent HIV-1 reservoir is shaped by when proviruses integrated into host genomes. These integration dates can be estimated by phylogenetic methods like root-to-top (RTT) regression. However, RTT does not accommodate variation in the number of substitutions over time, uncertainty in estimating the molecular clock or the position of the root in the tree. To address these limitations, we implemented a Bayesian extension of RTT as an R package (bayroot), which enables the user to incorporate prior information about the time of infection and start of antiretroviral therapy. Taking an unrooted maximum likelihood tree as input, we use a Metropolis-Hastings algorithm to sample three parameters (the molecular clock, the location of the root, and the time associated with the root) from the posterior distribution. Next, we apply rejection sampling to this posterior sample of model parameters to simulate integration dates for HIV proviral sequences. To validate this method, we use the R package treeswithintrees to simulate time-scaled trees relating samples of actively- and latently-infected T cells from a single host. We find that bayroot yields significantly more accurate estimates of integration dates than conventional RTT under a range of model settings.