Accurate detection of HIV transmission clusters from phylogenetic trees using a multi-state birth-death model

HIV transmission networks are highly clustered, and accurate identification of these clusters is essential for effective targeting of public health interventions. This clustering affects the transmission dynamics of the HIV epidemic, which affects the pathogen phylogenies reconstructed from patient samples. We present a new method for identifying transmission clusters by detecting the changes in transmission rate provoked by the introduction of the epidemic into a new cluster. The method employs a multi-state birth-death (MSBD) model where each state represents a cluster. Transmission rates in each cluster decrease exponentially over time, simulating susceptible depletion in the cluster. This model is fitted to the pathogen phylogeny using a Maximum Likelihood approach. Using simulated datasets we show that the MSBD method is able to reliably infer both the cluster repartition and the transmission parameters from a pathogen phylogeny. In contrast to existing cutpoint-based methods for cluster identification, which are dependent on a parameter set by the user, the MSBD method is consistently reliable. It also performs better on phylogenies containing nested clusters. We present an application of our method to the inference of transmission clusters using sequences obtained from the Swiss HIV Cohort Study. The MSBD method is available as an R package.