Inference of Multiple Mergers while Dating a Pathogen Phylogeny.

The vast majority of pathogen phylogenetic studies do not consider the possibility of multiple merger events being present, where a single node of the tree leads to more than 2 descendent branches. These events are however likely to occur when studying a relatively small population or if there is high variability in the reproductive chances. Here we consider the problem of detecting the presence of multiple mergers in the context of dating a phylogeny, that is determining the date of each of the nodes. We use the Lambda-coalescent theory as a modeling framework and show how Bayesian inference can be efficiently performed using a Billera-Holmes-Vogtmann space embedding and a customized Markov Chain Monte Carlo sampling scheme. We applied this new analysis methodology to a large number of simulated data sets to show that it is possible to infer if and when multiple merger events occurred, and that the phylogenetic dating is improved as a result of taking this information into account. We also analyzed real data sets of Vibrio cholerae and Mycobacterium tuberculosis to demonstrate the relevance of our approach to real pathogen evolutionary epidemiology. We have implemented our new methodology in a R package, which is freely available at https://github.com/dhelekal/MMCTime.