Detecting hybridization by likelihood calculation of gene tree extra lineages given explicit models

Abstract

1. Explanations for gene tree discordance with respect to a species tree are commonly attributed to deep coalescence (also known as incomplete lineage sorting [ILS]), as well as different evolutionary processes such as hybridization, horizontal gene transfer and gene duplication. Among these, deep coalescence is usually quantified as the number of extra linages and has been studied as the principal source of discordance among gene trees, while the other processes that could contribute to gene tree discordance have not been fully explored. This is an important issue for hybridization because interspecific gene flow is well documented and widespread across many plant and animal groups.<br /><br />2. Here, we propose a new way to detect gene flow when ILS is present that evaluates the likelihood of different models with various levels of gene flow, by comparing the expected gene tree discordance, using the number of extra lineages. This approach consists of proposing a model, simulating a set of gene trees to infer a distribution of expected extra lineages given the model, and calculating a likelihood function by comparing the fit of the real gene trees to the simulated distribution. To count extra lineages, the gene tree is first reconciled within the species tree, and for a given species tree branch the number of gene lineages minus one is counted. We develop a set of r functions to parallelize software to allow simulations, and to compare hypotheses via a likelihood ratio test to evaluate the presence of gene flow when ILS is present, in a fast and simple way.<br /><br />3. Our results show high accuracy under very challenging scenarios of high impact of ILS and low gene flow levels, even using a modest dataset of 5–10 loci and 5–10 individuals per species.<br /><br />4. We present a powerful and fast method to detect hybridization in the presence of ILS. We discuss its advantage with large dataset (such as genomic scale), and also identifies possible issues that should be explored with more complex models in future studies.