Inferring reticulate evolution networks from consensus gene trees

Motivation: A major unresolved problem in phylogeny is to reconstruct the evolutionary history of a set of species when it cannot be modeled by a simple tree. When a reticulation event, such as hybrid speciation or horizontal transfer, occurs, separate regions of a species’ genetic data may evolve along different trees. Our approach uses the gene trees for each region to solve the inverse problem of reconstructing an underlying network that contains them. This “separate analysis” approach, introduced by Maddison [7], was explored by Nakhleh et al. [10] who give an efficient method for reconstructing a gt-network—a network whose reticulation cycles are nodedisjoint—from a pair of binary trees. A serious problem with this solution is its lack of robustness in the face of errors or unresolved branching in the gene trees. An alternative is to use non-binary consensus trees, created from the top-ranked trees for each genetic region. Consensus trees are more likely to be error-free, but reconstructing networks from them is more difficult. Nakhleh et al. only give a solution in this case for gt-networks with at most one reticulation cycle, which greatly restricts the domain. Our goal is to provide a practical, complete solution to the problem. Results: We give an efficient algorithm for reconstructing a minimal gt-network with any number of reticulation cycles from a set of any number of consensus trees. Availability: Implementation is in progress. Experimental results should be available in the near future. Contact: [email protected]