Browsing by Author "Nguyen, Quan"

Probabilistic phylogenetic trees are widely considered as the most powerful and reliable method for phylogenetic analysis. However, in reality, processes like hybridization, horizontal gene transfer, and recombination result in reticulation, which means that the evolutionary process can no longer be accurately described by a tree-like graph. A phylogenetic network, which is a general version of a phylogenetic tree is more appropriate in this situation. Unfortunately computational challenges arise when handling likelihood-based phylogenetic networks. Earlier methods often require the hypotheses to be in the neighborhood of the underlying true phylogeny and to be specified as a backbone tree or the number of possible reticulation events. Nevertheless their running time is still often too slow to be really helpful in many realistic scenarios. We propose a method called PhyloDAG, which is significantly faster than earlier methods, and thus restrictions on the network search can be removed. As a consequence the inference is more likely to be accurate. The key idea to speed up phylogenetic network inference by the proposed method, Stochastic Structural Expectation Maximization, which is an EM like algorithm, where in the E step it samples missing data while in the M step it optimizes both the parameters and the structure of the phylogenetic network on pseudo-complete data. Experiments on simulated data as well as real biological and textual data demonstrate that the proposed method, PhyloDAG, can efficiently infer accurate phylogenetic networks.