Divide-and-Conquer Tree Estimation: Opportunities and Challenges

Large-scale phylogeny estimation is challenging for many reasons, including heterogeneity across the Tree of Life and the difficulty in finding good solutions to NP-hard optimization problems. One of the promising ways for enabling large-scale phylogeny estimation is through divide-and-conquer: a dataset is divided into overlapping subsets, trees are estimated on the subsets, and then the subset trees are merged together into a tree on the full set of taxa. This last step is achieved through the use of a supertree method, which is popular in systematics for use in combining species trees from the scientific literature. Because most supertree methods are heuristics for NP-hard optimization problems, the use of supertree estimation on large datasets is challenging, both in terms of scalability and accuracy. In this chapter, we describe the current state of the art in supertree construction and the use of supertree methods in divide-and-conquer strategies, and we identify directions where future research could lead to improved supertree methods. Finally, we present a new type of divide-and-conquer strategy that bypasses the need for supertree estimation, in which the division into subsets produces disjoint subsets. Overall, this chapter aims to present directions for research that will potentially lead to new methods to scale phylogeny estimation methods to large datasets.