Using INC Within Divide-and-Conquer Phylogeny Estimation

Thien Le, Aaron Sy, Erin K. Molloy, Qiuyi Richard Zhang, Satish Rao, Tandy J. Warnow

Abstract: In a recent paper (Zhang, Rao, and Warnow, Algorithms for Molecular Biology 2019), the INC (incremental tree building) algorithm was presented and proven to be absolute fast converging under standard sequence evolution models. A variant of INC which allows a set of disjoint constraint trees to be provided and then uses INC to merge the constraint trees was also presented (i.e., Constrained INC). We report on a study evaluating INC on a range of simulated datasets, and show that it has very poor accuracy in comparison to standard methods. We also explore the design space for divide-and-conquer strategies for phylogeny estimation that use Constrained INC, and show modifications that provide improved accuracy. In particular, we present INC-ML, a divide-and-conquer approach to maximum likelihood (ML) estimation that comes close to the leading ML heuristics in terms of accuracy, and is more accurate than the current best distance-based methods.