Species tree estimation using Neighbor Joining

Abstract

Recent theoretical work has demonstrated that Neighbor Joining applied to concatenated DNA sequences is a statistically consistent method of species tree reconstruction. This brief note compares the accuracy of this approach to other popular statistically consistent species tree reconstruction algorithms including ASTRAL-II Neighbor Joining using average gene-tree internode distances (NJst) and SVD-Quartets+PAUP*, as well as concatenation using maximum likelihood (RaxML). We find that the faster Neighbor Joining, applied to concatenated sequences, is among the most effective of these methods for accurate species tree reconstruction.