PartFastTree: Constructing large phylogenetic trees and estimating their reliability

Inferring phylogenies from alignments of thousands of sequences is becoming a known computational problem as DNA sequencing accelerates and gene families are growing rapidly. We present a method, PartFastTree, to construct large phylogenetic trees and estimate their reliability, which is improved from FastTree, an approximate Maximum-Likelihood method for constructing phylogenetic trees. Instead of using improved Neighbor-Joining method, PartFastTree adopts PartTree method in the phase of constructing an initial tree. It reduces the memory required from O(nsa+n^1.25) to O(ns) and at the same time reduces the computation time from O(n^1.25sa) to O(nlog(n)s), where n is the number of sequences, s is the width of the alignment, and a is the size of the alphabet. PartFastTree and FastTree are implemented and the evaluation on them is also presented, while PartFastTree is faster than FastTree with a little reduced accuracy when running on the datasets of from 250 to 237,882 sequences.