Hybrid tree reconstruction methods

Publisher Website
Google Scholar
Abstract
A major computational problem in biology is the reconstruction of evolutionary trees for species sets, and accuracy is measured by comparing the topologies of the reconstructed tree and the model tree. One of the major debates in the field is whether large evolutionary trees can be even approximately accurately reconstructed from biomolecular sequences of realistically bounded lengths (up to about 2000 nucleotides) using standard techniques (polynomial-time distance methods, and heuristics for NP-hard optimization problems). Using both analytical and experimental techniques, we show that on large trees, the two most popular methods in systematic biology, Neighbor-Joining and Maximum Parsimony heuristics, as well as two promising methods introduced by theoretical computer scientists, are all likely to have significant errors in the topology reconstruction of the model tree. We also present a new general technique for combining outputs of different methods (thus producing hybrid methods), and show experimentally how one such hybrid method has better performance than its constituent parts.

This publication has 14 references indexed in Scilit: