Imputing supertrees and supernetworks from quartets

Abstract

A contemporary and sometimes contentious problem in genome phylogeny is to reconcile the fact that an accurately reconstructed gene tree does not necessarily correspond to a species phylogeny. Thus, in practice, species phylogenies are commonly obtained by applying consensus tree/supertree methods to collections of gene trees. However, such methods can suppress true conflicts in gene trees arising from processes such as gene transfer and gene duplication/loss. To help deal with this dilemma, Holland et al. 2004 proposed constructing consensus networks (Holland and Moulton 2003) instead of consensus trees. This requires that all genes are sequenced for all of the taxa in question, a shortcoming that was circumvented in (Huson et al. 2004) where the alternative Z-closure method for generating supernetworks as opposed to supertrees was introduced. Here, we present a new method to generate supernetworks called Q-imputation [Syst. Bio., to appear]. It works by sequentially inserting all missing taxa into a set of partial gene trees, after which a consensus network is constructed. To insert a missing taxon, a score function is used that rewards inserting the taxon into the partial gene tree in such a way that the resulting tree has as many quartet subtrees as possible in common with the other original gene trees. Theoretical results, simulations, and studies of real data sets indicate that Q-imputation and Z-closure supernetworks have complementary strengths and weaknessess. We therefore expect that Q-imputation will provide a useful additional tool for computing supernetworks. This work has been accepted for publication in Systematic Biology. © Springer-Verlag Berlin Heidelberg 2006.