Home

Background Gene trees carry important information about specific evolutionary patterns which characterize the evolution of the corresponding gene families. However, a reliable species consensus tree cannot be inferred from a multiple sequence alignment of a single gene family or from the concatenation of alignments corresponding to gene families having different evolutionary histories. These evolutionary histories can be quite different due to horizontal transfer events or to ancient duplication events which cause the emergence of paralogs within a genome. Many methods have been proposed to infer a single consensus tree from a collection of gene trees. Still, the application of these tree merging methods can lead to the loss of specific evolutionary patterns which characterize some gene families or some groups of gene families. Thus, the problem of inferring multiple consensus trees from a given set of gene trees becomes very relevant.

Results We present a new fast method for inferring multiple consensus trees from a given set of phylogenetic trees (i.e. additive trees or X-trees) defined on the same set of species (i.e. objects or taxa). The traditional consensus approach yields a single consensus tree. We use the popular k-medoids partitioning algorithm to divide a given set of trees into several clusters of trees. We propose novel versions of the well-known Silhouette and Calinski-Harabasz cluster validity indices, which are adapted for tree clustering with k-medoids. The efficiency of the new method was assessed using both synthetic and real data, such as a well-known phylogenetic dataset including 47 gene trees inferred for 14 archaeal organisms.

Conclusions The presented method allows for inferring multiple consensus trees from a given set of gene trees. It can be used in order to identify groups of gene trees having similar intragroup and different intergroup evolutionary histories. The main advantage of our method is that it is much faster than the existing tree clustering approaches, while providing similar or better clustering results in most cases. This makes it particularly well suited for the analysis of large genomic and phylogenetic datasets.