Building species trees from larger parts of phylogenomic databases

Gene trees are leaf-labeled trees inferred from molecular sequences. Due to duplication events arising in genome evolution, gene trees usually have multiple copies of some labels, i.e., species. Inferring a species tree from a set of multi-labeled gene trees (MUL trees) is a wellknown problem in computational biology. We propose a novel approach to tackle this problem, mainly to transform a collection of MUL trees into a collection of evolutionary trees, each containing single copies of labels. To that aim, we provide several algorithmic building stones and describe how they fit within a general species tree inference process. First of all, we propose to separately preprocess MUL trees in order to remove their redundant parts with respect to speciation events. For this purpose, we present a tree isomorphism algorithm for MUL trees that can be applied to the pairs of subtrees hanging from duplication nodes. This preprocess lowers the number of duplication nodes in gene trees. For the gene trees that still have duplication nodes, we define the topological information of a MUL tree that can be thought of as being unambiguously related to speciation events. When the MUL tree contains a coherent speciation signal, we show that we can replace the MUL tree with a single-labeled tree representing its speciation information. Otherwise, we propose to extract a maximum subtree that is free of duplication events. Most algorithms have a linear-time complexity, except for an FPT algorithm proposed for a problem that we show to be intractable. The algorithms described in this paper are used to analyse the hogenom database, a database of homologous genes from fully sequenced genomes .