Genome evolution-aware gene trees Method Manuscript

A gene family tree is traditionally inferred from a multiple alignment of homologous sequences according to a model of sequence evolution. Trees for several genes families are thus constructed independently from each other. They often carry unresolutions or bad resolutions. Information for their full resolution may lie in the poorly exploited dependency between gene families, each bringing information for the resolution of the others. We propose to use several kinds of such dependencies in the construction of gene trees: information from a species tree through a model of gene content evolution by duplication, speciation and loss, infor-mation from extant synteny through ortholog predictions, and information from ancestral synteny through a model of gene neighborhood evolution. We develop several “correction ” techniques, yielding a software package called “RefineTree”. We report some tests on simulated data and an application on the full set of gene families from the Ensembl database. We perform a genome-wide analysis of duplication and loss pat-terns on the history of 65 eukaryote species, including ancestral genes and gene orders of all ancestors along this phylogeny. We show that according to several measures including running time, likelihood, stability of genome content and linearity of ancestral chromosomes, trees corrected by RefineTree are arguably more plausible than the ones stored by Ensembl. We discuss the quality criteria in the light of gene definitio