A Bayesian Approach for Fast and Accurate Gene Tree Reconstruction

Recent sequencing and computing advances have enabled phylogenetic analyses to expand to both entire genomes and large clades, thus requiring more efficient and accurate methods designed specifically for the phylogenomic context. Here we present SPIMAP, an efficient Bayesian method for reconstructing gene trees in the presence of a known species tree. We observe many improvements in reconstruction accuracy, achieved by modeling multiple aspects of evolution, including gene duplication and loss rates, speciation times, and correlated substitution rate variation across both species and loci. We have implemented and applied this method on two clades of fully-sequenced species, 12 Drosophila and 16 fungal genomes as well as simulated phylogenies, and find dramatic improvements in reconstruction accuracy as compared to the most popular existing methods, including those that take the species tree into account. We find that reconstruction inaccuracies of traditional phylogenetic methods overestimate the number of duplication and loss events by as much as 2 to 3 fold, while our method achieves significantly higher accuracy. We feel the results and methods presented here will have many important implications for future investigations of gene evolution.