Automatic selection of representative proteins for bacterial phylogeny

BACKGROUND: Although there are now about 200 complete bacterial genomes in GenBank, deep bacterial phylogeny remains a difficult problem, due to confounding horizontal gene transfers and other phylogenetic "noise". Previous methods have relied primarily upon biological intuition or manual curation for choosing genomic sequences unlikely to be horizontally transferred, and have given inconsistent phylogenies with poor bootstrap confidence. RESULTS: We describe an algorithm that automatically picks "representative" protein families from entire genomes for use as phylogenetic characters. A representative protein family is one that, taken alone, gives an organismal distance matrix in good agreement with a distance matrix computed from all sufficiently conserved proteins. We then use maximum-likelihood methods to compute phylogenetic trees from a concatenation of representative sequences. We validate the use of representative proteins on a number of small phylogenetic questions with accepted answers. We then use our methodology to compute a robust and well-resolved phylogenetic tree for a diverse set of sequenced bacteria. The tree agrees closely with a recently published tree computed using manually curated proteins, and supports two proposed high-level clades: one containing Actinobacteria, Deinococcus, and Cyanobacteria ("Terrabacteria"), and another containing Planctomycetes and Chlamydiales. CONCLUSION: Representative proteins provide an effective solution to the problem of selecting phylogenetic characters.