The genome revolution has produced an abundance of protein sequence data. Traditional homology-based computer methods make it possible to establish evolutionary relationships between large numbers of these proteins. Yet among any set of new protein sequences, say from the complete genome sequence of a new organism, a significant fraction of the proteins cannot be assigned fuctions by traditional methods. A new sequence may have no recognizable homologs in other organisms, or it may have recognizable homologs, the cellular functions of which are yet unknown. The critical need to make some kind of functional inferences for the vast numbers of proteins that could not be functionally annoted by traditional homology methods led in 1999, and in the years that followed, to new ideas for inferring ‘functional linkages’ between proteins not related to each other by homology. These ‘non-homology’ or ‘genomic context’ methods included the ‘Phylogenetic Profile’ method and the ‘Rosetta-Stone’ method (both pioneered principally by Edward Marcotte and Matteo Pellegrini when they were postdocs with Eisenberg and Yeates), and others. Subsequent work has aimed to extend those ideas. One recent extension of Phylogenetic Profiles (developed by Peter Bowers and Shawn Cokus) involves an application of logic analysis to uncover proteins whose presence vs. absence across organisms is related to the presence or absence of two other proteins, taken in logical combination. These kinds of higher order relationships are expected to be abundant in the cell, but are not detected by the original Phylogenetic Profile method, which looks for direct similarity between the profiles of just two proteins at a time.
Our computational genomics work has touched on many other subjects as well: disulfide bonding in thermophiles, repetitive protein sequences, genomic encoding of unusual amino acids such as selenocysteine and pyrollysine, detection of protein targeting sequences, and the function of bacterial microcompartments.
