Towards a structural classification of phosphate binding sites in protein-nucleotide complexes: an automated all-against-all structural comparison using geometric matching

Analysis of completed genomes from a number of organisms reveal that about half of all gene products are classified as functionally unknown hypothetical proteins, and structures are already being solved prior to any knowledge of function (1,2). This will make the structure-based prediction of molecular function of increasing importance in the future. The ability to detect and classify local atomic level similarity will allow the development of rules for the prediction of function directly from structure (the premise of Structural Genomics). The determination of protein function from sequence and structure is major goal of bioinformatics techniques. In many cases sequence patterns are sufficient to identify function. The idea of using sequence patterns and profiles underlie the PROSITE(3), PRINTS(4) and Pfam(5) methodologies. However, structure can often provide a more powerful evolutionary link than sequence alone. The structural classification schemes of protein families (e.g. the CATH, SCOP, DALI databases) have proved invaluable tools for structural biologists frequently allowing functional inferences to be made between members of a family even when sequence similarity is statistically low. Furthermore, th