Finding protein domain boundaries: an automated, non-homology-based method

A sequence-based methodology identifies the boundaries of structural domains in proteins. The method doesn't depend on knowledge of a protein's structure or on sequence homologs. We developed a Bayesian approach based on the statistical analysis of word content used in other fields. Our method first catalogs "pattern" frequencies - occurrences of groups of amino acids - in a nonredundant database of known protein domains and then uses the distributions of these patterns to identify regions of protein sequence that appear to signal the beginnings and ends of domains. The domain-delineating signals we've produced using amino acid patterns show great promise in providing further insight into both the biochemistry and structural biology of proteins.