A Bayesian algorithm identifies structural domains in proteins using amino acid sequence information only. This approach differs from other sequence-only approaches, which are typically sequence-homology-based, not fully automated, or dependent on the structure being known. This approach catalogs "pattern" frequencies-occurrences of groups of amino acids-in a nonredundant database of known protein domains to identify those that appear to signal the beginnings and ends of domains. It uses those patterns to score new sequences and find their domain boundaries. Inspecting the patterns that appear significant in marking the fronts or backs of domains reveal subtle differences in amino acid use along each domain's length. These patterns might elucidate differences in function between chemically similar amino acids.

This article is part of a special issue on data mining in bioinformatics.