Automating the identification and analysis of protein beta-barrels.

beta-Barrels are widespread and well-studied features of a great many protein structures. In this paper an unsupervised method for the detection of beta-barrels is developed based on techniques from graph theory. The hydrogen bonded connectivity of beta-sheets is derived using standard pattern recognition techniques and expressed as a graph. Barrels correspond to topological rings in these connectivity graphs and can thus be identified using ring perception algorithms. Following from this, the characteristic topological structure of a barrel can be expressed using a novel form of reduced nomenclature that counts sequence separations between successive members of the ring set. These techniques are tested by applying them to the detection of barrels in a non-redundant subset of the Brookhaven database. Results indicate that topological rings do seem to correspond uniquely to beta-barrels and that the technique, as implemented, finds the majority of barrels present in the dataset.