Bayesian classification of protein structural elements

Heuristic Bayesian classification is applied to the problem of defining and enumerating structural motifs present in protein macromolecular systems. The classification system used, Autoclass III, estimates the most probable number of classes and describes class membership in terms of both mean attribute value and expected variance. A database of 9556 protein structure segments, each five amino acids in length, was compiled from high resolution, high quality X-ray crystal structures in the Brookhaven Protein Databank. Coordinates for these segments were transformed to a translationally and rotationally invariant local reference frame, then classified.<>