A-PRIORI AND A-POSTERIORI MIXTURE DISTRIBUTIONS FOR USING DATABASES IN PROTEIN-STRUCTURE DETERMINATION

The twists and turns of protein molecules correspond to the rotational angles of the main and side chains in their constituent amino acid re sidues. Owing to stereochemical restrictions the joint distribution of these angles falls into several well-separated clusters. Consequently , we fit a statistical finite mixture model to data from the Brookhave n Protein Data Bank (PDB), obtaining a new classification of conformat ional states of amino acids. The results of this database modeling are important for knowledge-based approaches to protein structure determi nation and analysis and can be used in conjunction with experimental d ata in the determination of protein spatial structure. As parr of this process, the mixture distributions we have fit can be used as a prior i distributions. A posteriori distributions corresponding to these a p riori mixture distributions are illustrated. (C) 1997 Elsevier Science B.V.