PubMed1998
Molecular surface areas of proteins are responsible for selective binding of ligands and protein-protein recognition, and are considered the basis for specific interactions between different parts of a protein. This basic principle leads us to study the interfaces within proteins as a learning set for intermolecular recognition processes of ligands like substrates, coenzymes, etc., and for prediction of contacts occurring during protein folding and association. For this purpose, we defined interfaces as pairs of matching molecular surface patches between neighboring secondary structural elements. All such interfaces from known protein structures were collected in a comprehensive data bank of interfaces in proteins (DIP). The up-to-date DIP contains interface files for 351 selected Brookhaven Protein Data Bank entries with a total of about 160,000 surface elements formed by 12,475 secondary structures. For special purposes, the inclusion of additional structures or selection of subgroups of proteins can be performed in an easy and straightforward manner. Atomic coordinates of the constituents of molecular surface patches are directly accessible as well as the corresponding contact distances from given atoms to their neighboring secondary structural elements. As a rule, independent of the type of secondary structure, the molecular surface patches of the secondary structural elements can be described as quite flat bodies with a length to width to depth ratio of about 3:2:1 for patches consisting of more than ten atoms. The relative orientation between two docking patches is strongly restricted, due to the narrow distribution of the distances between their centers of mass and of the angles between their normal lines, respectively. The existing retrieval system for the DIP allows selection (out of the set of molecular patches) according to different criteria, such as geometric features, atomic composition, type of secondary structure, contacts, etc. A fast, sequence-independent 3-D superposition procedure was developed for automatic searches for geometrically similar surface areas. Using this procedure, we found a large number of structurally similar interfaces of up to 30 atoms in completely unrelated protein structures.