AUTOMATED PROTEIN-PROTEIN AND SUBSTRATE-PROTEIN DOCKING. Arthur J. Olson, Ph.D., Department of Molecular Biology, The Scripps Research Institute, La Jolla, California

We have developed methods to computationally dock proteins with other proteins and with flexible small molecules and have predicted the correct interactions both in known test cases, and in unknown cases that have subsequently been verified by experiment. Prediction of biomolecular interactions is critical in the understanding of fundamental biological processes as well as in the design of bioactive compounds for medicine, agriculture and other technological applications. Because of the complexity of the systems involved, computational approaches to the docking problem must balance the need for an accurate physical description with the requirements of computational feasibility.

This paper describes the approaches we have taken in the codes AutoDock (1), for docking flexible ligands to protein receptors and SurfDock, for predicting protein-protein interactions. In each case we describe the nature of the models built, the scoring functions developed and the approximations used, as well as the strategies for searching the large configurational space involved. AutoDock uses atomic affinity grids for rapid energy evalutation of intermolecular interactions and has the option of two search strategies; simulated annealing and a local/global search based on a genetic algorithm.

SurfDock uses an analytical surface-based representation of protein shape and properties based upon the expansion of spherical harmonic functions(2). It performs a multi-resolution search of the positional and orientational degrees of freedom using an evolutionary programing algorithm.

Examples involving HIV protease and inhibitors, Human Tissue Factor complexes, and beta lactamase with a protein inhibitor will be discussed.