Computational methods for structure-based and combinatorial drug design

Computational methods, both ligand- and structure-based, have made important contributions to drug design. More recently, combinatorial synthesis and high-throughput screening have begun to impact significantly the drug design process. An emerging drug design paradigm is the partnering of computational methods with combinatorial chemistry to integrate information, such as target structure and pharmacophore preferences, in library design. This approach capitalizes on complementary capabilities of the methods to compensate for the difficulties of each.;The Multiple Copy Simultaneous Search (MCSS) method described in Chapter two is an efficient computational method for determining favorable positions and orientations (defined as local minima in the force field of the target) for functional groups in the binding site of a macromolecule. The resulting functional group maps form the basis of computational and experimental methods for generating compound libraries targeted for a selected macromolecule.;Chapter three describes a method for constructing compound libraries in silico. The method uses Monte Carlo simulated annealing to find near-optimal combinations; of MCSS functional group minima which form molecular skeletons. These skeletons are functionalized by an exhaustive search of side chain functional group minima. The construction of a virtual library for HLA-B27, a Major Histocompatibility Complex (MHC) Class I protein, is presented.;Chapter four illustrates how structure-based design methods can be used to focus combinatorial libraries for a target macromolecule. In this approach, MCSS functional group maps are analyzed by visualization, clustering, interaction energy comparisons, and diversity methods to form a hypothesis regarding ligands for HLA-DR4, an MHC Class II protein. Combinatorial synthesis is used to fill in the details of the hypothesis by generating a library of compounds that satisfy key points of the hypothesis. Parallel screening in vitro allows rapid evaluation of the library.;Chapter five describes an empirically-derived model for protein-ligand binding affinity. A predictive model is generated using a simulated neural network to fit experimentally-determined binding affinities using a small number of structure-based parameters. This model may be useful for screening virtual libraries.;Concluding remarks consider the prospects of an integrated computational and experimental approach to drug design.