Toward a more efficient handling of conformational flexibility in computer-assisted modelling of drug molecules

Summary

Several computational search techniques are described to map the conformation space of flexible organic molecules. A vast multiplicity of geometries is produced that has to be minimized according to a particular energy function. Comparative studies on a nine-membered cyclic lactam are taken as an example. They show that thoroughly tailored search conditions can obtain roughly comparable search efficiencies. Out of the vast multiplicity of geometrically possible and computationally accessible conformers, only a limited number will be of relevance for the problem under consideration. In ligand design for drug discovery, a relative energy ranking determined on isolated conformers is only of limited use for the selection of biologically relevant conformers. This is due to an unsatisfactory transferability of energy scales between different energy functions and the strong modulation of conformational energies of isolated molecules once exposed to a structured molecular environment. A knowledge-based approach, using torsionangle libraries as retrieved for common fragments in small-molecule crystal structures, allows one to map more efficiently the biologically relevant part of conformation space. The relevance of these libraries for the conditions at the binding pocket of a protein is evidenced by experimental data. Sets of well-distributed conformers can be used to compare different drug molecules binding to common targets. Such comparisons reveal new modes of structural superposition of the molecules and consideration of their physicochemical properties leads to interesting pharmacophore hypotheses. They indicate possible binding geometries at the recognition site of a protein and highlight the structural similarities and differences that correlate with changes in the biological properties. Comparisons of GP IIb/IIIa receptor antagonists and of thrombin inhibitors are discussed.