W0050

A Multisolution Strategy for the Validation of Ligand Placement in Macromolecular Crystal Structures. John Badger and Sándor Szalma Molecular Simulations Inc., 9685 Scranton Road, San Diego, CA 92121

The focus of many crystallographic studies is the determination of a reliable description of protein:ligand interactions. This information is central to many methods for computer-aided drug design. Low resolution data, high ligand mobility, poor phases and user bias are all hazards to be faced when fitting ligands into electron density maps. Once a ligand is placed within a map the amount of positional change that can be achieved by conventional refinement procedures is limited and map bias towards the ligand model may prevent the detection of a misoriented ligand for manual refitting.

To verify the ligand placement within its binding site, a multicopy sampling and optimization strategy has been developed which can selects all conformations that have low chemical and x-ray energies. In contrast to the qualitative criteria used for the initial ligand fitting (maximum overlap of the ligand with the electron density and avoidance of van der Waals overlaps) these solutions are quantitatively scored against the energy function. An analysis of the best solutions either shows that the ligand was correctly fitted or provides alternative binding modes that had not been previously considered. The optimization takes some account of structural flexibility in the protein and does not require highly accurate co-ordinates.

Tests with the trypsin:benzamidine structure show that a few minutes of CPU time are sufficient to generate, optimize and score a sufficiently complete range of conformation to confirm the correct docking mode. Methods involving variable initial ligand conformations and/or torsion angle dynamics are being investigated to handle more difficult examples, in which the ligand contains many rotatable bonds.