Many biological processes involve the binding of a small molecule by a receptor protein, for example the binding of a substrate by an enzyme, or the binding of an antigen by an antibody. In computer aided drug design, a database of molecules is screened against a protein binding site to identify those with the greatest binding affinity. The process involves two steps: 1) �Docking� , in which potential ligands are bound to a receptor and 2) �Scoring�, in which the receptor/ligand complexes are ranked by a scoring function. A molecule which has been shown by experiment to bind to a receptor may be mis-ranked by a scoring function if it is not properly positioned within the binding site. Thus, docking involves not only sampling different ligands, but also sampling different orientations, or �poses� of the same ligand within the binding site.

The Ligand-Protein Database (LPDB) includes a set of incorrectly posed ligands, termed �decoys�, for a large number of protein-ligand complexes. Using this database, we tested the ability of a force field based scoring function to discriminate the native pose among a set of decoys. Our study involved 40 different receptor/ligand complexes, with 50-70 decoys for each. For each complex, we minimized each decoy in the binding site and calculated energies. For 35 out of the 40 complexes (87.5%), the lowest energy complex, ie: the highest ranking decoy, was below 2 Angstrom in root mean square deviation (RMSD) from the experimental (crystal) structure. This demonstrates that in most cases, our scoring function successfully identified the native pose from a set of decoy structures. Future work will involve scoring a larger set of complexes, identifying the weaknesses in the scoring function which lead to high scoring decoys, and improving aspects of the scoring function.