Molecular similarity based on DOCK-generated fingerprints.

An alternative method for defining molecular similarity is presented. By using the docking program DOCK and a reference panel of protein binding sites, fingerprints for a set of molecules have been generated, based on calculated interaction energies. These binding patterns allowed us to calculate matrices of similarity coefficients which subsequently were used for nearest-neighbor searches within the database. Our results indicate that the method is suitable for finding significant similarities of compounds of the same biological activity. Although the overall performance of a traditional 2D similarity method is better in the test systems investigated, our 3D approach can be regarded as complementary since it is able to detect similarities independent of the covalent structure of the compounds. Thus it should be a useful 3D database-searching tool for rational lead discovery.