Three-dimensional protein-ligand interaction scaling of two-dimensional fingerprints

We introduce a computational scaling methodology that utilizes protein-ligand interaction information extracted from complex crystal structures to enrich similarity searching using structural fingerprints with compound class-specific information. Scaling factors are derived to emphasize fingerprint bit positions that result from interacting fragments of bound ligands and correspond to frequently occurring structural features. Through interaction-based scaling, this information is transferred to standard fingerprints of multiple reference compounds. In systematic search calculations, fingerprints scaled on the basis of three-dimensional information are found to produce higher recall rates of active compounds than alternative types of scaled and non-scaled fingerprints.