Pharmacophore models aim to comprise the features of protein-ligand interactions that are most crucial for binding and for the subsequent biological activity. These models are used for virtual screening (VS) to identify potential new actives or to generate ligand alignments for subsequent three-dimensional quantitative-structure-activity-relationships (QSAR) calculations. Pharmacophore models are typically derived from structural features common to biologically active ligands that are hypothesized to be important for biological activity.
We have devised a method, PharmPose, to generate protein-based pharmacophore models that in contrast to those standard ligand-based pharmacophore models, does not require a priori knowledge of active ligands and whose models are not biased by the chemical space of previously identified actives. We carefully optimized the pharmacophore-generation process by reproducing native contacts for a large number of experimentally-determined protein-ligand complexes. The method has been successfully applied to VS and binding pose (i.e. ligand conformation in binding site) identification.