Abstrakt
Using pharmacophores in virtual screening of large chemical compound libraries proved to be a valuable concept in computer-aided drug design. Traditionally, pharmacophore-based screening is performed in 3D space where crystallized or predicted structures of ligands are superposed and where pharmacophore features are identified and compiled into a 3D pharmacophore model.
However, in many cases the structures of the ligands are not known which results in using a 2D pharmacophore model. We introduce a method capable of automatic generation of 2D pharmacophore models given previous knowledge about the biological target of interest.
The knowledge comprises of a set of known active and inactive molecules with respect to the target. From the set of active and inactive molecules 2D pharmacophore features are extracted using pharmacophore fingerprints.
Then a statistical procedure is applied to identify features separating the active from the inactive molecules and these features are then used to build a pharmacophore model. Finally, a similarity measure utilizing the model is applied for virtual screening.
The method was tested on multiple state of the art datasets and compared to several virtual screening methods. Our approach seems to exceed the existing methods in most cases.
We believe that the presented methodology forms a valuable addition to the set of tools available for the early stage drug discovery process.