A New Method of Computing Ligand-Based Pharmacophore Models for Flexible Chemical Compounds

A detection of the three-dimensional binding mode between a protein and a chemical compound is valuable to optimize drug candidates for high-throughput screening. Pharmacophore models are essential functional groups of atoms in the proper three-dimensional position to interact with a given receptor, and widely used for drug design. Recent pharmacophore models can be classified into two categories, that are receptor-based pharmacophores and ligand-based pharmacophores. For a receptor with a known three-dimensional structure, receptor-based pharmacophores have been studied which are based on the famous concept of a key for the lock [1, 2]. On the other hand, since there are many proteins whose three-dimensional structures have not been known, ligand-based pharmacophore models are still useful [3]. Traditionally, ligand-based pharmacophore models are computed by extracting common features among three-dimensional structures of compounds which are known to interact with a target protein [4]. To do this, possible conformers of compounds should be previously enumerated. However, calculation of common features among the conformers of compounds with many rotatable bonds takes much computational time since such compounds have an extremely large number of conformers. We now focus on ligand-based pharmacophore models and propose a new method to shorten the computational time for generating pharmacophores of a set of chemical compounds with many rotatable bonds. Our method consists of the following processes: First, we divide bonds of each compound into two parts, where one is a set of rotatable bonds and another is a set of non-roratable bonds. Second, we compute hard part sets which mainly includes non-rotatable bonds. We then generate a few conformers for each hard part set. Finally, we superpose the compounds each other by overlapping the three-dimensional structures among the hard parts and extracting common substructures as molecular graphs among the other parts of compounds. By using our method, since we do not have to enumerate all the conformers of a compound, we can save much computational time for ligand-based pharmacophore modeling.