Abstract
Most docking programs are able to generate conformations similar to crystallographically determined protein/ligand complex structures. However, scoring function have problems at distinguishing conformations from the set of docked poses. The task becomes even more challenging, if the precise induced fit structure is not known. Functional data may provide indirect structural information, and biochemical structure mapping methods have been developed to provide structural information for hard to crystallize biomolecules. In this study we wish to report a novel approach utilizing a statistical evaluation of docking poses with respect to their ability to fit experimental data.The FlexE docking tool was applied to dock nine benzodiazepines into an array of homology models of the benzodiazepine binding site of the α1β2γ2GABAA-receptor. Docking into an array of models with FlexE takes into consideration both receptor uncertainty and side chain flexibility. Subsequently, we screened for ligand orientations which support the maximum common binding mode [1] hypothesis. In an initial coarse step, poses were clustered merely using a RMSD matrix of their common scaffold (csRMSD). Cluster number was set to 30, resulting in an average scaffold-RMSD of about 2Å within a cluster. Only three out of 30 clusters were able to accommodate all ligands at this coarse level. In a second more precise step, for each of the three clusters, receptor atoms forming the binding site for the common scaffold were identified. Finally the scaffold atoms and identified receptor atoms of each pose within a cluster were used to calculate a fused RMSD (fRMSD) matrix. Subsequently in each cluster the fRMSD matrix was used to form homogeneous groups that contain all ligands. These groups of poses were then evaluated in view of homogeneity, energetic scoring values and accordance with biological data. This procedure enabled us to identify one single homogenous group of docking poses which harbors 9 different benzodiazepines.