Improving the highly variable performance of virtual docking and screening for recognition of active molecules is a critical issue for the structure based lead discovery and predictive polypharmacology.  We compiled a comprehensive set of induced conformational ensembles for over 2000 protein binding sites, the Pocketome, and used a fraction of those ensembles to study the cross docking performance and recognition of actives vs inactives (or decoys) extracted from the ChEMBL database.  Three main steps improved the performance of the docking poses and scores: 1) correct protonation, orientation and refinement of the pocket side chains; 2) availability of multiple crystallographic conformations; and 3) selecting a small subset, the team, of the complementary pocket conformations with optimal combined performance.  In some cases heterogeneous solvation models can improve performance over the homogeneous continuous solvation models.  The chemical fields of co-crystallized ligands can also be used for docking and scoring with a high level of success.  The approach was tested on the nuclear receptor and GPCR families.  In an application to the estrogen receptor it was shown that the docking models could predict estrogenic activity of mycotoxin zearalenone, chemically dissimilar to most other estrogen mimics, while the ligand based methods fail at this task.

Accurate ligand docking and screening: Lessons from the Pocketome

245th National Meeting (2013)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.