Identification of molecular targets and pathways affected by drugs, environemtal compounds or metabolites is one of great challenges of computational structural biology. A number of methods addressing the problem at the level of chemistry-based machine learning have been developed. We attempted to develop an approach that would be based on a complete collection of crystallographically determined pockets, the Pocketome, complemented with the co-crystallized ligands. The newest release of the Pocketome contains around two thousands ensembles that can be converted to predictive models, and futher characterized by the noise distribution needed to derive the probability values, as well as, in some cases, additional training sets with pKd, IC50 or other activity data. Each new molecule is docked into those pockets or cumulative pharmacophoric fields and scored according to the model. We demonstrate how this system identifies previously unknown targets associated with adverse effects of some drugs and environmental compounds, as well as repurposing drugs to a different indication. A larger evaluation of this platform is underway a group of computational researchers from Novartis.

Toward predictive structural polypharmacology via flexible docking of ligands to the organismal pocketomes

251st National Meeting (2016)

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