In the context of molecular modeling and drug discovery research, it is imperative to specify which conformations of a given molecule are responsible for the observed biological activity. The reliable approach to obtain the bioactive conformation is to use the crystal structure of a ligand-protein complex; however, such information is limited because of experimental difficulty, especially for structures of transmembrane receptors. Therefore, we propose a ligand-based approach, named multiple-instance learning (MIL) via embedded instance selection (MILES), to be combined with 1-norm support vector machines (SVM). MILES-SVM enables the identification of biological active conformations and the construction of a quantitative structure-activity relationship (QSAR) model at the same time. We have applied MILES-SVM to study the binding affinity of various compounds interacting with different biomolecular targets including P-glycoprotein, cannabinoid receptors, and protein kinases. MILES-SVM performed comparably to the conventional QSAR approaches in terms of predictive abilities.

Multiple-instance learning (MIL): A framework to identify bioactive conformations

243rd National Meeting (2012)

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