A main objective of quantitative structure-activity relationship (QSAR) is to elucidate the contribution of structural features to an observed biological property and, thus, enable the application of such knowledge in subsequent drug discovery efforts. Essentially, QSAR is a screening data-driven method aiming to extract knowledge for biological data interpretation and, for predictive modeling purposes. In a typical scenario, QSAR models are developed on an assay screening dataset of a specific target of interest and are then used for understanding chemical structure-activity relations or, as filters, i.e. to predict the biological behavior of unknown molecules on the same or a related target. The overwhelming majority of QSAR models have a limited life span. Some exceptions do exist, with 'global' QSAR models that are repeatedly used to predict certain properties, often related to ADMET.Research efforts in the QSAR field have largely focused on the investigation of predictive modeling algorithms, molecular descriptors, similarity measures and training data collection and pre-processing. More recently, efforts in assessing and expanding the applicability domain of models as well as applications for selectivity prediction have been reported in the literature. A potentially useful yet unexplored research direction is that of SAR summarization and the identification of 'persistent' SAR transferable across screening datasets and, potentially, targets of interest.In this presentation we present our initial results in summarizing SAR knowledge from screening datasets and identifying specific SAR with 'cross-assay' potential. We introduce a method for the detailed encoding and storage of SAR using scaffold-based analysis and reasoning. Techniques for SAR knowledge comparison and applicability analysis for potential reuse are also described. Results from the application of the approach to multiple datasets are presented as are potential uses of the method. A discussion on lessons learned, issues to be resolved, and future development directions concludes the presentation.

Scaffold-based reasoning approaches for cross-assay structure-activity relationship identification

246th National Meeting (2013)

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