Discovery of new pharmaceutical substances is currently boosted by the possibility of utilization of the Synthetically Accessible Virtual Inventory (SAVI) library, which includes about 283 million synthesizable molecules. The SAVI database is well-suited for ligand-based methods in virtual screening to select molecules for experimental testing. In this study, we compare the performance of three approaches for the analysis of structure-activity relationships that differ in their criteria for selection of “active” and “inactive” compounds included in the training sets.<br/>The PASS (Prediction of Activity Spectra for Substances) software based on the modified Naïve Bayes algorithm was used since it has been shown to be robust and to provide reasonable prediction of many biological activities based on just the structure of a compound even when the information in the training set is incomplete. Traditionally, PASS considers compound as “active” if information about this activity is included in the training set and as “inactive” if such information is not given. However, availability of data about confirmed active and inactive compounds allows comparing different approaches to virtual screening in case studies.<br/>We used different subsets of kinase inhibitors for this study because many data are currently available for this important class of drug-like molecules. Based on the subsets of kinase inhibitors extracted from the ChEMBL 20 database we performed the PASS training, and then applied the model to ChEMBL 23 compounds not yet present in ChEMBL 20 to identify novel kinase inhibitors.<br/>The highest classification accuracy and earliness of actives’ detection were obtained by using an individual training set for each kinase containing only experimental data. Nevertheless, other training strategies can provide acceptable results even in the absence of data about known inactives, which is often the case with novel targets. We assessed the applicability domain of our classifiers: while classifiers trained using individual sets showed strong dependency of the prediction quality on the predicted compounds’ novelty, training strategies assuming the usage of merged sets are much less sensitive to the novelty of predicted compounds.<br/>Taken together these facts allow us to suggest that one may benefit most from using a combination of different training strategies when exploring huge chemical libraries containing diverse structures of heretofore unexplored chemical compounds.

How to achieve better results using ligand-based virtual screening of big chemical databases

256th National Meeting (2018)

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