QSAR study of Akt/protein kinase B (PKB) inhibitors using support vector machine

A three-class support vector classification (SVC) model with high prediction accuracy for the training, test and overall data sets (95.2%, 88.6% and 93.1%, respectively) was developed based on the molecular descriptors of 148 Akt/protein kinase B (PKB) inhibitors. Then, support vector regression (SVR) method was applied to set up a more accurate model with good correlation coefficient ( r 2) for the training, test and overall data sets (0.882, 0.762 and 0.840, respectively). Enrichment factors (EF) and receiver operating curves (ROC) studies of database screening were also performed either using the SVR model alone or assisted with the SVC model, the results of which demonstrated that the established models could be useful and reliable tools in identifying structurally diverse compounds with Akt inhibitory activity. [Display omitted] Three-class SVC and SVR models with high prediction accuracy for Akt inhibitors were developed. The SVR assisted with SVC model was further proved as a reliable tool in identifying Akt inhibitors in database screening.