Application of the PharmPrint methodology to two protein kinases

The PharmPrint methodology developed by McGregor and Muskal(1,2) was used to construct quantitative structure-activity relationship (QSAR) models for the prediction of cyclin-dependent kinase-2 (CDK2) and vascular endothelial growth factor receptor-2 (VEGFR2) inhibition. The QSAR models were constructed based on a binary description of biological activity-a value of zero for inactive and one for active compounds. Subsets of "active" kinase inhibitors (that is, inhibitors with pIC(50) greater than or equal to 6.0) along with a subset of MDDR3 compounds serving as the recommended set of inactive compounds were used for model development. The predicted activities for the training set compounds were in excellent agreement with the assigned binary activities with greater than 92% of the compounds correctly classified. However, when the QSAR models were applied to the subsets of "inactive" kinase inhibitors (that is, inhibitors with pIC(50) < 6.0), greater than 67% were incorrectly predicted to be active. Identical results were obtained with our CDK2 and VEGFR2 validation sets, where the majority of the inactive kinase inhibitors were predicted to be active. In efforts to improve the predictive performance of the QSAR models, simple, but important modifications were made to the PharmPrint methodology. On the basis of these modifications, a second set of QSAR models was constructed and applied to our validation sets to assess their predictive performance. Significant improvements were seen with the modified version of PharmPrint over the original. The results from both versions of PharmPrint are compared and discussed.