High performance computational virtual screening tools: development and application to the discovery of kinase inhibitors

Abstract

Virtual screening (VS) can provide valuable contributions in hit and lead compound discovery. Numerous software tools have been developed for this purpose. However, the insufficient coverage of compound diversity, high false positive, high false negative prediction and lower speed of screening compound libraries are also required to address in the development of virtual screening methods. In this work, training-sets of diverse inactive compounds are used to improve the performance of Support vector machine (SVM) virtual screening tools. In retrospective database screening of active compounds of single mechanism (HIV protease inhibitors, DHFR inhibitors, dopamine antagonists) and multiple mechanisms (CNS active agents) from large libraries of 2.986 million compounds, the yields, hit-rates, and enrichment factors of our SVM models are compared to those of structure-based VS and other ligand-based VS tools in screening libraries of &lt 1 million compounds. The hit-rates are comparable and the enrichment factors are substantially better than the best results of other VS tools. SVM appears to be potentially useful for facilitating lead discovery in VS of large compound libraries.

Virtual screening performance of SVM depends on the diversity of training active and inactive compounds. We also evaluated the performance of SVM trained by sparsely distributed actives in six MDDR biological target classes composed of high number of known actives of high, intermediate, and low structural diversity. The results show SVM has substantial capability in identifying novel active compounds from sparse active datasets at low false-hit rates.

c-Src and VEGFR-2 are two important kinases that play various roles in tumour progression, invasion, metastasis, angiogenesis and survival. The successes of their inhibitors and the encountered problems have led to further efforts for discovering new inhibitors for c-Src and VEGFR-2. We applied our developed SVM based virtual screening tools for searching c-Src and VEGFR-2 inhibitors from large compound libraries. SVM models showed around 60% accuracy for independent testing sets and >99.9% accuracy for non-inhibitors (very low false hit-rate) that is favorable for selecting potential leads to further study in wet-lab experiment.

Multi-target agents have been increasingly explored for enhancing therapeutic efficacies and improving safety and resistance profiles by selectively modulating the elements of these counter-target and toxicity activities. In the final part of my thesis, combinatorial support vector machines (C-SVMs), virtual screening tools for searching multi-target agents are developed based on our previous high performance SVM based virtual screening tools. C-SVMs models were tested for searching dual-inhibitors of 11 combinations of 9 anticancer kinase targets (EGFR, VEGFR, PDGFR, Src, FGFR, Lck, CDK1, CDK2, GSK3). Moreover, C-SVMs were compared to other VS methods DOCK Blaster, kNN and PNN against the same sets of kinase inhibitors and 1.02M Zinc clean-leads dataset. C-SVMs produced comparable dual-inhibitor yields, slightly better false-hit rates for kinase inhibitors, and significantly lower false-hit rates for the Zinc clean-leads dataset.