Study On The Prediction Of The Activity And Selectivity Of Kinase Inhibitors

Protein kinases play a central role in the regulation of cell metabolism,growth,movement,differentiation and division,as well as signaling pathways related to the formation and progression of many human diseases,including cancer,vascular disease,diabetes,inflammation,and degenerative disorders,which makes them attractive target for drug reseach and development.Since the approval of imatinib in 2002,the US Food and Drug Administration has approved 55 small molecule kinase inhibitors,Currently,there are many candidate drugs for small molecule kinase inhibitors,which are currently undergoing clinical trials.Most small molecule kinase inhibitors bind to the highly conserved ATP binding site of the kinase catalytic domain,leading to potential non-targeted kinase interactions(low selectivity),which in turn leads to adverse side effects.Therefore,finding new effective and selective protein kinase inhibitors is still an important challenge in drug development plans.In order to effectively solve the selective problem of kinase inhibitors,the industry and academia have made more and more efforts.A large-scale biochemical screening platform for a variety of protein kinases has been developed to detect the interaction between thousands of small molecule inhibitors and hundreds of human protein kinases.Therefore,it is experimentally feasible to conduct detailed experiments on small molecules of most kinases.Even large pharmaceutical companies need to spend a lot of time and regularly test hundreds of molecules.In the past 20 years,a large number of kinase-related biological activity measurements have been collected and are freely available as open resources,including Ch EMBL,Pub Chem,and Binding DB.Using these resources,computational methods,including structure-based(molecular docking)and ligand-based methods,are widely used to build silicon for predicting kinase inhibitor activity,hit recognition,lead optimization,and offtarget prediction.These include structure-based(molecular docking)and ligand-based methods.It is still an urgent and challenging task to develop selective kinase inhibitors for human diseases.Since experiment-based large-scale kinase profiling of small molecules is still an expensive and time-consuming task,the development of computational methods to discover and selectively predict kinase inhibitors can provide clues for the design of subsequent experimental verification.In this study,we used the naive Bayesian method to construct 882 models based on the atomic center fragments of 147 kinases.The best model for each corresponding kinase target achieved performance.The average area under the receiver operating characteristic curve(AUC)score was 0.907,and the accuracy of the internal validation set was 72.28%.In addition,based on the established prediction model and 5,692,677 compounds that can be screened and purchased,we have developed a comprehensive webbased platform(Kin Pro Pred)to support kinase-inhibitor selectivity prediction and large-scale and proficient virtual screening capabilities for both experts and non-experts in the field.Kin Pro Pred is freely available at http://www.idruglab.com/Kin Pro Pred/.