Identifying Cyclin-dependent Kinase-2 Inhibitors By A Multistage Virtual Screening Method

Objective: Cyclin-dependent kinases(CDKS)are members of the serine/threonine protein kinase family and are able to regulate cellular processes by influencing G/S and G2/M cycles.As a catalytic subunit of CDKS,CDK2 has a strong regulatory effect on the G1-S phase,which is mainly responsible for mitosis and DNA replication.Abnormal expression of CDK2 has been found in many cancers,leading to cell cycle disorders.Many anti-tumor CDK2 inhibitors have been developed,but only a few have entered clinical trials.Therefore,a reasonable virtual screening scheme should be established to screen out CDK2 inhibitors with novel scaffold structure and potent anti-tumor activity.Methods: The multilayer virtual screening method based on ligand and structure combination was established to screen the promising compounds with inhibitory activity against CDK2 protein,and then the leading compounds with the most stable protein binding were screened through molecular dynamics and MM/PBSA free energy calculation.The multi-layer virtual screening method mainly establishes and optimizes three models: support vector machine(SVM),protein ligand fingerprint(PLIF)pharmacophore and molecular docking.Molecular dynamics mainly includes RMSD binding stability analysis and RMSF binding residue analysis.Finally,MM/PBSA binding free energy is calculated by calculating van der Waals energy,electrostatic energy,polar solvation and non-polar solvation energy.Results: 1.The SVM model established by LibSVM is used for parameter and descriptor optimization: when C=32 and ?=0.125 use the optimal descriptor selected by genetic algorithm,the overall prediction accuracy of the training set and test set was 99.79% and 92.72% respectively.2.PLIF method showed the interaction of amino acid residues Glu81 and Leu83 with the generation of most aggregates,and five pharmacophores groups were generated accordingly,of which the production rate and hit rate of pharmacophore 3 on the training set were 89.6% and 12.5%,and the overall accuracy of the test set was 94.32%.Therefore,it was selected to establish a multi-layer virtual screening method.3.The calculation results of NDCG show that the Chemscore scoring function in GOLD can truly reflect the real situation of the combination of CDK2 and ligand,and when the Chemscore scoring threshold is set to 7.24 by the recursive partitioning method,it can have a low false positive rate and a high overall accuracy for the test set.4.The established virtual screening model was tested on the validation set(375 inhibitors,155,528 negative compounds),with a hit rate of 80.1%,enrichment factor of 332.83,yield of 73.8% and final time of 1.62 h,indicating that this multi-layer virtual screening model can screen CDK2 inhibitors in the chemical database.5.This model was used to screen 6 hits compounds from BindingDB,Pubchem and NCI databases.After molecular dynamics and MM/PBSA free energy calculation,2 compounds with strong affinity to CDK2 and stable binding structure with new nucleation structure were finally determined.Conclusion: 1.The multi-layer virtual screening method combines ligand-based and structure-based virtual screening methods,and takes into account and makes use of the information of receptor and ligand,which greatly improves the screening efficiency and accuracy compared with any single method.2.The calculation results of MM/PBSA binding energy and molecular kinetic results of the two lead compounds finally screened were similar to the CDK2 inhibitor Milciclib,indicating that these two compounds had similar binding patterns,stability and affinity with Milciclib.3.The results of this experiment show that the established multi-layer virtual screening method can be used to screen CDK2 inhibitors,and the screened compounds have new nucleotides and high binding stability and affinity to CDK2 protein,so it is of great significance to further develop this nucleotides as CDK2 inhibitors.