Identification of molecular features and pharmacophores for selective inhibition of cyclindependent kinases: Application of counterpropagation artificial neural networks

Protein kinases are one of the largest enzyme families, consisting of ۲% of the translated human genome.Kinase inhibitors have enormous medical use for treating mortal diseases such as cancer and Parkinson. Dueto the wide range of activities and functions for all kinases, there is a vital need for the development ofselective kinase inhibitors for targeting only some specific types of them for effective treatments. The mainaim of this project is to find isoform-selective pharmacophores and molecular features for different types ofKinase receptors. In order to achieve this goal, a total of ۴۲۰۱ drug-like molecules with recorded inhibitionactivity for the inhibition of CDK۱, CDK۲, CDK۴, CDK۵, and CDK۹, were collected from Binding-Database. The variable importance in projection (VIP) method was used to select the molecular descriptorsfrom the ۳۲۲۴ descriptor pool calculated via DRAGON ۵.۵ software. The dataset was divided into training(۷۰%) and test (۳۰%) sets, randomly. Counter propagation artificial neural network (CPANN) and supervisedKohonen networks (SKN), were used for the classification of the molecules. Some general parameters suchas mean square distance index, number of Pyrazoles atoms, hydrophobicity, aromaticity index, and numberof hydroxyl groups were found to be important parameters for describing the inhibition behavior of CDK’sinhibitors. Generally, the performances of classification models were evaluated according to the statisticalparameters derived from the confusion matrices. The classification rates range from ۸۲ % to ۷۹% for thetraining and validation procedure for the optimized CPANN models. The high accuracy values of theobtained classifiers for the training and test sets demonstrate that the information provided is reliable fordescribing and predicting the activity of CDK inhibitors. The reliable statistical values of the classifiedmodels can be applied by researchers in the pharmaceutical sciences whom aim to design selective kinaseinhibitors.

کلیدواژه ها:

Classification ، Kinase ، Selective drug design ، structure-activity relationship ، Virtual screening