Supervised Kohonen networks as tools for virtual screening of PubChem database: A case study with cyclin dependent kinases

The Cyclin-dependent kinase (CDK) is a family of serine/threonine kinases that plays essential roles in regulating the cell cycle, transcription, and cell migration. In addition, they control metabolism and apoptosis . Improper regulation of kinases had been clinically proven to be associated with di↵erent diseases including cancers, and inflammatory and cardiovascular diseases. CDKs are in eleven isoforms with specific biological roles and identifying the characteristics of their selective inhibitors is of great importance. The main aim of this project is to find a series of general structure-selectivity relationship patterns for CDKs inhibitors. To achieve this goal, ۴۲۰۱ active inhibitors of CDK۱, CDK۲, CDK۴, CDK۵, and CDK۹ were collected from Binding DB and analyzed using machine learning techniques. A total of ۳۲۲۴ physicochemical properties were calculated for each molecule using DRAGON ۵.۵ software. As a method for selecting discriminatory molecular features, the variable importance in projection (VIP) approach was used. Counter propagation artificial neural networks (CPANN) and supervised Kohonen networks (SKN) were used for the classification of molecules based on their therapeutic targets. The developed multivariate classifiers were used for ligand-based virtual screening of two million random molecules of the PubChem database. The average values of the enrichment factor (EF۱۰%) for the SKN and CPANN models were ۷.۰۳ and ۴.۷۰, respectively. In addition, the average values of the area under the receiver operating characteristic (ROC) curves were more than ۰.۶۵ and ۰.۸۴ for the CPANN and SKN models, respectively. The VIP-selected molecular descriptors in this work defined a well-separated subspace for discriminating molecules based on the isoforms of CDKs. The information obtained from this study is a stepping stone to help pharmacists and medicinal chemists to produce drugs with better e cacy and fewer side effects

کلیدواژه ها:

Classification ، Cyclin dependent kinase ، PubChem database ، Selective drug design ، Structure-activity relationship (SAR) ، Virtual screening