Binding prediction of drugs and machine learning analysis of putative compounds

Abstract

Kinases are one of the many enzymes that are absolutely essential for the regulation of cell division and in the performance of so many actions in the cell. They are implicated with many diseases, most notably cancer, which underlines the fact that they might be valuable therapeutic objectives. In the present study, we incorporated machine learning methods alongside molecular docking to determine the binding energies of the candidate compounds for improving the opportunity of drug discovery. The binding interactions between a curated set of 369 ligands, primarily protein kinase inhibitors, and ten selected kinase proteins critical to cancer signaling pathways were investigated. Ligands were meticulously chosen from the PKIDB database based on their relevance to anticancer drug development. Protein structures were obtained from the RCSB Protein Data Bank and prepared for docking using Chimera to ensure biologically relevant conformations. Active binding sites were identified with DogSitescorer, and docking was performed using AutoDock Tools and Vina_split, facilitating systematic analysis of protein-ligand interactions. The binding affinities ranged differently and it was observed that compounds having hydroxyl groups on benzene rings showed higher affinities. However, eribulin had the highest binding affinity of 12.4 kcal/mol for erbB2 while Bemcentinib had the following affinities: 11.9 kcal/mol with PLK1 and 12. 13 kcal/mol with RET amongst the multiple kinases. The aforementioned analysis exposed details about interactions like the hydrogen bonds and the Van der Waals which enabled understanding of the molecular mechanism in binding effectiveness. Specifically, this study focuses on the application of these ligands as potential inhibitors in therapeutic fields along with helpful guidelines for the kinase targeted anti-cancer drug design accompanied by machine learning based compound predicting model.