Computational Identification of High-Affinity Kinase Inhibitors through Molecular Representation Learning and Drug–Target Interaction Modeling

Abstract

The identification of high-affinity kinase inhibitors is a key objective in computational drug discovery because protein kinases regulate numerous biological processes associated with cancer and other complex diseases. This study developed a molecular representation learning framework integrated with drug–target interaction modeling to predict binding affinity between small-molecule inhibitors and kinase proteins. Experimentally validated interaction data from the KIBA and Davis benchmark datasets were used for model development and external validation. Drug molecules represented by canonical SMILES strings and kinase protein sequences were transformed into informative numerical representations to enable supervised affinity prediction. The proposed framework demonstrated satisfactory predictive performance, with superior accuracy achieved on the KIBA dataset while maintaining good generalizability on the independent Davis dataset. High-affinity drug–target interactions were successfully prioritized, indicating that representation-based learning effectively captured the structural patterns governing kinase–inhibitor binding. The findings demonstrate that molecular representation learning offers a reliable and scalable strategy for computational identification of kinase inhibitors while reducing dependence on extensive experimental screening. This framework can support early-stage drug discovery by improving affinity prediction and facilitating the prioritization of promising therapeutic candidates for subsequent experimental validation.