Machine Learning-Based Prediction of Protein-Ligand Binding Affinity for Kinase Inhibitor Discovery

Background: Accurate prediction of protein-ligand binding affinity is central to structure-based drug design, and protein kinasesremain among the most heavily pursued drug target classes because of their roles in cancer and inflammatory disease. Objective:This narrative review synthesises the published literature on machine learning (ML) and deep learning (DL) approaches to bindingaffinity prediction, with emphasis on their application to kinase inhibitor discovery. Methods: Peer-reviewed and preprint literaturewas identified through PubMed, Google Scholar, Crossref and bioRxiv searches covering feature representations, modelarchitectures, benchmark datasets and validation strategies for protein-ligand affinity prediction. Findings: Structure-basedthree-dimensional convolutional and graph models such as KDEEP, Pafnucy, OnionNet and PIGNet achieve Pearson correlations ofroughly 0.7-0.8 against the CASF-2016 PDBbind core-set benchmark, while sequence-based deep networks such as DeepDTA andGraphDTA report concordance indices near 0.88-0.89 on the kinase-centred Davis and KIBA datasets. Progress is constrained bydataset size, structural redundancy and scaffold bias, motivating scaffold-split and temporal validation schemes andphysics-informed architectures that improve generalisation to novel chemotypes. Conclusions: ML/DL affinity predictors nowapproximate or complement classical scoring functions and free-energy methods for kinase inhibitor triage, but rigorous,leakage-free benchmarking and integration with structural biology resources such as AlphaFold remain priorities for translatingthese models into prospective drug discovery pipelines.