This paper introduces the AFBAP model, a novel machine learning model that leverages transfer learning benefits from pre-trained transformers ProtBert and ChemBERTa for feature extraction, and utilises a CNN-based prediction module prefaced by a task-adaptive feature transformation to predict protein-ligand binding affinity with state-of-the-art accuracy. It accepts one-dimensional sequential inputs for both proteins and ligands, in the form of amino acid strings and SMILES strings respectively. AFBAP’s performance over a number of datasets using standard evaluation metrics validates the fact that the model achieves higher accuracy with lower training times and lower compute. AFBAP democratizes access to computational methods of optimizing drug discovery, paving the way for rapid and accessible innovation in drug discovery research.