Drug-Target Affinity Unleashed: Integrating CNN and Graph Transformers for Superior Predictions

In computational biology and drug discovery, accurate prediction of drug affinity is essential. In this paper, we presents a novel drug target affinity (DTA) prediction model that utilizes multi-modal data representation and advanced neural network architecture to increase prediction accuracy and ranking reliability. Using the Davis and KIBA datasets, which provide extensive information on drug-protein interactions, the model processes chemical and biological properties to predict affinity, which is essential for early drug discovery. Our approach integrates a convolutional neural network (CNN) for protein sequence encoding and a graph variable with an attention mechanism for drug molecule representation. Evaluation metrics including Mean Squared Error (MSE), Concordance Index (CI) and R²m were used to measure model performance. The proposed model outperformed established DTA prediction methods showing lower MSE, higher CI and improved R²m. Additional metrics—precision, accuracy, and recall—were introduced to strengthen the robustness of the model in practical applications, yielding consistent gains across both datasets. Our results validate the model’s potential for accurate alignment of drug-target pairs, thereby increasing its utility in therapeutic prediction and accelerating pathway discovery in the drug development pipeline.