DTBA-net: Drug-Target Binding Affinity prediction using feature selection in hybrid CNN model

In drug discovery, virtual screening and repositioning rely on accurate Drug-Target Binding Affinity (DTBA) prediction to develop effective therapies. However, DTBA prediction remains challenging due to limited annotated datasets, high-dimensional biochemical data, and heterogeneous data sources, including chemical structures, biological sequences, and molecular interactions. These complexities hinder the development of unified deep-learning frameworks. To address these challenges, we propose DTBA-Net, a novel hybrid neural network model that enhances DTBA prediction accuracy and efficiency. DTBA-Net integrates optimal feature selection within a CNN architecture to predict DTBA. Protein sequences and compound structures are processed through a hybrid CNN that includes convolutional layers, a flattened layer, a Modified JAYA Algorithm for optimal feature selection, and dense blocks. The Modified JAYA algorithm selects relevant features, reduces computational overhead, and improves predictive performance. DTBA-Net was evaluated on two benchmark datasets, KIBA and DAVIS. On the DAVIS dataset, DTBA-Net attained an R-squared value of 0.95 and a Mean Absolute Error (MAE) of 0.17. Further validation using the drug Nirmatrelvir resulted in an R-squared value of 0.96, showcasing the model's robustness and scalability. Integrating a hybrid neural network with an optimized feature selection process accelerates model training and enhances prediction accuracy. DTBA-Net demonstrates promising potential for scalable, efficient, and accurate DTBA prediction, facilitating faster and more reliable drug discovery.