MView-DTI: A Multi-view Feature Fusion-Based Approach for Drug-Target Protein Interaction Prediction

Drug-Target protein Interaction (DTI) prediction is a crucial task in the field of drug discovery. Prediction methods based on deep learning have been demonstrated to significantly enhance the accuracy of DTI prediction. Existing approaches mainly extract features from drug molecular sequences and then utilize networks for learning and prediction. However, drug molecular images can clearly display features such as atoms, structures, and chemical bonds, which are difficult to capture in sequences. Therefore, this study introduces a deep learning approach based on multi-view feature fusion, leveraging Transformer to combine the graph structure and image features of drug molecules, thereby learning more comprehensive drug features. This enables the model to learn more intricate interaction features between amino acids and atoms during DTI simulation. The proposed model was evaluated on three benchmark datasets and demonstrated significant improvements over the latest baselines. Furthermore, to validate the efficacy of capturing drug image feature information, ablation experiments were conducted, indicating a notable enhancement in accuracy upon incorporating image data.