Explainable Deep Learning for Predicting Drug–Target Interactions in Personalized Medicine

Drug–target interaction (DTI) prediction is crucial for accelerating drug discovery and enabling personalized medicine. While deep learning models have shown remarkable performance in DTI prediction, their black-box nature limits clinical adoption. This paper presents an explainable deep learning framework that combines graph convolutional networks (GCNs) for drug representation and convolutional neural networks (CNNs) for protein target encoding, integrated with attention mechanisms, SHAP values, and Grad-CAM for interpretability. We evaluated our approach on the BindingDB and Davis datasets, achieving an AUROC of 0.96 and AUPRC of 0.94, outperforming existing methods by 3-5%. The explainability module provides molecular substructure importance and binding site identification, enabling clinicians to understand model predictions. Our framework demonstrates that high performance and interpretability are not mutually exclusive in DTI prediction for personalized medicine applications.

Hali tarjima qilinmagan