Deep Learning for Predicting Drug-Drug Interactions and Patient Safety in Dental Clinical Trials
Annotatsiya
This paper introduces a new model, named Deep Federated Transformer Graph Attention Network (DFT-GAT), to predict drug-drug interaction (DDIs) to improve patient safety in dental clinical trials. The proposed model is based on the PyTorch 2.x and includes multimodal data sources: molecular structures, clinical records, and treatment notes, which are processed with the help of an attention-based fusion mechanism to learn features holistically. With the federated learning paradigm, model training can be done in more than one dental center, and the data privacy and adherence to the clinical rules must be provided. Graph attention module focuses on capturing the inter-drug associations, whereas SHAP-based explainability enables easy understanding of the results of decisions, enhancing clinical trust and interpretability. Experimental analyses showed better performance with accuracy of 93.8% and AUC-ROC of 0.962 and exceeding the current deep learning benchmarks. These findings indicate that the framework can provide scalable, interpretable, and privacy-aware predictions and that DFT-GAT is a powerful AI-based instrument in dental pharmacological analysis to provide safe and efficient predictions.