Abstract:
The MedPharma AI project addresses the critical issue of polypharmacy related Drug to Drug Interactions (DDIs) where current reactive detection methods often fail to prevent adverse effects. The proposed solution is an AI-driven framework utilizing Graph Neu ral Networks (GNNs), Explainable AI (GNE), and Large Language Models (LLMs) to proactively predict DDIs using the DDInter and DrugBank datasets. The methodology involves representing drugs as nodes in a graph structure for complex interaction analysis with GNE providing textual justifications for transparency. The model’s results show high performance, achieving an overall accuracy of 88% and strong F1-scores across all severity categories, notably 89% for ’Major’ interactions. In conclusion, MedPharma AI offers a robust, scalable, and proactive clinical decision support system designed to reduce adverse drug reactions and significantly improve patient safety
