NL Journal of Medical and Pharmaceutical Sciences
(ISSN: 3108-0502)

---

A Novel Graph Neural Network for Predicting Polypharmacy Risk

Author(s) : Dr. Anusha Sunder*, Nikhilesh Anand, Samyuktha Sunkara, Kirtikaa Chezhian.
DOI : 10.71168/NMP.02.03.135

---

Abstract

Drug–drug interactions (DDIs) are a major contributor to adverse drug reactions, particularly in patients undergoing polypharmacy for chronic conditions such as diabetes. Although databases such as DrugBank provide validated interactions, their coverage remains incomplete, leaving many potential DDIs unreported. As a result, many computational approaches treat these unknown drug pairs as non-interacting, thereby introducing label noise and reducing prediction reliability. In this study, we propose a Graph Neural Network (GNN)-based framework that integrates heterogeneous biomedical data with Positive–Unlabeled (PU) learning to address these challenges. Known interactions are treated as positive samples, while unknown pairs are handled as unlabeled to identify reliable non-interactions through a data-driven approach. A heterogeneous graph incorporating drug–drug, drug–protein, protein–protein, and pathway relationships is constructed and further enriched with side effects and Gene Ontology annotations. The enriched model demonstrates improved predictive performance and generalization, and external validation on unseen diabetes drug combinations shows strong agreement with clinical literature. Overall, the proposed framework provides a clinically relevant and data-driven approach for DDI prediction, while also enabling extension to polypharmacy risk assessment. Keywords: Graph Neural Networks, Drug–Drug Interaction Prediction, Positive–Unlabeled Learning, Polypharmacy Risk Assessment, Heterogeneous Biomedical Networks.

This article licensed under the Creative Commons Attribution 4.0 International License CC-BY 4.0., which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are properly credited.