Advanced Data Science and Computational Intelligence

(Annual (One issue per year))

ABSTRACT

Drug repurposing-the identification of novel therapeutic indications for existing, clinically approved compounds-has emerged as a cost-effective and time-efficient alternative to de novo drug discovery, particularly in the context of rare diseases, emerging pathogens, and treatment-resistant conditions. Conventional repurposing pipelines, however, remain constrained by the combinatorial vastness of the chemical-biological interaction space and the inherent limitations of similarity-based or single-omics inference methods. This paper introduces DeepCure Pro, a novel computational framework that leverages Graph Neural Networks (GNNs) to model and predict drug-disease associations within a heterogeneous biomedical knowledge graph integrating drug-target interactions, protein-protein interaction networks, gene-disease associations, and pathway-level annotations. DeepCure Pro employs a multi-relational graph attention architecture augmented with inductive node embedding strategies, enabling the framework to generalize to previously unseen drug and disease entities. The model is trained and validated using benchmark datasets, including DrugBank, DisGeNET, and STRING, with performance assessed via area under the receiver operating characteristic curve (AUROC), precision-recall metrics, and case-study validation against literature-confirmed repurposing events. Experimental results demonstrate that DeepCure Pro substantially outperforms traditional matrix factorization and network-propagation baselines, achieving superior predictive accuracy while maintaining biological interpretability through attention-weight analysis. This work contributes a scalable, extensible, and interpretable GNN-based architecture to the network medicine literature, offering significant implications for accelerating precision pharmacology and pandemic-response drug discovery pipelines.

Keywords: Graph Neural Networks; Drug Repurposing; GraphSAGE; Explainable Artificial Intelligence; Retrieval-Augmented Generation; Biomedical Knowledge Graphs; Link Prediction; Heterogeneous Graph Learning Intelligence