From matrix factorization to graph neural networks: Advances in computational drug repositioning

Drug repositioning, which explores new uses for existing drugs, offers a cost-effective alternative to traditional discovery. Advances in artificial intelligence (AI) and computational biology accelerate this process by analyzing large-scale biomedical data. This review summarizes AI-driven strategies, ranging from classic matrix factorization to advanced models like graph neural networks and transformers. Applications across oncology, infectious, neurological, autoimmune, and rare diseases are supported by in silico predictions and real-world validations. Key challenges addressed include data heterogeneity, model interpretability, and the crucial need for early absorption, distribution, metabolism, excretion, and toxicity (ADME-Tox) and safety integration to reduce late-stage attrition. Future directions highlight multi-omics integration, foundation models, and generative AI. This article provides insights into leveraging computational approaches for translational drug discovery and underscores the role of AI in therapeutic innovation.