AI-driven drug repurposing: Methods, platforms, and translational applications

The conventional drug discovery pipeline is resource-intensive, time-consuming, and characterized by high attrition rates, often requiring more than a decade and billions of dollars to bring a single therapeutic agent to market. Drug repurposing identifying new therapeutic indications for approved or investigational drugs offers a pragmatic alternative by leveraging established safety profiles and known pharmacokinetics. Despite growing interest, comprehensive syntheses that span the full spectrum of artificial intelligence (AI) methodologies from classical machine learning (ML) through deep learning (DL), graph neural networks (GNNs), and generative AI while also critically evaluating platforms, validation status, and translational readiness remain limited. This review addresses that gap. A narrative review methodology was employed. Literature was searched across PubMed, Scopus, Web of Science, and Google Scholar using a structured keyword strategy (terms: drug repurposing, AI drug discovery, machine learning, deep learning, graph neural networks, generative AI, reinforcement learning, clinical translation combined with Boolean operators). Studies published between 2018 and 2025 were prioritized, with foundational works included regardless of date. Inclusion criteria encompassed primary research articles, systematic reviews, and platform descriptions reporting AI-based drug repurposing methods or outcomes. Non-English publications, conference abstracts without full text, and purely theoretical works without empirical validation were excluded. We identified and synthesized evidence across six thematic domains: (1) classical ML strategies, including target-based and phenotypic screening; (2) DL architectures, including convolutional and recurrent networks; (3) GNNs for biological network modeling; (4) generative AI and reinforcement learning for de novo molecular design; (5) disease-specific applications in oncology, neurology, and COVID-19; and (6) bioinformatics databases underpinning model development. We explicitly compared ML, DL, GNN, and generative AI approaches in terms of data requirements, interpretability, scalability, and translational readiness and summarized these dimensions in a comparative table. Key platforms reviewed RepurposeDrugs, NeuroCADR, and COVID-19 repurposing frameworks were selected using predefined criteria that emphasized methodological transparency, accessibility, and reported validation. Across approaches, AI models frequently demonstrate strong internal predictive performance in silico, with graph-based and transformer models representing the current state of the art on benchmark datasets. However, external validation, reproducibility, and successful clinical implementation remain comparatively scarce. We therefore place particular emphasis on external cohort validation, prospective and experimental follow-up, and the distinction between computational promise and realized clinical benefit. Persistent challenges include dataset bias, limited model interpretability, regulatory uncertainty, and infrastructure gaps in low-resource settings. AI-driven drug repurposing holds considerable translational promise, but many claims remain preliminary. Interdisciplinary collaboration, shared benchmarks, transparent reporting of negative results, and prospective clinical validation frameworks are essential for realizing the full therapeutic potential of these approaches.