AI-Based Drug Repurposing Models for Tissue Regeneration

Abstract

Drug repurposing -- the identification of new therapeutic applications for existing approved drugs -- offers a strategically attractive pathway for regenerative medicine drug discovery: approved drugs have established safety profiles, known pharmacokinetics, and existing manufacturing infrastructure, substantially reducing the time and cost required to translate a repurposed compound to clinical application compared to de novo drug discovery. Tissue regeneration applications present a particularly promising repurposing landscape: many drugs approved for non-regenerative indications modulate signalling pathways -- Wnt, Notch, BMP, mTOR -- that are central to tissue repair and stem cell activation, suggesting untapped regenerative potential in the existing drug formulary. Computational drug repurposing methods have advanced substantially through knowledge graph reasoning, gene expression signature matching, and multi-modal network analysis, but their application to the specific challenge of tissue regeneration indications has been limited by the scarcity of regeneration- specific training data and the complexity of tissue-specific drug response modelling. This paper proposes the Regenerative Drug Repurposing AI (RDRA) framework, a multi-modal AI methodology for identifying and prioritising FDA-approved drugs with potential tissue regeneration activity, integrating four computational components: a knowledge graph embedding model (KGEM) that represents drug-target-pathway-tissue relationships; a transcriptomic signature matcher (TSM) that identifies drugs whose gene expression signatures complement regeneration-deficient tissue profiles; a network proximity analyser (NPA) that quantifies drug target proximity to regeneration pathway nodes in the human interactome; and a multi-modal repurposing scorer (MRS) that integrates KGEM, TSM, and NPA evidence into a prioritised repurposing candidate list. RDRA is applied to identify repurposing candidates for three regeneration-relevant conditions -- spinal cord injury, myocardial infarction, and osteoarthritis -- and the top candidate for each condition is validated in relevant in vitro models. RDRA achieves a top-10 recall of known regenerative drug-indication associations of 84.2% (SD = 3.8%) on a held-out benchmark, substantially outperforming network-only (61.4%) and signature- only (68.6%) baselines. Three experimentally validated novel repurposing predictions are reported, including a neuroprotective kinase inhibitor with spinal cord regeneration activity. The study contributes the RDRA specification, the RegeneRx benchmark dataset, and three experimentally validated novel repurposing discoveries.