AI Drug Repurposing — Giving Old Drugs New Life

Why Drug Repurposing Matters

Of the roughly 10,000 known drug molecules with established safety data, most have been tested for only a handful of indications. Each molecule interacts with hundreds of biological targets, creating a vast unexplored space of potential therapeutic applications. AI can systematically explore this space in ways manual research never could.

The COVID-19 pandemic demonstrated the power of repurposing. Dexamethasone, a decades-old steroid, was identified as an effective COVID treatment through AI-assisted analysis of clinical data — saving hundreds of thousands of lives faster than any new drug could have been developed.

For rare diseases, repurposing is often the only economically viable path to treatment. The patient populations are too small to justify $2 billion in new drug development, but a repurposed drug with an established safety profile can reach patients at a fraction of the cost and timeline.

AI Approaches to Drug Repurposing

Network pharmacology uses AI to map the complex web of interactions between drugs, proteins, genes, and diseases. Graph neural networks analyze these biological networks to identify drugs that could modulate disease pathways they were never designed to target. This approach has identified promising candidates for Alzheimer's, cancer, and autoimmune diseases.

Molecular docking simulations powered by AI predict how drug molecules bind to disease-related protein targets. Deep learning models like AlphaFold have revolutionized protein structure prediction, enabling virtual screening of existing drugs against newly characterized disease targets at unprecedented speed and accuracy.

Natural language processing mines millions of published research papers, clinical trial reports, and electronic health records to discover hidden connections between drugs and diseases. AI identifies patterns that no human researcher could find by reading papers sequentially — connecting findings across disciplines and decades of research.

Real-World Electronic Health Records

One of the most powerful AI repurposing strategies analyzes real-world patient data. When millions of patients take a drug for condition A, some of them also have condition B. AI can detect whether patients on the drug experience unexpected improvements in condition B at rates significantly above baseline.

This retrospective analysis requires sophisticated statistical methods to account for confounders — patient demographics, co-medications, disease severity, and selection bias. AI models trained on large-scale EHR databases can control for these factors and identify genuine therapeutic signals hidden in the noise of real-world clinical practice.

Notable discoveries from EHR mining include metformin's potential anti-cancer properties, observed because diabetic patients on metformin showed lower cancer incidence than expected. These signals then guide targeted clinical trials that validate or refute the AI-generated hypothesis.

Clinical Trial Acceleration

Because repurposed drugs already have extensive safety data, they can skip Phase I trials entirely and often proceed through abbreviated Phase II studies. AI further accelerates this process by identifying optimal patient populations, predicting likely response rates, and designing adaptive trial protocols that reach conclusions faster.

AI-powered patient matching identifies individuals most likely to benefit from a repurposed drug based on their genetic profile, biomarkers, and disease characteristics. This precision enrollment improves trial success rates from the industry average of 10% to 25-35% for AI-guided repurposing candidates.

Success Stories and the Pipeline

Thalidomide, once infamous for birth defects, was repurposed as a groundbreaking treatment for multiple myeloma after AI analysis revealed its anti-angiogenic properties. Sildenafil was developed for hypertension before becoming Viagra. These serendipitous discoveries are being replaced by systematic AI-driven identification.

As of 2026, over 200 AI-identified drug repurposing candidates are in clinical trials across oncology, neurodegenerative diseases, infectious diseases, and rare genetic conditions. The success rate for AI-identified candidates is roughly 2.5x higher than traditional repurposing approaches, validating the technology's value.

Challenges and Limitations

Patent and regulatory challenges complicate the economics. Many drugs targeted for repurposing are off-patent, meaning generic competition makes it difficult to recoup development costs for new indications. Orphan drug designations and extended exclusivity periods help, but the business model remains challenging.

Data quality and access remain barriers. EHR data is fragmented across health systems, inconsistently coded, and subject to privacy restrictions. Federated learning approaches — where AI models train on distributed data without centralizing it — are emerging as a solution that preserves privacy while enabling large-scale analysis.

Key Takeaways

• Drug repurposing cuts development costs from $2.6B to $300M and time from 15 to 3-5 years
• AI network pharmacology maps drug-target-disease relationships at scale
• EHR mining reveals therapeutic signals hidden in real-world patient data
• AI-guided candidates show 2.5x higher clinical trial success rates
• 200+ AI-identified candidates are currently in clinical trials worldwide