AI Drug Repurposing: How Machine Learning Is Finding New Uses for Existing Medicines

Thousands of tested compounds are sitting in pharmaceutical freezers. AI is working out which ones could treat diseases they were never designed for.

Nine out of ten experimental drugs fail to reach their original therapeutic target. That’s the reality of pharmaceutical R&D: a decade of work and billions in investment, with a 90% attrition rate.

But here’s what makes AI drug repurposing so compelling. Those failed experiments didn’t produce nothing. They produced safety data, pharmacokinetic profiles, manufacturing protocols, and years of clinical observations. Pharmaceutical companies have thousands of already-tested compounds sitting in their libraries. Many passed initial safety screening. Some were approved for one condition before new opportunities emerged. Others were shelved when funding dried up.

AI systems are now mining these molecular libraries at a scale and speed that human researchers cannot match, cross-referencing genomics data, digital pathology, population health records, and molecular databases to find connections between existing drugs and diseases they were never designed to treat.

The global drug repurposing market was estimated at $26–30 billion in 2024 and is projected to grow at a 6–8% CAGR through 2030. AI-driven discovery platforms are becoming an increasingly significant segment of that market.

This article breaks down how the technology works, who’s leading the field, what results are reaching patients, and what obstacles remain.

What Is AI Drug Repurposing and Why Does It Matter?

Drug repurposing (also called drug repositioning or reprofiling) is the process of finding new therapeutic uses for medications that already have established safety profiles. It has been around for decades: sildenafil was developed for cardiovascular disease before becoming Viagra; thalidomide was repurposed for multiple myeloma after its withdrawal as a morning sickness treatment; minoxidil moved from blood pressure medication to hair loss therapy.

What makes AI drug repurposing different is the scope and speed of analysis. Traditional repurposing relied on serendipity or narrow hypothesis-driven research. AI systems can screen millions of drug-disease pairs simultaneously, identify non-obvious molecular relationships, and rank the most promising candidates for validation before any wet-lab work begins.

The economics are hard to ignore. Traditional drug development takes 10–15 years and costs upwards of $2.6 billion per approved drug. Drug repurposing can shave three to four years off that timeline and cut R&D costs by up to 60%. For investors, that translates to lower capital requirements, reduced regulatory risk, and faster paths to market.

Why AI Drug Repurposing Is Accelerating Now: Data, Regulation, and Economics

Several forces are converging to make AI drug repurposing practical at scale.

The data infrastructure finally exists. The UK Biobank released a multimodal dataset covering 20,000 subjects with integrated imaging and genomics data. FinnGen has compiled genome data from over 550,000 Finnish participants. The NIH’s All of Us programme is building one of the largest and most diverse biomedical datasets in the world. These resources give AI systems the biological signal they need to map drug-disease relationships across populations.

Regulators are moving toward real-world evidence. The FDA has been developing frameworks that support data-driven repositioning submissions. Major funders including the NIH and Wellcome Trust increasingly encourage applicants to consider existing compounds and resource-efficient strategies before pursuing entirely new chemical entities.

Investment is flowing in. Pharma companies and biotech investors see drug repurposing as a capital-efficient strategy. For VCs and PE firms, AI-driven repurposing offers lower capital requirements, reduced regulatory risk, and established safety data. The result is an accelerating pipeline of funded programmes.

How AI Discovers New Drug-Disease Connections

Modern AI drug repurposing pipelines go well beyond simple pattern matching. They use layered computational approaches to uncover relationships that human researchers would take years to identify.

Graph neural networks integrate genes, biological pathways, patient phenotypes, and molecular structures into interconnected knowledge maps. These networks surface under-appreciated links. For example, they might reveal that certain antifungals could modulate neuroinflammation, or that beta-blockers could potentially dampen cancer metastasis. Harvard’s TxGNN model, published in Nature Medicine in 2024, demonstrated this approach at scale, ranking drugs as potential treatments across more than 17,000 diseases, including conditions with no existing therapies.

Contrastive learning models identify hidden relationships by comparing how different molecules affect cellular systems. They can recognise that a drug designed for one mechanism might coincidentally influence entirely different biological pathways.

Causal inference engines use population-scale datasets to test whether AI-discovered connections actually translate to clinical benefit. By analysing genetic proxies and real-world health outcomes across hundreds of thousands of patients, these systems can predict drug efficacy before a single new patient is enrolled in a trial.

What once took a decade of screening can now happen in months, with AI ranking millions of drug-disease pairs and validating the most promising candidates through computational simulations before any lab work begins.

Leading AI Drug Repurposing Companies

Several companies are at the forefront of AI drug repurposing, and their pipelines are moving beyond research demonstrations into clinical trials.

• Recursion Pharmaceuticals has accumulated over 20 petabytes of cellular imaging data and built one of the largest biological datasets in drug discovery. The company’s collaboration with Roche and Genentech, initiated in 2021 with a $150 million upfront payment, spans neuroscience and oncology programmes. In 2024, Roche exercised a $30 million option on Recursion’s first neuroscience phenomap. Recursion also completed a landmark merger with Exscientia in late 2024, combining their platform capabilities and expanding their partnership portfolio to include Sanofi, Bayer, and Merck KGaA.

• BenevolentAI gained prominence when its knowledge graph platform identified baricitinib as a potential COVID-19 treatment in January 2020. The drug, originally approved for rheumatoid arthritis, received FDA Emergency Use Authorisation in November 2020 after clinical trials confirmed it reduced mortality in hospitalised COVID-19 patients by 38%. The company has since established partnerships with pharmaceutical companies including Merck, focusing on oncology, neurology, and immunology targets.

• Healx specialises in rare diseases, combining patient registry data with omics analysis through its Healnet platform. The company advanced HLX-0201 (sulindac, originally an anti-inflammatory) to Phase 2a clinical studies for Fragile X syndrome after its AI platform identified the drug through gene expression matching. Healx received FDA IND approval in 2021 and has been evaluating multiple AI-identified compounds as potential combination therapies.

• Insilico Medicine used its PandaOmics platform to identify lifitegrast, an approved dry eye medication, as a candidate for endometriosis. Preclinical mouse studies published in Advanced Science in December 2024 validated reduced lesion growth. The company has also extended its generative chemistry platform to mine shelved compounds for age-related diseases.

• Verge Genomics applies network biology and AI to identify neuroprotective candidates, including mining discontinued pharmaceutical libraries for compounds that could address neurodegenerative diseases.

Real-World AI Drug Repurposing Successes

AI-discovered repurposing candidates are reaching patients. These are not theoretical projections; real drugs are entering clinical trials and, in some cases, reaching market based on AI-driven repositioning.

Baricitinib for COVID-19. BenevolentAI’s platform analysed biomedical data and molecular interactions to identify this rheumatoid arthritis drug as an inhibitor of both viral entry and inflammation in COVID-19 patients. The ACTT-2 and COV-BARRIER clinical trials validated the prediction, leading to FDA Emergency Use Authorisation and subsequent regulatory approvals globally.

Lifitegrast for endometriosis. Insilico Medicine’s PandaOmics platform identified this dry eye medication as a candidate through target discovery analysis. Mouse model validation showed the drug effectively suppressed endometriotic lesion growth, and the compound is advancing toward clinical trials for this new indication.

HLX-0201 for Fragile X syndrome. Healx’s AI platform identified sulindac (an established NSAID) through gene expression matching with the disease profile. Preclinical validation showed the drug modified behaviours characteristic of Fragile X. The compound received FDA IND approval for Phase 2a clinical studies.

Tretinoin for rare diseases. Researchers at Harvard Medical School identified this widely used acne medication as a leading candidate for treating rare diseases, demonstrating how AI can spot non-obvious therapeutic connections across disparate disease areas.

Multi-drug combinations for Alzheimer’s. A research team used the DeepDrug AI approach to identify a combination of five approved drugs, including tofacitinib and pravastatin, that target synergistic pathways in Alzheimer’s disease. This represents a new frontier: using AI not just to repurpose individual drugs, but to discover novel combinations of existing medications.

Challenges in AI Drug Repurposing: Data Bias, IP, and Clinical Validation

AI drug repurposing faces real obstacles that the industry is actively working to address.

Data bias limits generalisability. Large biobanks still over-represent European ancestry populations, which means efficacy signals in more diverse groups may be missed. Initiatives like the NIH’s All of Us dataset and H3Africa cohorts are working to close this gap, but progress is gradual.

Intellectual property is complex. Composition-of-matter patents on older molecules have often expired. Companies pursuing repurposing strategies must rely on formulation innovations, novel delivery mechanisms, or orphan drug exclusivity to maintain competitive advantages. This creates a commercial tension: the drugs most likely to be repurposed are often the hardest to protect commercially.

Clinical validation still requires real investment. Even after AI screening, repurposed drugs fail approximately 30% of the time in Phase III trials. AI reduces the front-end discovery timeline, but it doesn’t eliminate the need for rigorous clinical testing. Each candidate still requires funding, regulatory submissions, and patient enrolment.

Regulatory frameworks are still evolving. The distinction between “label expansion” and “new indication” continues to shift. The FDA and EMA are working to harmonise guidance for repurposed drugs, but formal frameworks are still developing. For companies navigating this landscape, regulatory ambiguity adds cost and delays.

The Future of AI Drug Repurposing: What to Expect by 2028

Several developments are expected to reshape the field within the next few years.

Regulatory agencies may increasingly accept validated real-world evidence to shorten approval pathways for repurposed molecules. Longitudinal electronic health record data, combined with genomics, could generate supportive evidence alongside traditional clinical trials.

Pharmaceutical companies will increasingly spin out “legacy asset” programmes, partnering with AI platforms on success-fee models to monetise dormant compound libraries without dedicating full internal resources. The Recursion-Roche collaboration is an early model for this kind of arrangement.

Foundation models for chemistry will merge structure-based generation with phenotypic screening, creating closed-loop systems that repurpose, redesign, and validate compounds in continuous cycles. Harvard’s TxGNN and similar graph foundation models are early indicators of this trajectory.

The convergence of these trends points toward a future where every approved drug is systematically evaluated as a potential treatment for multiple diseases, unlocking a reservoir of therapeutic value from molecules that have already passed safety screening.

FAQ: AI Drug Repurposing

What is AI drug repurposing?

AI drug repurposing uses artificial intelligence and machine learning to identify new therapeutic uses for existing medications. AI systems analyse large-scale datasets including genomics, molecular structures, patient health records, and clinical trial data to find connections between approved drugs and diseases they were not originally designed to treat. Because these drugs have already passed safety screening, they can potentially reach patients years faster than newly developed compounds.

How much does AI drug repurposing save compared to traditional drug development?

Drug repurposing can reduce development timelines by three to four years and cut R&D costs by up to 60% compared to developing a new chemical entity from scratch. Traditional drug development averages $2.6 billion per approved drug over 10–15 years. Repurposed drugs benefit from existing safety data, established manufacturing processes, and in many cases, prior regulatory review, all of which reduce the financial and time investment required.

What are examples of drugs successfully repurposed using AI?

Baricitinib, originally approved for rheumatoid arthritis, was identified by BenevolentAI’s platform as a COVID-19 treatment and received FDA Emergency Use Authorisation. Healx’s AI platform identified sulindac (an anti-inflammatory) as a candidate for Fragile X syndrome, now in Phase 2a trials. Insilico Medicine used its PandaOmics platform to identify lifitegrast (a dry eye drug) as a potential endometriosis treatment, validated in preclinical studies.

What AI technologies are used in drug repurposing?

The primary AI approaches include graph neural networks (which map relationships between genes, drugs, diseases, and biological pathways), contrastive learning models (which compare molecular effects across cellular systems), and causal inference engines (which use population-scale health data to validate predicted drug-disease associations). Large language models and foundation models like TxGNN are also being applied to rank repurposing candidates across thousands of diseases simultaneously.

What are the main challenges facing AI drug repurposing?

Key challenges include data bias in training datasets (which over-represent certain populations), intellectual property complications for off-patent molecules, the ongoing need for clinical validation despite AI screening, and evolving regulatory frameworks for repurposed drugs. Additionally, translating computational predictions into clinical benefit requires significant investment in trials and regulatory submissions.

What Comes Next

AI drug repurposing is shifting from proof-of-concept to production pipeline. The baricitinib story demonstrated that AI can identify a life-saving therapy in weeks rather than years. Companies like Recursion, BenevolentAI, and Healx are building on that precedent with expanding clinical programmes and major pharma partnerships.

For pharmaceutical executives and R&D leaders, the strategic question is not whether AI-driven repurposing works. The clinical evidence is accumulating. The question is how to build or access the data infrastructure, AI capabilities, and partnership models needed to extract value from existing compound libraries.

For investors, the economics are attractive: lower capital requirements, shorter timelines, and established safety profiles reduce the risk profile compared to de novo drug discovery.

For patients, the promise is tangible: proven molecules reaching new disease populations years faster than traditional development would allow.

Stay ahead of what’s next in healthcare.

Healthy Innovations is my weekly newsletter delivering strategic analysis of emerging biotech and digital health.

No spam. Unsubscribe anytime.