AI Can Help Find New Uses for Old Drugs; Advancing Them to Patients Is the Hard Part

Contributed Commentary by Dr. David Fajgenbaum 

September 18, 2026 | In 2010, I was a third-year medical student when I became critically ill with idiopathic multicentric Castleman disease (iMCD), a rare condition that landed me in the hospital and on the brink of death. After seven chemotherapies temporarily held my symptoms at bay, another relapse led to the words no patient wants to hear: there were no more options. My disease would kill me if I didn’t act quickly. 

Waiting for a brand-new therapy, the kind that can take 10 to 15 years and billions of dollars to create, was not a viable option for me. Repurposing an existing drug was my only real shot at survival.  

For many diseases, drugs already approved for other conditions may act on relevant biological pathways that are shared with other conditions. Rather than starting from scratch, we can build on previous research, find existing drugs that work through similar mechanisms, and use low-cost, widely available drugs “off-label” to help treat patients with rare diseases. 

While waiting for my next relapse to occur, I started collecting my own blood samples and studying them in the lab. We found an overactive pathway and tried a 25-year-old, already FDA-approved, transplant drug called sirolimus, to calm the overactive signal. This drug had never been used for iMCD before. 

It worked, and I have now been in remission for more than twelve years. This experience reshaped how I think about medicine and about time.  

What made a difference in my case was not a single breakthrough, but the ability to quickly connect scattered biomedical evidence, identify how an existing drug might influence a disease pathway, and treat each possibility as a testable hypothesis. Off-label prescribing is relatively common in medicine, but many more potential treatments hide in plain sight.  

My story revealed a powerful opportunity to systematically identify more repurposing opportunities for the thousands of diseases and millions of patients the traditional drug development system leaves behind. In 2022, I co-founded Every Cure to bring this opportunity to life. 

Today, AI can generate more drug-repurposing hypotheses than clinicians and researchers can evaluate manually. But finding possible connections is not the same as proving which ones are safe, effective and clinically useful. At Every Cure, we are working end to end, from discovery to patient impact, to ensure that promising and scientifically validated repurposing opportunities help the people who need them most.  

The FDA’s recent announcement of a new initiative to collect information about drug repurposing is an important step forward. In May, the agency stated that, “drug repurposing can make better use of available scientific data to deliver effective treatment options for patients in need.” This recognition reflects a growing understanding that drug repurposing represents a faster way to bring treatments to patients with unmet medical needs.  

The Repurposing Gap 

In 2014, I began taking sirolimus, the FDA-approved transplant drug that saved my life. Through my 12 years in remission, I continue to grapple with the question: How many existing drugs could treat diseases they were never designed for, if we had a systematic way to find them? 

Since it’s not profitable for companies to repurpose existing drugs, especially the 80% that are already generic, no one is incentivized to invest in the funding to prove that these drugs work. For this reason, Every Cure is structured as a nonprofit organization, solely dedicated to saving and improving lives by repurposing drugs. 

Our approach treats drug repurposing as a ranking and validation problem: first identifying plausible drug-disease links across millions of possible pairings, then moving the strongest candidates through expert review. Instead of starting with a single drug or a single disease, we make predictions across roughly 3,000 FDA-approved drugs and 22,000 recognized diseases, evaluating over 65 million possibilities to identify the opportunities with the greatest potential for patient impact. We currently have 15 active repurposing programs across a variety of diseases, ranging from ultra-rare like Bachmann-Bupp syndrome to more common diseases like breast cancer.  

What AI Can and Cannot Prove 

The biological world is inherently about connections: drugs, diseases, genes, pathways, interactions, phenotypes, and the evidence that science has discovered to describe them. To capture these relationships at scale, Every Cure uses an AI-driven platform built on knowledge graphs and machine learning. We recombined and extended existing biomedical knowledge graphs into one integrated system, creating a foundation for large-scale computational analysis.  

A graph database provides a way to compute and query relationships at scale. ​​For each drug-disease pair, the platform looks for paths linking a drug’s known targets to genes, pathways or phenotypes implicated in the disease. A candidate might rise in the rankings because it acts on a pathway that appears repeatedly in patient data and in related diseases. It can also fall if the evidence depends on a weak animal model, a single paper or a contraindication that would make clinical use unsafe. 

Using graph analytics in Neo4j Graph Data Science with modeling tools such as XGBoost and scikit-learn, we treat drug repurposing as a ranking and a search problem. Our platform systematically evaluates and prioritizes millions of possible drug-disease relationships, allowing work that would traditionally take years to be completed in days. This graph-based architecture also enables us to move beyond periodic “all-vs-all” searches and continuously prioritize the most promising candidates for focused review and validation. 

Clinical expertise remains essential throughout this process. No team can manually triage a search space with up to 65 million possibilities, but AI alone is not enough. Our feedback loop between humans and technology is the operational heart of the model. It keeps the system grounded in clinical and biological judgment, while continuously improving how the platform prioritizes opportunities. What once required months of manual literature review and expert consensus-building can now begin with a ranked shortlist generated in hours. 

Computational drug repurposing or “computational pharmacophenomics” also has its limitations: it can make a hypothesis appear more certain than the underlying evidence supports. A drug may affect a pathway associated with a disease, but still fail in patients because of factors such as dose, tissue exposure, disease timing, comorbidities or safety risks. For rare diseases, the validation problem is even harder because patient populations are small and randomized trials may be slow, expensive or impractical. To move the field forward, we need better ways to combine evidence, clinical data, registries, natural history studies and carefully designed trials. 

Alongside Every Cure, a growing number of groups are working to make drug repurposing more systematic and scalable. Academic teams have built graph-based models such as TxGNN, developed by Harvard Medical School’s Marinka Zitnik, which rank drug candidates across thousands of diseases and provide interpretable paths for clinician review. We know that AI can produce plausible drug-disease hypotheses, but we’re still working on how medicine should evaluate, fund, and act on these opportunities. 

Programs like the FDA’s Office of Orphan Product Development and Rare Neurodegenerative Disease Grants Program help support drug repurposing with research dollars. International organizations like the Drugs for Neglected Diseases initiative help treat rare populations in Africa, Latin America, and other underserved regions. Together, we can help change the system to focus on patients today, instead of waiting for the next billion-dollar drug development program. 

Every Cure has an ambitious goal of unlocking repurposed treatments for 15 to 25 diseases by 2030. In 2025 alone, Every Cure reviewed more than 9,000 repurposing opportunities, and our team currently evaluates approximately 1,000 unique opportunities each month. Feedback from these reviews is captured as structured data, allowing the platform to continuously retrain and refine how it prioritized candidates.  

But success depends less on generating leads than on proving which candidates can responsibly reach patients and actually getting them to patients. That will require stronger infrastructure for validating and advancing repurposed therapies, including proof-of-concept funding, clinical trials, rare disease registries, real-world evidence systems, and clearer reimbursement pathways for well-supported off-label uses. The FDA’s recent initiative to gather input on drug repurposing is a very positive step forward. 

We are beginning to see what that path can look like. Every Cure has been working with Dr. Luke Chen and other experts to evaluate lenalidomide and dexamethasone for Rosai-Dorfman disease (RDD), a rare disorder in which certain blood cells accumulate in the body. In June 2026, following new research co-authored by our team and others, the National Comprehensive Cancer Network named the combination a preferred treatment option for appropriate RDD patients. The work now shifts to making sure physicians and patients know that option exists and can access it. 

Drug repurposing has traditionally depended partly on luck: a side effect noticed by a physician, a mechanism spotted by a researcher or a desperate patient with no standard options left. AI and knowledge graphs now make it possible to search for those connections systematically. But generating hypotheses is only the beginning. The next challenge that we’re trying to solve right now is building the evaluations systems needed to prove efficacy and turn plausible hypotheses into treatments doctors can responsibly prescribe and patients can access. 

David Fajgenbaum, MD, MBA, MSc, is co-Founder & President of Every Cure and one of the youngest faculty members ever to receive tenure at the University of Pennsylvania School of Medicine. A physician-scientist and patient battling a deadly disease, he discovered and repurposed a treatment that saved his own life. Dr. Fajgenbaum has since advanced 13 more repurposed treatments for cancers and rare diseases and co-founded Every Cure to unlock more hidden cures from existing medicines. He can be reached at [email protected].