When every drug becomes a possibility

Dr. David Fajgenbaum remembers the moment clearly. Lying in a hospital bed, the medical student turned patient faced a harsh reality: the drugs designed for his rare disease, Castleman disease, weren’t effective. In fact, he had almost already died five times. But David then had a radical thought: What if the answer wasn’t a new drug at all, but an old one used in a completely different way?

Twelve years later, David is still enjoying remission from the existing drug he found after much slow and meticulous research.

Cures hidden in plain sight

David’s transformation from a desperate patient to a physician-scientist aiming to expand drug repurposing with AI highlights a shift in thinking. For decades, pharma followed a single approach: one disease, one drug, 10-15 years, and roughly a billion dollars. But what if, among the 4,000 FDA-approved drugs, there are treatments for many of the 14,000 diseases that currently lack approved therapies?

This compelling promise inspired two recent ‘AI in action’ webinars: ‘Impact and potential for effective drug repurposing‘ and ‘Applying Elsevier biological relationship datasets in R&D workflows‘. The first features David alongside two thought leaders from the frontlines – heading patient-centric organizations desperately searching for treatments for devastating conditions. The second webinar offers a deeper look into the technological infrastructure that supports their mission.

Out of necessity

The urgency is impossible to overstate. Nina Frost, who leads RARE Hope’s research program for Alternating Hemiplegia of Childhood (AHC), shares a sobering statistic: 30% of children born with a rare condition will not live to see their fifth birthday. For her daughter Annabelle and hundreds of other children with AHC, time is not a luxury they have.

“We’re navigating this path out of necessity,” Nina explains. “The clock is ticking for our patients, and we can’t afford not to take this path.” Similarly, Cara O’Neill from the Cure Sanfilippo Foundation describes watching children, including her daughter, who appear normal at birth, gradually lose their ability to speak, walk and swallow as toxic fragments build up in their nervous systems. Despite the disease being known since the 1960s, no approved treatments are available six decades later.

These aren’t isolated cases. The traditional drug development model, while responsible for tremendous medical advances, simply cannot address the vast landscape of rare and neglected diseases quickly enough. The economics don’t work for small patient populations. The science takes too long. And thousands of patients run out of options – just like David did.

Throw AI at it and see what sticks

David’s answer, crystallized in his nonprofit Every Cure, is not about starting with a disease and spending years developing a drug to match it or, alternatively, starting with a drug and trying to expand its indications. Rather, Every Cure begins with all drugs and all diseases simultaneously. Using AI-powered platforms that analyze vast biomedical knowledge graphs, they quantify the strength of evidence between every possible drug-disease combination.

“We might learn that lidocaine is a potential treatment for preventing recurrence of breast cancer,” David notes. “A priori, you would never have thought to be looking for some new use for lidocaine.”

The approach has already yielded impressive results. When a patient with Castleman disease failed to respond to all available treatments, Every Cure’s platform identified adalimumab, a TNF inhibitor never before used for this condition, as the top candidate. The patient has been in remission for over two and a half years. In another case, the platform helped identify folinic acid for children with cerebral folate deficiency – a condition that, within days of treatment, allowed them to speak their first words.

However, these successes depend on a foundation that’s often overlooked: data. Comprehensive, high-quality, structured biological data.

The knowledge infrastructure

This is where Elsevier’s EmBiology knowledge graph comes into play – the technological foundation that enables AI-driven drug repurposing at scale.

Maria Shkrob, a molecular biologist and consultant in Elsevier’s Professional Services Group, explains how EmBiology addresses one of the biggest challenges in biomedical research: making the world’s scientific literature computable. With over 36 million MEDLINE abstracts and 7.6 million full-text articles (and counting), it’s impossible for anyone to keep up without assistance. Meanwhile, traditional literature searches help researchers find specific papers but don’t provide the comprehensive, systematic analysis needed to uncover hidden repurposing opportunities. EmBiology transforms this unstructured text into a structured network of 1.5 million biological concepts – including genes, proteins, diseases, drugs, and pathways – connected by over 18 million relationships automatically extracted from the literature. Unlike large language models that function as “black boxes,” every connection in this knowledge graph is traceable back to its source publication.

“No hallucinations, no false sources, no made-up references,” Maria emphasizes. “Every connection has solid references to support it.”

Triangulating truth

For patient organizations venturing into AI-assisted drug repurposing, trust is paramount. Nina Frost describes RARE Hope’s deliberate approach: using three parallel methodologies to identify drug candidates for AHC. One was a drug prediction algorithm from Unravel Bioscience via the rareSHIFT program. Another was extensive collaboration with disease experts to hypothesize mechanism-of-action categories. The third was Elsevier’s EmBiology platform, mining published literature with disease-specific inputs.

“Individually, all of the approaches had their limitations,” Nina acknowledges. “But when we combined the approaches, that’s when we started to feel like we had something compelling. It’s like the premise of GPS tracking, it’s the intersection of three points that becomes meaningful.”

The result was an “AHC Smart Library” of ranked drugs that could be validated across multiple lab systems and model organisms, a practical tool built on cautious optimism rather than hype.

This measured approach reflects a broader theme: AI and sophisticated data platforms aren’t magic solutions that replace human judgment. They’re hypothesis generators that identify promising starting points, which must then be rigorously validated through laboratory studies and clinical trials.

The integration challenge

What makes EmBiology particularly powerful is its interoperability. Since every biological concept carries standardized public identifiers, researchers can connect literature-derived insights with experimental data, clinical trial results, chemical databases and proprietary internal data sources. David’s team, for instance, uses the platform as one component of a multi-layered approach that also incorporates proteomics data, flow cytometry and other experimental evidence.

Meanwhile, the integration extends beyond biology into chemistry and safety. Researchers can identify a target in EmBiology, then use Elsevier’s Reaxys database to find chemical compounds active against that target, analyze scaffolds and even plan synthesis routes. Alternatively, they can explore a target’s safety profile using PharmaPendium, which contains databases of FDA-approved documents and adverse event reports.

As Cara notes, “To effectively repurpose drugs and find these safe and effective treatments, we do need to be relying on a lot of different data sources.”

Looking forward

Amid all the technological sophistication, the webinar participants kept returning to what matters most: the patients. Children like Annabelle, who has a “superpower” of making anyone love her. A young boy like Ryan, who spoke his first unprompted words after receiving folinic acid – he told his father who was leaving for work that he loved him. “Of course, his dad cried for the next 15 minutes,” David recalls. “We call ourselves Every Cure because this is an opportunity to truly help everyone with the medicines we already have,” David concludes. His current goal – identifying and advancing over 20 repurposed treatments for devastating conditions by 2030 – is ambitious. But with advanced biological knowledge graphs transforming the world’s scientific literature into actionable insights, and with patient organizations bravely embracing new technologies while maintaining strict standards, that vision is becoming achievable.

And for the millions of patients with rare diseases, that future can’t come soon enough.

Related Articles

Every Cure: turning old drugs into new hope

Turning rare disease research into a common goal

LLMs as a Jury: Bringing Quality to Quantity in GenAI-Aided R&D

Repurposing drugs for rare diseases, cancer and beyond