The Pharmacy of Second Chances
By 2040, a patient diagnosed with a rare autoimmune disease will no longer hear the words “there is no treatment.” Instead, a doctor will open a digital catalogue of thousands of molecules—most of which failed their first clinical trial. Some were abandoned because they caused liver damage in rats. Others were shelved because the business case was too weak. A few were simply misunderstood. But each one will carry a second label: a note explaining exactly why it failed, and exactly how to fix it. That future is not science fiction. It is being built today, inside a small company in Cambridge, UK, where a chemist turned data scientist is teaching artificial intelligence to read the obituaries of dead drugs.
For decades, the drug development pipeline has followed a brutal arithmetic. For every 10,000 molecules that enter preclinical testing, only one reaches a patient. The rest fail. Some fail because they are toxic. Some fail because they do not work. But a large number fail for reasons that have nothing to do with the molecule itself. A drug might cause an off-target effect in a specific organ, or it might be metabolized too quickly in the human body. In many cases, the underlying science was sound. The problem was that nobody had the tools to diagnose the failure correctly.
Before artificial intelligence entered the field, a failed drug was simply a dead end. A pharmaceutical company would spend 10 to 15 years and roughly 2.6 billion dollars bringing a candidate to Phase II or Phase III trials. If the drug failed, that investment was written off. The molecule was locked in a vault, never to be tested again. The knowledge about why it failed—the metabolic pathway that went wrong, the protein it inadvertently bound to, the genetic variation in the trial population—was often lost, buried in internal reports that no one outside the company ever saw.
That is the world that Layla Hosseini-Gerami walked into when she began her career. She studied chemistry at the University of Leeds between 2014 and 2018. [2] At that time, artificial intelligence was still an emerging technology in drug discovery. There were no AI modules in her undergraduate degree. Her first exposure came during a one-year industrial placement in 2016, at a drug-discovery software company called Optibrium, based in Cambridge. [3] There, she worked as a machine-learning intern. She saw for the first time how computational models could predict how a molecule would behave inside the body.
In 2021, Hosseini-Gerami co-founded Ignota Labs with Jordan Lane and Sam Windsor. [1] The company is based in Cambridge, UK. Its mission is singular: to determine why drugs have failed in clinical trials, and then re-engineer the most promising therapies for a second attempt. The company uses artificial intelligence to analyze the chemical and biological data from failed trials. The AI looks for patterns that human researchers might miss. It asks questions like: Did the drug bind to the wrong receptor? Did it accumulate in a particular tissue? Was it broken down by an enzyme that varies between patients?
In February 2025, Ignota Labs closed a 6.9-million-dollar funding deal. [1] Since then, the company has built a strong pipeline of promising drugs. The pipeline includes treatments for autoimmune diseases and blood cancers. Hosseini-Gerami serves as the company’s chief data-science officer. In April 2025, Forbes magazine included her in its “30 under 30” list for European science and health care. [4] The citation recognized her work using artificial intelligence to accelerate the process of bringing safe drugs to market.
The core insight behind Ignota Labs is deceptively simple. Most drugs fail not because they are bad molecules, but because they are good molecules tested in the wrong context. A molecule that causes liver toxicity in a patient with a specific genetic variant might be perfectly safe in a different population. A drug that fails because it is metabolized too quickly might succeed if it is reformulated with a different delivery system. The problem is that pharmaceutical companies rarely have the resources or the incentive to investigate these nuances. Once a drug fails, the company moves on to the next candidate.
Ignota Labs flips that logic. Instead of starting from scratch, the company starts from failure. The AI models are trained on vast datasets of chemical structures, protein interactions, and clinical outcomes. They learn to map the gap between a molecule’s intended effect and its actual behavior in the body. When a drug fails, the AI can pinpoint the exact mechanism of failure. It can say: this drug failed because it activated a receptor in the liver that caused inflammation. Or: this drug failed because it was broken down by an enzyme that is overexpressed in 30 percent of the population.
Once the failure is understood, the company can re-engineer the molecule. Sometimes this means adding a chemical group to block a binding site. Sometimes it means changing the formulation. Sometimes it means identifying a different patient population that would benefit from the drug. The goal is to give the molecule a second chance.
This approach has parallels in other fields. In materials science, researchers routinely recycle failed experiments. A polymer that did not work as a battery electrolyte might be repurposed as a membrane for water filtration. In aerospace engineering, crash data from failed prototypes is used to design safer aircraft. But in medicine, the practice of learning from failure has been slow to take hold. The regulatory system is designed for new drugs, not for recycled ones. The financial incentives favor blockbusters, not second chances.
The approach of learning from failure has parallels in other fields. In materials science, researchers routinely recycle failed experiments. A polymer that did not work as a battery electrolyte might be repurposed as a membrane for water filtration. In aerospace engineering, crash data from failed prototypes is used to design safer aircraft. But in medicine, the practice of learning from failure has been slow to take hold. The regulatory system is designed for new drugs, not for recycled ones. The financial incentives favor blockbusters, not second chances.
The technology behind Ignota Labs is part of a broader movement in drug discovery. Other groups are working on similar problems. At the Massachusetts Institute of Technology, researchers have developed machine-learning models that predict drug toxicity from chemical structure alone. [7] At Stanford, scientists are using AI to repurpose existing drugs for new diseases. [8] At the University of Cambridge, a team is building a database of failed clinical trials, so that the knowledge is not lost. [9] Ignota Labs sits at the intersection of these efforts, but with a specific focus on the mechanism of failure.
The technology behind Ignota Labs is part of a broader movement in drug discovery. Other groups are working on similar problems. At the Massachusetts Institute of Technology, researchers have developed machine-learning models that predict drug toxicity from chemical structure alone. [7] At Stanford, scientists are using AI to repurpose existing drugs for new diseases. [8] At the University of Cambridge, a team is building a database of failed clinical trials, so that the knowledge is not lost. [9] Ignota Labs sits at the intersection of these efforts, but with a specific focus on the mechanism of failure.
The company’s approach has already yielded results. In its pipeline, there are drugs that failed in Phase II trials for autoimmune diseases. The AI identified that the drugs were being metabolized too quickly by a liver enzyme that varies between ethnic groups. The company is now reformulating the drugs with a different delivery system that bypasses the liver. Another candidate failed in a blood cancer trial because it caused bone marrow suppression. The AI found that the suppression was caused by a metabolite that could be blocked with a second compound. The company is now testing the combination.
What makes Ignota Labs different from traditional drug repurposing is the depth of the analysis. Traditional repurposing asks: what else can this drug treat? Ignota Labs asks: why did this drug fail, and how can we fix it? That distinction matters. A drug that failed because it was toxic at high doses might be repurposed at a lower dose for a different disease. But a drug that failed because it was toxic in a specific organ might need structural modification, not just a new indication.
The company is also building a library of failure mechanisms. Each molecule that passes through the pipeline adds to the dataset. Over time, the AI will become better at predicting failure before it happens. That is the long-term vision. In the future, a pharmaceutical company could submit a candidate molecule to Ignota Labs before starting clinical trials. The AI would analyze the molecule and say: this drug will fail in 70 percent of patients because it binds to a receptor in the heart. The company could then modify the molecule before spending billions on trials.
For a child born today, the implications are profound. That child will grow up in a world where the cost of drug development is lower, because the industry has learned to learn from failure. The time from molecule to medicine will shrink from 15 years to perhaps 5 or 6. Diseases that are currently considered untreatable will have multiple candidates in the pipeline, because the pool of viable molecules is no longer limited to those that succeed on the first try. The child will never know a world where a promising drug was abandoned simply because nobody understood why it failed.
The future that Ignota Labs is building is not a future of miracle cures. It is a future of systematic, data-driven recovery. It is a future where failure is not a dead end, but a signpost. It is a future where a chemist in Cambridge can look at a molecule that was written off a decade ago, and say: I know why you failed. And I know how to fix you. That is the promise of learning from failure, and it is a promise that is already being kept.
Sources
1. Ignota Labs
3. Optibrium
4. Forbes
6. Scottish Medicines Consortium