This is a 9AM Business interview with Dr Irina Babina, CEO of cancer startup Concr.
Why are AI and digital twins becoming central to drug development?
Drug development is expensive and slow, and clinical trials operate like a lottery. Over 96% of cancer drugs fail between Phase I and approval, with each failed drug representing an estimated ÂŁ4.4 billion in wasted investment and thousands of patients denied potentially life-saving treatments. Over 50% of that failure is attributed to lack of efficacy, heterogeneity and inability to outperform standard of care.
Adapting a multi-dimensional dynamic view of individual patients rather than populations solves this bottleneck, and AI and digital twins are the enablers. A digital twin is a virtual model of an individual patient’s cancer biology built from combining their genomic, molecular, and clinical data. It allows you to simulate how that patient’s biology might respond to different treatments before they’re given.
For drug developers, this means you can test trial designs, identify the right patient populations, and predict outcomes computationally before dosing real patients. That reduces risk, saves time, and means fewer patients are exposed to treatments that won’t work for them, sparing them physical and emotional toil, and cutting the enormous cost of failure that ripples through health systems and drug programmes alike.
What makes this moment different is that regulators are increasingly coming on board. Earlier this year, the U.S Food and Drug Administration (FDA) released draft guidance on using simulation-based and Bayesian approaches in clinical development, and this week EMA published theirs. It puts the wind in our sails as the pieces are finally coming together.
How are astrophysics methods being applied to predict cancer treatment outcomes?
Astrophysics and cancer biology share the same core challenge: understanding something enormously complex that you can’t directly observe, with limited, often messy and fragmented data.
When astrophysicists study dark matter or black holes, they can’t see the thing itself. As a workaround, they measure its effects on surrounding matter and use Bayesian statistical methods to infer what’s really happening. Cancer works the same way. We measure genomic mutations, RNA expression, clinical symptoms – all indirect signals – and try to infer how a tumour will behave and respond to treatment.
At Concr, we adapted these Bayesian frameworks from astrophysics to build a foundation model of cancer biology. Our FarrSight® platform integrates molecular and clinical data to create digital twins of individual patients’ cancers. It can then predict how each patient is likely to not only respond, but also benefit, from existing and experimental therapies, under the conditions of clinical practice. So, each patient is treated like a galaxy, and their true treatment response is the dark matter.
The key advantage of Bayesian methods is their ability to handle uncertainty. Most AI models need vast amounts of clean, uniform data to work. Our approach requires significantly less data and is tailored to the conditions of the real-world clinical practice, using existing diagnostic tests. This is critical for rare cancers with small patient cohorts where data is inherently scarce.
Why may the biggest opportunity in oncology not be new drugs, but better patient matching?
Our own research in over 3000 patients across 32 cancer types showed that by better capturing a patient’s biology with multi-modal data using digital twinning, median overall survival more than doubled, from 4.1 to 9.8 years. No new tests, no new drugs, just more sophisticated stratification.
There’s a lot of focus in oncology on developing better drugs, which makes sense and matters a lot. But there are already a lot of good drugs out there. The question I keep coming back to is: what if we had all the drugs? The real bottleneck isn’t the drugs themselves, but our inability to match the right drug to the right patient at the right time and correctly predict how that patient will respond to it throughout their treatment journey. Because very often in oncology, one patient would go through several treatments, and we need tools to adapt to the dynamic and evolving nature of the disease.
Diagnostics, predictive and decision support tools are critical to make the best match, but historically have been chronically underfunded. It’s high-risk, capital-intensive, and the financial returns are a fraction of what a drug can generate. But the patient impact of getting this right could be far greater than any single new drug, and investors are beginning to recognise this. If you could reliably predict which treatment would work for which patient, you’d transform outcomes without lengthy, expensive and risky development cycles.
How are TechBio startups helping pharma companies reduce clinical trial risk?
Pharma is under enormous pressure right now. A wave of blockbuster drugs is going off patent in 2027 and 2028, which is expected to wipe out over $180 billion in annual revenue, representing almost 12% of the global market. That means pharma needs to replenish pipelines fast, but clinical trials remain slow, expensive, and prone to failure.
This is where agile TechBio startups can really fuel change. Again, a lot of focus is on drug discovery – identifying the initial compound with high promise. But failure in development, particularly in late-phase trials, is the ultimate killer.
Companies like ours use predictive modelling and simulation to help pharma de-risk their trials before they spend hundreds of millions recruiting patients. We can simulate how different biomarker strategies capture patient populations likely to derive benefit from the new drug, and test trial designs computationally. That means fewer surprises in Phase II and III, where the costs escalate.
For pharma, the biggest value isn’t only cost savings but time. Getting a drug to market two years earlier can mean billions in additional revenue while it’s still under patent. So the economic case for working with companies that can accelerate timelines and improve the probability of success is very strong. Some companies we work with use several modelling solutions to test their hypotheses, really maximising probability of success before deploying large capital.
And because regulators are increasingly supportive of this direction, there’s a meaningful shift in how clinical evidence will be generated and evaluated. So early adopters can gain a sustained competitive edge.
What should investors understand about the growing digital twins market in healthcare?
The digital twins market in healthcare is still nascent, but growing rapidly. It was valued at approximately US$ 2.69 billion in 2024 and is expected to grow at a compound annual growth rate of around 68% over the next five years. For investors, three things matter.
First, look at where the technology is validated. They must distinguish between companies that have tested their models in real clinical settings, working with real patients, and those that only theorise and make educated guesses.
Second, consider the commercial model. There aren’t yet enough data points to converge on a single model, because it really depends on how value is generated. The pharma use case is most immediate because the economics of de-risking a clinical trial are so compelling. But the longer-term opportunity in clinical decision support for healthcare systems is arguably even bigger, and still largely untapped. Most investors want to know: is it a platform or a product? In this new age, there are more options than that.
Third, understand the regulatory direction. The FDA and European Medicines Agency (EMA) are both moving towards accepting simulation and modelling as part of the evidence base for drug approvals. But there are already proven, albeit difficult, paths to approval, captured in this recent paper. Important to remember though, regulatory approval doesn’t mean adoption and reimbursement.
Can you share more about your background and previous roles?
I trained as a geneticist and spent about 12 years in academia, developing targeted therapies for breast and gastric cancer at the Royal College of Surgeons in Ireland and the Institute of Cancer Research in London. I was passionate about science, but I kept seeing the same problem repeat itself: good ideas dying because there wasn’t the right support or funding to take them further.
That frustration pushed me out of the lab and into funding: firstly government research funding, and then venture capital after completing an MBA. That really unlocked my view of how scientific solutions can drive meaningful progress, so I worked with many scientific founders at early stages to help them shape their business plans.
Concr was one of those companies I started advising in 2021. It stood out to me because the technology was genuinely different: applying established, proven astrophysics methods to a problem that desperately needed a fresh approach. I got more and more involved, joined full-time, and was eventually appointed CEO in January 2024.
Do you have a view on the impact AI will have on one jobs – considering that the ‘godfather’ of AI and others have said it will replace jobs in a way we’ve not seen?
I’d agree with that. Perhaps not replace necessarily, but augment. When I was still a researcher, it took me weeks to put together a literature review, then days drawing up figures. It’s something that can now be done in seconds. I can see how more junior roles in certain fields, like software engineering for example, will become redundant.
In science what that will also do though, is unlock time and bandwidth to ask more important questions and carry out the experiments. Hypotheses still have to be tested in the physical world. The scientific judgment, reasoning, the clinical insight, the decision about what to do next, those remain entirely human. It’s how we approached development of our own technology, making sure humans are always in the loop and that biological insights we generate are fully explainable.
So, I see AI as augmenting what we can do. The complexity of biology is such that you still absolutely need clinicians, scientists, and researchers making judgments. What AI does is help them process information at scale and at speeds that are simply not possible for a human brain alone. In the meantime it poses an interesting challenge for educators: how to train the next generation of scientists so they can thrive in this digitally-enabled world.