5 Questions to Romain Clement, Founder and CEO at ArcaScience

Q1: In a recent keynote at Health.tech, you spoke about your own journey as a patient for the first time. Can you tell us more about this defining moment, and how this led to ArcaScience? 

I was 19 when a radiologist pointed at a 9-centimeter mass on my MRI scan.  Being with my mom at that moment, I cared more about what she felt rather than what I was supposed to assess. The diagnosis was presented to me as a cold, logic fact, and I was being treated as a statistic, rather than an individual.  He didn’t have the full picture; he only had the ‘average’ of what that shadow usually meant. At that moment, I was just a university student who studied literature and coded in my free time. And I realized that in medicine, patients are regarded as averages: from the beginning, it felt like the system failed to understand, prospectively,  what a treatment really is to a patient, and how we are supposed to integrate it into our lives, with its benefits and its risks.

The technical blueprint for ArcaScience was born from that moment. Our goal became Total Integration: building a system that treats every piece of fragmented evidence - every paper, every safety report, every real-world data point, every patient journey - as a connected thread in a single person's story. Our architecture is designed to eliminate that 'standard protocol’ by providing a continuous, auditable flow of truth so that no patient is ever left as a mere statistic. 

I wanted to hand out to clinicians what I would have wanted to feel when that radiologist pointed his pen towards the mass: considered and informed with MY own chances and MY own path back to life, back to normality.

Q2: Why is "Standard Protocol" such a dangerous phrase in modern oncology? 

'Standard protocol' is dangerous because it is the ultimate compromise. It is a treatment designed for the 'median' patient - a person who doesn't actually exist.

When we rely on standard protocols, we are essentially admitting that we don’t understand the Benefit-Risk profile of the drug well enough to specialize. Most  drugs in development fail, because we are trying to make “One size fits all” medicine. At ArcaScience, we believe the danger lies in the opacity of the standard. If you don't know why a drug helps 'Group A' but harms 'Group B' before the trial starts, you aren't practicing medicine; you're practicing probability. We want to replace 'Standard Protocol' with 'Precision Reality.’

Q3: In 2020, ArcaScience was selected by the French government to tackle the COVID-19 crisis. Looking back, how did that high-stakes environment bridge the gap between "identifying candidates" in a dataset and actually supporting confident clinical decisions?

When the pandemic hit, the French government needed more than a search engine: they needed a decision engine. There was a mountain of fragmented data: pre-prints, early trial results, and molecular signals, all of which were scattered across the globe. Our task was to harvest and merge every piece of relevant data to identify which existing drugs could be repurposed to save lives.

That was a defining moment for our infrastructure. COVID-19 was our proof of concept for speed and scale. It proved that our AI could harmonize chaos into actionable insights. But more importantly, it pushed us to concentrate ArcaScience around the mission of Prospective Benefit-Risk Evaluation. We learned that the platform had to provide the transparency and rigor required to move a drug from a 'potential hit' to a ' Best-in-class for specific patients.

Q4: You mentioned a turning point involving a cancer patient who asked for help, how did that defining moment change the actual code and infrastructure of your platform? 

Yes, in 2020 a patient with glioblastoma asked for our help. In 2020, we couldn’t help her. That night I went home and sat in silence. It was a defining moment for me both in my personal life and in my journey as an entrepreneur. At that time, we could show what might work, but we were still refining how and for whom. We were able to identify promising candidates and biomarkers by looking at our Benefit-Risk datasets, but identifying leads is not the same as supporting a life-or-death treatment decisions. We had the signals and the hypotheses, but we realized that to truly move the needle in oncology or rare diseases, we needed a more integrated, auditable model - one that doesn't simply rank candidates but provides the 'Confidence Score' a clinician needs to act.

We didn’t have what she needed. We needed to change.

Our code was excellent at finding signals, but it wasn't yet built to provide accountability. We went back to the infrastructure and built what we now call our stack of 24 specialized models. We needed to avoid the problems of a 'Generative AI' that could hallucinate; we needed a Traceability Engine, so we hard-coded the ability to 'walk back' every single insight to its source. We integrated one of the original scientists behind the Transformer architecture (the underlying technology of ChatGPT, Claude, Mistral and many others) specifically to ensure that our infrastructure was grounded in clinical rigor. We shifted the focus from 'What does the data say?' to 'What is the most auditable benefit-risk path for this specific profile?'The patient who asked for help was a compass, and following her direction, we’re beginning to redefine the industry standard.

Despite huge advances in research and technology in the last decades, 95% of drugs still fail in clinical trials. What does a world look like where ArcaScience’s "World Model" is the industry standard? 

It looks like a world where we finally stop betting on billion-dollar molecules. Right now, we land probes on comets because physics is predictable; drug development feels unpredictable only because our information is fragmented, and our  AI models not well framed. At ArcaScience, we streamline benefit‐risk analysis by integrating every relevant piece of evidence (quantitative and qualitative) into one coherent, auditable model.

In a world where our World Model is the standard, the '95% failure rate' becomes a relic of the past. We would see 'Prospective Benefit-Risk' modeling used before Phase II, not as a post-mortem after a failed Phase III. It means drugs are approved for the right people faster, and 'off-label' use isn't a shot in the dark, but a data-driven choice. Most importantly, it means that the woman I met at the hospital, who asked me for help, wouldn’t be left in the dark. She would be looking at a screen that gives her clarity, confidence, and - most importantly - time. 'Standard' will finally mean standard excellence, not standard compromise.