Helical raises $10M for in silico reproducibility

Drug-discovery AI has a weird problem. The models keep getting better, but the work around them is still messy. A scientist gets a promising prediction, an ML team runs a notebook, somebody tweaks a parameter, and pretty soon nobody can fully retrace how a decision got made. That is the gap Helical is trying to sell into. On April 14, the company said it raised a $10 million seed round to build what it calls a virtual AI lab for pharma. ### What is Helical actually building? Basically, not another single foundation model. Helical is pitching an application layer that sits on top of biological models and turns them into repeatable workflows scientists can run, validate, and defend. The product has two sides — a Virtual Lab for biologists and translational scientists, and a Model Factory for ML engineers and data scientists — all working off the same data and results. (tech.eu) ### Why does “reproducible” matter so much here? Because drug discovery is not a demo problem. A flashy model output is useless if a pharma team cannot show what data went in, what settings changed, what evidence supported the call, and whether another team can reproduce the same result later. That is the real bottleneck Helical keeps pointing at — not raw model quality, but the handoff from computational guess to scientific decision. (tech.eu) ### Who backed the round? The seed round was led by redalpine. Gradient, BoxGroup, and Frst also joined, along with angels including Cohere CEO Aidan Gomez, Hugging Face CEO Clement Delangue, and footballer Mario Götze. That mix is telling — part AI infrastructure signal, part startup-finance signal, part visibility play. ### What problem are pharma teams running into? (tech.eu) Turns out a lot of AI drug-discovery work still lives in one-off analyses. Bench scientists and ML engineers often work in separate systems, and new model architectures arrive faster than organizations can operationalize them. The result is duplication, fragile notebooks, and outputs that are hard to transfer across programs or therapeutic areas. ### Does Helical have real customers yet? It says yes. Helical says it is already working with several top-20 pharmaceutical companies. One profile of the company names Pfizer on predictive safety biomarkers and Tanabe Pharma, and says early deployments compressed some discovery timelines from years to weeks. That claim should be read as company-reported, but it explains why investors were willing to fund the workflow layer instead of another model shop. (tech.eu) ### Why not just let each pharma company build this itself? Because the hard part is not only running a model. It is creating a shared system where biology teams and ML teams can iterate on hypotheses the way a wet lab would — but computationally, with versioning, traceability, and evidence that survives handoffs. Think of it like moving from scattered spreadsheets to an actual operating system. The model is one ingredient; the lab process is the product. (tech.eu) ### Where does the new money go? Helical says the $10 million will fund deeper deployments with existing pharma clients, expansion into more top-20 pharma programs and therapeutic areas, and more work on its “evidence layer,” which is the part meant to make in-silico experiments trustworthy and reusable across diseases. ### Bottom line? (markets.businessinsider.com) This round matters because it shows where the next AI-in-pharma fight is moving. Not just toward smarter models, but toward software that makes those models legible inside big drug companies. If Helical is right, the valuable layer is not the prediction itself. It is the system that makes the prediction auditable enough to act on. (tech.eu)