From Data Factory to Better Drugs: Inside Recursion's AI-native Product Engine

In our new video, we take you behind the scenes of Recursion’s AI-native data factory and product engine – the first, and one of the largest, of its kind. Capturing Cells AI Can Decipher It starts with perturbing cells via CRISPR and other compounds — dozens of methods at nano scale, generating up to 2 million experiments a week. Recursion’s labs are built to randomize experiments so models learn what's real biology versus noise. Cells are imaged and sequenced at scale, yielding thousands of measurements per experiment — high-dimensional data that trains our foundation models and can be mined again and again. Creating Maps of Biology Our foundation models turn every perturbation — imaging, genetic, and patient data — into a unique fingerprint. Mapping these fingerprints lets us compare genetic manipulations, spot ones with similar effects, and generate new hypotheses for what drives or rescues a disease state. De-Risking Discoveries & Designing New Drugs We validate hypotheses at the bench against the same high bar pharma partners hold themselves to. Each target becomes a chemistry problem, solved in Centaur — our integrated design environment running hundreds of ML and physics-based models. Starting from target properties, we generatively design novel molecules, improving with each cycle. A chemist can run a full design cycle in about a day, using tools like Nesso-1 and in-house AI agents to explore novel chemical space and plan design synthesis. Hundreds of models — ADME, potency, simulations — then narrow millions of molecules to a short, makeable list. We’ve produced 10+ development candidates, and each held to that same high bar. Synthesizing Compounds Compounds then move directly into automated testing. Scientists describe an experiment in plain language and the platform generates the protocol, cutting pharmacology costs roughly in half compared to industry. Every result feeds back into the models, letting us reach a development candidate with about 90% fewer molecules synthesized than industry average — one closed loop, built to make fewer molecules, not more. Carrying Insights Into the Clinic Clinical candidates enter our ClinTech platform. Our AI compares lab response data to real patient tumors to predict who benefits before enrollment even starts. Our Site Finder tool mines hundreds of millions of patient records to find high-quality trial sites in hours, improving enrollment 30–60% and slashing feasibility and startup timelines. Once running, a biometrics engine turns dose-escalation decisions that used to take 7–10 days into 24–48 hours. This is an end-to-end AI-native product engine that has already put multiple programs in the clinic and is already changing the way medicines are made. 👉 Watch the full video: https://www.youtube.com/watch?v=0T8MlHDruo8&feature=youtu.be #AI #drugdiscovery #drugdesign #biology #chemistry #TechBio #AIdrugdiscovery