Finding New Drugs Is Laborious—This AI Startup Is Skipping the Lab Work
Potential Sciences, a UK government-backed startup, is betting that training AI models on computer-generated biochemical data can accelerate the plodding process of developing new drugs.
By Joel Khalili | November 5, 2026
Drug discovery has long been one of the most expensive and time-consuming endeavors in science—typically spanning a decade or more and costing billions of dollars per approved therapy. Much of that effort is consumed by iterative laboratory work: synthesizing compounds, testing them against biological targets, and discarding the vast majority that fail.
Potential Sciences, a startup backed by the UK government, is proposing a different path. Rather than relying solely on physical experiments to generate training data, the company is betting that AI models trained on computer-generated biochemical data—synthetic datasets that mimic real-world molecular behavior—can shortcut the early stages of drug development.
The premise rests on a familiar constraint in AI-driven biotech: high-quality experimental data is scarce, expensive, and slow to produce. By simulating biochemical interactions at scale, Potential Sciences aims to give its models far more training material than any single lab could feasibly generate, potentially narrowing the search for viable drug candidates before a single physical experiment is run.
It is an approach that reflects a broader shift in the pharmaceutical industry as of 2026, where AI-assisted discovery platforms have moved from novelty to mainstream investment. Yet skepticism remains: synthetic data is only as useful as the fidelity of the simulations behind it, and the ultimate test of any computational shortcut is whether it produces molecules that hold up in clinical trials. Potential Sciences—and its government backers—will be watching closely.
via Wired AI