A new British company is taking an unconventional route into AI drug discovery, betting that synthetic chemical data generated through physics simulations can train better models than the limited experimental datasets available today. Potential Sciences, which launched on November 5, plans to build large-scale datasets by simulating atomic interactions and scaling those simulations up to small molecules, amino acids, and proteins. The goal is to create prediction systems that can identify promising drug candidates while weeding out compounds unlikely to succeed in clinical testing.
Why the Protein Data Bank Is Not Enough
Much of the current work in AI drug discovery relies on the Protein Data Bank, a public repository of 3D atomic structures that has grown over decades through experimental research. Hatem Helal, the founder and CEO of Potential Sciences, argues that this dataset is too small to train models capable of reliably accelerating the search for viable drugs. He frames the problem as one of scale: the number of possible molecular arrangements is vast, and finding the right ones requires navigating and pruning that space quickly.
To overcome the data bottleneck, Potential Sciences is generating its own synthetic data by simulating how chemical compounds interact. The theory is that far larger training pools will produce more advanced AI models and prediction systems. Helal’s stated objective is to make molecules that will never work in the clinic fail quickly, which requires better predictors of how compounds behave when they meet human biology.
Government Backing and a Physics Angle
The startup has seed funding in the tens of millions of dollars from SovAI, the UK government’s AI venture capital fund, and Twin Path Ventures, among other investors. The founders declined to disclose the precise amount or the company’s valuation. SovAI was established earlier this year with a mandate to invest in British startups that could become essential parts of the global AI supply chain. The fund has roughly $800 million at its disposal and can offer portfolio companies access to the UK’s national AI supercomputer.
Joséphine Kant, head of ventures at SovAI, said that approaching drug discovery through the physics of chemistry sets Potential Sciences apart from other companies. She has joined the startup’s board as part of the investment. Access to computing resources, she noted, allows the company to work at a scale that would be difficult for a small startup to achieve otherwise. Potential Sciences joins other AI drug discovery companies backed by SovAI, including Basecamp Research and Isomorphic Labs, as well as materials discovery startup CuspAI.
Scientific Promise and Practical Limits
Academic observers see potential in the simulation-based approach. Alan Fersht, an emeritus professor of chemistry at the University of Cambridge who specializes in protein engineering, said that accurately calculating interactions between all molecules in a cell and their structural fluctuations could predict where molecules bind, and that this would revolutionize drug development. He added that he is optimistic this will eventually happen.
Peter Wolynes, a professor of chemistry at Rice University, expects the approach will yield an improvement, but he cautioned that protein engineering remains at an early stage of development. His assessment suggests that the engineering challenges are substantial, even if the underlying idea is sound.
Even if AI can identify promising drug candidates more quickly, the cost and time required to test their safety and efficacy remain. Wolynes noted that the cost of designing a drug is small compared with determining what it will do to mice, to sick patients, and to healthy people. In his view, design is not the rate-limiting step. That means new treatments will continue to take a long time to reach the market regardless of advances in AI.
Competition and Dual-Use Concerns
AI drug discovery is a crowded field. Potential Sciences will face competition from numerous entrepreneurs and academics racing to solve the same bottlenecks from different angles. The company also has to contend with the possibility that technologies able to identify effective treatments could be misused to develop dangerous chemicals. Helal acknowledged that tools designed to minimize toxicity could, in principle, be flipped to maximize it.
Potential Sciences has not settled on a way to generate revenue from its technologies, nor has it decided whether to make its models freely available given the potential for misuse. Helal remains convinced that drug discovery is a fundamental, high-impact problem worth solving with AI. He noted that many promising breakthroughs have failed to make a real impact in the field, and his company is betting that its simulation-first strategy can change that.
