Potential Sciences launches with seed funding to cut lab work in AI drug discovery

A new British startup, Potential Sciences, launched on November 5 with a plan to speed up drug design using AI. It says it 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 give the precise figure or the valuation.

The plan is to build large-scale chemical datasets by using basic principles of physics to simulate interactions between atoms, then small molecules, then complex amino acids and proteins. With that data, the startup is betting it can train AI models and prediction systems capable of identifying highly targeted drugs for specific diseases, bypassing costly and time-consuming lab experimentation. Founder and CEO Hatem Helal told Wired: "We want molecules that are never going to work in the clinic to fail fast." That, he said, means having better predictors of how molecules behave as they interact with human biology.

Helal argues the current approach has a data problem. A large proportion of AI drug discovery labs build their models on a single public database of 3D atomic structures, the Protein Data Bank, which researchers have expanded over decades with the results of new experiments. According to Helal, that dataset remains too small to train models advanced enough to reliably accelerate the search for viable drug candidates. He describes drug discovery as a search problem: how fast you can navigate and prune a vast space of possible molecules into promising ones for a particular disease. Potential Sciences is trying to build far larger datasets of synthetic data, produced by simulating interactions between chemical compounds. The theory is that far larger pools of data will let it train far more advanced models.

Helal also says that cutting the cost of experimentation could make it economical for pharmaceutical companies to pursue drugs for rare conditions that have previously gone untreated, and could make more bespoke treatments possible: medicines as one-off molecular combinations tailored to a person's biology.

Wired frames the open question as whether Potential Sciences can predict how well drug candidates bind to disease-related proteins as accurately as it hopes, which would require a marked leap forward. Peter Wolynes, a professor of chemistry at Rice University, says he is sure it will be an improvement, but that currently "the engineering is at the level of the Romans, not of the Renaissance, in terms of how protein engineering is done." Alan Fersht, an emeritus professor of chemistry at the University of Cambridge who specializes in protein engineering, says that if the interactions between all molecules in the cell and the fluctuations in their structure could be calculated accurately, where molecules bind could be predicted, and that would revolutionize drug development. He says he is optimistic this will eventually happen.

SovAI was established by the UK government earlier this year with a mandate to invest in British startups capable of becoming indispensable components of the global AI supply chain. It has roughly $800 million at its disposal and can offer portfolio companies access to the UK's national AI supercomputer. With its investment, Potential Sciences joins AI drug discovery companies Basecamp Research and Isomorphic Labs, and materials discovery startup CuspAI. Joséphine Kant, head of ventures at SovAI, has joined the startup's board. She says approaching drug discovery from the physics of chemistry sets the company apart, and that computing access lets it work at a scale that would be very difficult for a small startup otherwise.

Wired lists the obstacles. AI drug discovery is a crowded market, and the startup faces steep competition from many entrepreneurs and academics. Faster identification of candidates does not remove the cost and time of testing safety and efficacy, so new treatments will keep taking a long time to reach the market. Wolynes puts it this way: the cost of designing a drug is nothing compared to figuring out what it will do to mice, to really sick patients and to normal people, so design is in some sense not the rate-limiting step. There is also a misuse risk. Helal says that if tools can be designed to perfectly minimize toxicity, it is a bit flip to maximize it as well. The startup has not settled on a way to generate revenue, nor decided whether to make its models freely available given that risk.

Key facts

Potential Sciences, a British startup, launched on November 5 and says it has seed funding in the tens of millions of dollars; the precise figure and valuation were not disclosed.
Investors named are SovAI, the UK government's AI venture fund with roughly $800 million at its disposal, and Twin Path Ventures, among others; SovAI's Joséphine Kant has joined the board.
The plan is to simulate atoms, small molecules, amino acids and proteins using physics, build far larger synthetic datasets than the Protein Data Bank, and train AI models to predict drug behaviour with less lab work.
Rice University's Peter Wolynes expects an improvement but says protein engineering is still at the level of the Romans, and that designing a drug is not the rate-limiting step compared with testing it.
The startup has no settled revenue model and has not decided whether to release its models openly, given the potential for misuse.

Why it matters

The pitch tackles a data bottleneck. Wired says many AI drug discovery labs build on one public database, the Protein Data Bank, which Helal considers too small to train models that reliably speed up the search for drug candidates. Potential Sciences wants to replace that limit with far larger synthetic datasets from physics simulation, working up from atoms to small molecules to amino acids and proteins. If it works, Helal says, cheaper experimentation could make drugs for rare, previously untreated conditions economical to pursue. It also fits a wider hope that drug discovery will be among the first scientific fields transformed by AI, a view Wired ties to figures such as Nobel laureate Demis Hassabis.

Who it affects

Pharmaceutical companies are the ones Helal says could find it economical to pursue drugs for rare conditions that have gone untreated, if experimentation gets cheaper. Patients could eventually see more tailored treatments, though the source gives no timescale. The story also touches the UK's AI industrial policy: SovAI was set up to back British startups that could become indispensable parts of the global AI supply chain, and Potential Sciences joins Basecamp Research, Isomorphic Labs and CuspAI in its portfolio. Rival AI drug discovery companies, entrepreneurs and academics face a new entrant working on the same bottlenecks.

How to use it

There is nothing to use yet. The startup has not decided how it will earn revenue from its technologies, or whether to make its models freely available. For readers in the field, the practical signal is the approach itself: simulate chemistry from physics to generate training data, rather than relying only on experimentally derived structures.

How solid is it

The funding and approach come from the company and are reported by Wired; the amount is described only as tens of millions of dollars, with no exact figure or valuation. No benchmark results, accuracy figures or working models are reported. Wired itself calls the central question open: whether the startup can predict drug-candidate binding to disease-related proteins as accurately as it hopes, which would need a marked leap forward. Outside academics are supportive but measured. Wolynes is sure it will be an improvement, and Fersht is optimistic the underlying calculation will eventually be possible.

Risks and caveats

Competition is steep: AI drug discovery is a crowded market, and the UK government's backing does not change that. Better candidate identification does not remove the cost and time of testing safety and efficacy, so treatments will still take a long time to reach the market. Wolynes says the cost of designing a drug is small next to learning what it does in mice, in very sick patients and in normal people. Dual use is a concern too: Helal acknowledges that tools able to minimize toxicity could be flipped to maximize it, and the company has not decided whether to release its models freely. Revenue is also unsettled.

“The cost of designing a drug is nothing compared to figuring out what it will do to mice; to really sick patients; to normal people”

— Peter Wolynes, professor of chemistry at Rice University