Table of Contents
A newly launched artificial intelligence startup called Potential Sciences aims to bypass traditional laboratory experimentation by simulating chemical interactions using basic physics. The company plans to construct massive synthetic datasets to train predictive AI models capable of identifying targeted drugs for specific diseases.
The Approach and Funding
Potential Sciences launched on November 5 with seed funding amounting to tens of millions of dollars. Investors include SovAI, which is the UK government’s AI venture capital fund, and Twin Path Ventures, alongside other backers. The founders chose not to disclose the precise funding figure or company valuation.
Founder and CEO Hatem Helal told WIRED that the startup wants failing molecules to fail quickly. By improving behavioral predictions during interactions with human biology, the company hopes to accelerate the development of new treatments and reduce experimentation costs. Helal stated that this approach could make it economical for pharmaceutical companies to pursue treatments for rare conditions that currently lack options, potentially enabling bespoke medicines tailored to individual human biology.
Industry Challenges and Simulation
The core challenge for Potential Sciences is whether it can predict drug candidate binding accuracy as effectively as hoped. Peter Wolynes, a chemistry professor at Rice University, noted that current protein engineering engineering remains at an early developmental stage. Meanwhile, Demis Hassabis and other industry leaders have long anticipated that AI advancements will drive major scientific breakthroughs.
Most AI drug discovery laboratories currently build models using a single public database containing three-dimensional atomic structures known as the Protein Data Bank. Helal explained that this dataset remains too small for training advanced models, noting that potential molecular combinations number around 1060. To solve this, Potential Sciences generates synthetic data by simulating interactions between chemical compounds, building pools large enough to train advanced prediction systems.
Academic experts see promise in this physics-based methodology. Alan Fersht, an emeritus chemistry professor at the University of Cambridge, stated that accurately calculating molecular interactions and structural fluctuations would revolutionize drug development, adding that he is optimistic about the approach.
Government Backing and Market Competition
Through its investment from SovAI, Potential Sciences joins other portfolio companies like Basecamp Research, Isomorphic Labs, and CuspAI. The UK government launched SovAI earlier this year with approximately $800 million to invest in startups that can become key components of the global AI supply chain, while also providing portfolio companies with access to the national AI supercomputer.
Joséphine Kant, head of ventures at SovAI and a new board member at Potential Sciences, told WIRED that the startup’s physics-based angle sets it apart, and the computing access allows them to operate at a scale difficult for small startups. However, the market remains crowded with competitors tackling similar bottlenecks from different angles.
Experts also point out that faster identification does not remove the time and cost needed for safety and efficacy testing. Wolynes observed that design costs pale in comparison to testing on animals and patients, meaning new treatments will still require significant time to reach the market.
Furthermore, Potential Sciences acknowledges the risk that tools designed to minimize toxicity could be inverted to maximize it. The startup has not yet finalized its revenue model or decided whether to make its models publicly available due to safety concerns, though Helal maintains that solving drug discovery remains a vital challenge.
