British startup uses physics simulations to speed up drug discovery

Simulating the building blocks

Potential Sciences is a new British startup attempting to accelerate the slow process of drug design using artificial intelligence. The company believes that applying the basic principles of physics can help simulate how atoms and molecules interact with one another. By modeling these interactions, they hope to identify targeted drugs for specific diseases without relying solely on physical lab tests.

Current drug discovery often relies on the Protein Data Bank, a public database of 3D atomic structures compiled over several decades. However, the startup's founder, Hatem Helal, argues that this dataset is too small to train advanced AI models effectively. He notes that there are many ways to assemble atoms into molecules, making the search for viable drugs a massive computational challenge.

To solve this, the company is constructing much larger datasets made of synthetic data. This data is produced by simulating the behavior of different chemical compounds at a fundamental level. The theory is that training AI on these massive, simulated pools will allow for better prediction systems that can navigate the vast space of possible molecules.

A physics-first approach

What sets Potential Sciences apart from other AI drug discovery firms is its focus on the physics of chemistry. Instead of just looking at existing biological data, the startup uses physics to predict how molecules will behave when they meet human biology. This approach is intended to help molecules that are unlikely to work in a clinical setting fail faster during the design phase.

The company's strategy aims to provide better predictors of molecular behavior. By understanding these interactions more deeply, researchers can prune away unpromising candidates early in the process. This could significantly reduce the cost and time associated with traditional experimentation, which is currently a major bottleneck in the pharmaceutical industry.

Experts acknowledge the difficulty of this task, noting that predicting how drug candidates bind to disease-related proteins requires a significant leap forward. While some academics are optimistic about the simulation-based approach, others point out that current protein engineering is still in its early stages. The goal remains to reach a level of accuracy that can reliably replace some of the most tedious manual laboratory work.

Funding and government backing

Potential Sciences launched on November 5 and secured seed funding in the tens of millions of dollars. The investment came from Twin Path Ventures and SovAI, which is the UK government's AI venture capital fund. SovAI was established earlier this year to invest in British startups that can become essential parts of the global AI supply chain.

The UK government's involvement provides the startup with significant resources, including access to the national AI supercomputer. This level of computing power is necessary to run the complex physics simulations required to generate synthetic data at scale. This support gives the small startup a competitive edge in a crowded market filled with other AI-driven drug discovery firms.

The investment also carries a geopolitical dimension, as the UK seeks to build a robust domestic AI infrastructure. By supporting companies like Potential Sciences, the government aims to provide the country with leverage in the global technology landscape. The startup joins other notable firms like Basecamp Research and Isomorphic Labs in the race to transform scientific discovery through AI.

The path to bespoke medicine

If the startup succeeds, the primary goal is to make it economically feasible for pharmaceutical companies to pursue drugs for rare conditions. Many of these diseases have gone untreated because the cost of traditional development is too high for small patient populations. By reducing the cost of the initial design phase, more niche treatments could become viable.

Hatem Helal envisions a future where medicines are one-off molecular combinations tailored to a person's unique biology. This concept of bespoke treatment relies on the ability to accurately predict how a specific molecule will interact with a specific human body. The startup hopes to move away from a one-size-fits-all model toward highly personalized medicine.

However, the company faces several hurdles, including the fact that identifying a candidate does not eliminate the need for safety testing. Testing a drug on mice or human subjects remains a slow and necessary part of the process. Additionally, the startup must navigate the ethical risks of its technology, as tools designed to minimize toxicity could theoretically be used to maximize it.

Challenges and next steps

Potential Sciences has not yet finalized its revenue model or decided whether its models will be made freely available. The company must balance the desire to solve a high-impact problem with the risks of potential misuse of its technology. Despite these unknowns, the founders remain committed to using AI to address the fundamental problems of drug discovery.

The startup enters a competitive field where many other entrepreneurs and academics are trying to solve the same bottlenecks. While the physics-based approach is a distinct strategy, the company will need to prove that its predictions are accurate enough to be useful. Success would mean a marked improvement over current methods, even if it does not completely replace laboratory testing overnight.

The broader scientific community watches these developments closely as a test case for AI's impact on science. Industry leaders have suggested that AI could lead to momentous breakthroughs in the same way it has transformed mathematics. Potential Sciences aims to be at the forefront of this transition, turning drug discovery into a more predictable and efficient search problem.