Where AI meets cell biology: designing a drug for leaky blood vessels

The problem: a missing drug

The leaking is caused by excess pressure on endothelial cells — the cells lining the inside of blood vessels. This pressure is regulated by proteins called Rho-GTPases, which can switch between an activated (“on”) and deactivated (“off”) state. In patients with leaky blood vessels, this balance is disrupted. Researchers are therefore looking for a drug molecule capable of flipping the protein from “off” back to “on.”

Machine learning as a design tool

The project brings together Bernd Ensing and Tati Fernández Ibáñez of the Van ‘t Hoff Institute for Molecular Sciences (HIMS) and Jaap van Buul of the Swammerdam Institute for Life Sciences (SILS) and Amsterdam UMC.

Rather than screening existing compounds in a lab, the team uses a computer simulation of the target protein and applies generative AI to design entirely new candidate molecules. As Ensing, Professor of AI for Chemistry, explains: “You can train a generative AI on a database of a million drug molecules. At some point, it starts recognising patterns in typical drug molecules. You can then generate millions of new ones.”

From there, the team uses a technique called Bayesian optimisation to iteratively improve candidates — making variations on the most promising molecules and refining them step by step toward an optimal design. The most promising results will then be tested in the lab at Amsterdam UMC.

Using already-approved drugs as a starting point is another deliberate strategy: it allows the team to skip several standard steps in drug development, accelerating the path from discovery to potential clinical use.

Unexplored ground

Ensing is candid about the novelty of the approach: “This is still just an idea, it has never been tried before. We’re very curious to put it to the test. There may be other groups working on this somewhere in the world, but I haven’t seen any publications on it yet.”

The project started as a UvA MMD TechHub project and has since been merged with a UvA Synergy grant, giving a PhD student four years to develop the work.

For Ensing, the interdisciplinary dynamic is as much a draw as the research itself: “We now have a really exciting team with people who are deeply passionate about things that are very different from what I normally work on. I’ll undoubtedly learn an enormous amount from them.”