University of Virginia researchers have created a trio of artificial intelligence tools that design drug molecules for moving, flexible protein targets. The open-access platform could help scientists worldwide develop better treatments faster and at lower cost.
Developing a new medicine today can take more than a decade, cost billions of dollars and still end in failure. A team at the University of Virginia School of Medicine believes artificial intelligence can change that equation.
Nikolay V. Dokholyan, a professor of neurology at the University of Virginia School of Medicine, and colleagues have built a suite of AI tools designed to rethink how drugs are discovered and engineered. The platform, made up of three interconnected programs called YuelDesign, YuelPocket and YuelBond, is meant to design drug molecules that fit their targets in the body far more realistically than most current methods.
At the heart of the system is YuelDesign, which uses a cutting-edge form of AI known as diffusion models to generate new drug candidates. Unlike many existing tools, it does not treat proteins as rigid objects. Instead, it tries to design drugs for proteins that bend, flex and change shape as they interact with other molecules.
Dokholyan compares the challenge to making a key for a lock that never sits still.
“Think of it this way: Other methods try to design a key for a lock that’s sitting perfectly still, but in your body, that lock is constantly jiggling and changing shape. Our AI designs the key while the lock is moving, so the fit is much more realistic,” Dokholyan said in a news release.
That moving lock is not just a metaphor. In real biology, proteins often shift their structure when a drug binds to them, a phenomenon scientists call induced fit. If a computer model ignores that flexibility, a drug that looks perfect on screen can fail in the lab or in patients.
Those failures are expensive. The average cost of bringing a new drug to market has been estimated at $2.6 billion or more, and nearly 9 out of 10 drug candidates fail once they reach human testing. A major reason is that the molecules do not bind to their intended targets in the body as predicted, or they bind in the wrong place and cause harmful side effects.
Artificial intelligence has already begun to speed up parts of drug discovery, from predicting protein structures to screening huge libraries of molecules. But Dokholyan’s team set out to push beyond tools that work with static, snapshot-like models of proteins.
YuelDesign tackles that by generating both the shape of the protein pocket where a drug will bind and the small molecule that fits into it at the same time. As the AI iterates, the protein and the potential drug are allowed to adapt to each other, more closely mimicking what happens inside a cell.
A second tool, YuelPocket, focuses on a different but equally crucial step: finding the right place on a protein for a drug to latch on. It uses graph neural networks, a type of AI that can understand complex relationships, to map out binding pockets even on protein structures predicted by other tools such as AlphaFold.
That distinction matters, according to first author Jian Wang, a member of the Dokholyan lab.
“Most existing AI tools treat the protein as a frozen statue, but that’s not how biology works. Our approach lets the protein and the drug candidate evolve together during the design process, just as they would in the body,” Wang said in the news release.
In tests, the researchers applied YuelDesign to a well-known cancer-related protein called CDK2, a common target in oncology drug development.
“We showed, for example, that when designing molecules for a well-known cancer-related protein called CDK2, only YuelDesign could capture the critical structural changes that happen when a drug binds,” Wang added.
The third component, YuelBond, checks the chemistry of the designed molecules, making sure the proposed bonds and structures are realistic and stable.
Together, the three tools are intended to form an end-to-end pipeline: identify a promising pocket on a protein, design a molecule that fits a moving target and verify that the chemistry holds up.
Mapping out protein pockets is central to virtually every aspect of modern development, from designing first-in-class drugs to repurposing existing medicines for new diseases. By improving how those pockets are identified and how molecules are tailored to them, the UVA team hopes to boost the odds that a drug candidate will work as intended.
Dokholyan believes the impact could be especially important for diseases where traditional approaches have repeatedly fallen short.
“This could make a real difference for patients with cancer, neurological disorders and many other conditions where we desperately need better drugs targeting these wiggly proteins but keep hitting dead ends,” he said.
Beyond improving success rates, the researchers see their work as a way to reduce costs and shorten timelines across the drug pipeline.
“Our ultimate goal is to make drug discovery faster, cheaper and more likely to succeed, so that promising treatments can reach patients sooner,” Dokholyan added.
In a move aimed at broadening the impact of their work, the team has released YuelDesign, YuelPocket and YuelBond as free tools for researchers worldwide. Dokholyan noted that he wants to democratize access to advanced drug design technology.
“We’ve made all of our tools freely available to the scientific community. We want researchers anywhere in the world to be able to use them to tackle the diseases that matter most to their patients,” he added.
The UVA scientists have detailed the development and testing of the tools in peer-reviewed papers in PNAS, JCIM and Science Advances.
For now, YuelDesign and its companion tools are research platforms, not products. The next steps will involve applying them to more targets, collaborating with experimental labs and, eventually, seeing whether AI-designed molecules can advance into clinical testing.
If they do, the UVA team’s vision is clear: a future where designing a drug for a shifting, complex protein is less a game of trial and error and more a precise, data-driven process that brings new treatments to patients faster.
Source: UVA Health