A team of scientists at the University of Virginia School of Medicine has created a breakthrough artificial intelligence approach that could fundamentally change how new drugs are discovered and developed. The researchers developed a suite of AI-powered tools called YuelDesign, YuelPocket, and YuelBond that work together to design drug molecules with unprecedented accuracy by accounting for how proteins actually move and change shape in the human body, rather than treating them as static targets .

Why Does Protein Movement Matter So Much in Drug Design?

The challenge that has plagued drug development for decades comes down to a fundamental mismatch between how scientists design drugs and how those drugs actually work in the body. Traditional drug design methods treat proteins like frozen statues, but in reality, proteins are constantly flexing and shifting shape. When a drug molecule tries to bind to a protein target, the protein often changes its structure in response, a phenomenon scientists call "induced fit." This dynamic interaction is critical to whether a drug will actually work .

"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," said Nikolay V. Dokholyan, PhD, of UVA's Department of Neurology.

Nikolay V. Dokholyan, PhD, Department of Neurology, University of Virginia

The consequences of ignoring this flexibility are severe. Drugs that look promising on a computer screen often fail when tested in humans because they don't actually bind correctly to their targets. This failure rate is staggering: almost 90% of new drugs fail during human testing . Additionally, the average cost to develop a single new drug has reached or exceeded $2.6 billion, making failures extraordinarily expensive .

How Do These New AI Tools Actually Work?

The centerpiece of this breakthrough is YuelDesign, which uses advanced artificial intelligence technology called diffusion models to simultaneously generate both the protein pocket structure and the small molecule drug candidate that fits into it. Rather than designing a drug molecule first and then hoping it fits a rigid protein target, YuelDesign allows both the protein and the drug to adapt to each other during the design process, just as they would in the actual human body .

The companion tools serve specific but equally important functions. YuelPocket uses graph neural networks, a type of AI architecture designed to understand relationships between connected objects, to identify precisely where on a protein a drug should bind. This is critical because most existing AI tools treat the protein as a frozen structure, missing the dynamic nature of real biology. YuelBond ensures that the chemical bonds in designed molecules are accurate and stable .

To demonstrate the power of this approach, the research team designed molecules for CDK2, a well-known cancer-related protein. Their results showed that only YuelDesign could capture the critical structural changes that happen when a drug binds to the protein, something existing methods completely missed .

Steps to Accelerate Drug Discovery Using These Tools

• Identify Protein Targets: Researchers can use YuelPocket to map out exactly where on a protein a drug should bind, even when working with predicted protein structures from tools like AlphaFold, eliminating guesswork about binding sites.
• Design Flexible Drug Molecules: YuelDesign generates drug candidates while accounting for protein movement and shape changes, creating molecules that will actually work in the dynamic environment of the human body rather than in static computer models.
• Validate Chemical Bonds: YuelBond checks that the chemical bonds in designed molecules are accurate and stable, preventing the creation of molecules that look good theoretically but fall apart in practice.
• Evaluate Existing Drugs: The tools can rapidly assess whether existing drugs might work for new purposes, accelerating the repurposing of already-approved medications for different diseases.

What Could This Mean for Patients?

The potential impact extends far beyond academic interest. Dokholyan and his team believe their technology could reduce drug development costs, improve the success rate of new drug candidates, and accelerate how quickly new treatments reach patients. The research team has described their work in papers published in prestigious scientific journals including PNAS, JCIM, and Science Advances .

Perhaps most importantly, the researchers have made all of their tools freely available to the scientific community. This democratization of drug discovery means that researchers anywhere in the world can use these tools to tackle the diseases that matter most to their patients, rather than having access limited to well-funded pharmaceutical companies .

"Our ultimate goal is to make drug discovery faster, cheaper and more likely to succeed, so that promising treatments can reach patients sooner," explained Dokholyan, adding that he wants to "democratize" drug discovery by putting new tools at scientists' fingertips.

Nikolay V. Dokholyan, PhD, Department of Neurology, University of Virginia

The research team, which includes Dr. Jian Wang, Dong Yan Zhang, Shreshty Budakoti, and Dokholyan, received support from the National Institutes of Health, the National Science Foundation, the Huck Institutes of the Life Sciences, and the Passan Foundation. Notably, the scientists have no financial interest in the work, ensuring that their research is driven purely by the goal of advancing medicine .

This breakthrough represents a significant step forward in addressing one of medicine's most pressing challenges: how to design drugs that actually work in the complex, dynamic environment of the human body. By finally accounting for the way proteins move and change shape, these AI tools could help bring better treatments to patients with cancer, neurological disorders, and many other conditions where current options have repeatedly failed.