New AI diffusion models to speed up drug development

Scientists at the University of Virginia School of Medicine have developed a new method of drug design, utilising AI diffusion models to identify and adapt to protein movements.

A new suite of artificial intelligence-powered tools, called YuelDesign, YuelPocket and YuelBond have been created by a team at the University of Virginia School of Medicine to make the drug development process more efficient, as well as identifying new ways to make existing medicine more effective.

The average cost of developing a new drug is estimated to fall anywhere from several hundred million to over $2.6 billion and continues to grow.  Around 90% of new drugs also fail to pass human testing trials.

This is due to several factors, one of the most significant being the unpredictable reactions of the drug’s molecules with their targets in the body, where treatments can end up having negligible or even negative effects.

Designing drugs that can respond to the shifting shapes of proteins

Proteins in the body are notorious for “induced fits”, where they change shape after binding with a drug.

This phenomenon is traditionally difficult to predict on computerised models, but advanced AI “diffusion models” designed in the project can generate the protein pocket structure and the small molecule that will fit into it, allowing for adaptive design.

“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. “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.”

The three tools work together to design the optimum shape:

• YuelDesign uses the diffusion model to design tailored drug molecule shapes
• YuelPocket uses graph neural networks to identifies precisely where a drug molecule can attach on a protein
• YuelBond ensures the chemical bonds in designed molecules are accurate

“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,” said researcher Dr. Jian Wang. “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.”

Speeding up the drug development process with AI

The team hope that their AI tools will make drug development processes more cost-effective and reach patients faster, hoping to “democratize” drug discovery.

“Our ultimate goal is to make drug discovery faster, cheaper and more likely to succeed, so that promising treatments can reach patients sooner,” Dokholyan said, adding: “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.”