UVA Scientists Unveil "Yuel" AI Suite to Accelerate Dynamic Drug Discovery

University of Virginia researchers develop Yuel AI tools to simulate flexible proteins, accelerating drug discovery and reducing pharmaceutical R&D costs.

By: AXL Media

Published: Apr 9, 2026, 11:07 AM EDT

Source: Information for this report was sourced from UVA Health System and PNAS.

Revolutionizing the "Lock and Key" Model

A research team led by Nikolay V. Dokholyan, PhD, from UVA’s Department of Neurology, has introduced a groundbreaking approach to molecular design that acknowledges the "wiggly" nature of human biology. While traditional drug discovery often treats protein targets as rigid locks, Dokholyan’s AI suite designs the "key" (the drug) while the "lock" (the protein) is in motion. This method, detailed in findings published in journals such as PNAS and Science Advances, accounts for "induced fit"—the phenomenon where proteins change shape upon binding. By simulating these dynamic interactions, the tools allow drug candidates and protein targets to evolve together during the design process, creating more realistic and effective medical interventions.

The Centerpiece: YuelDesign and Diffusion Models

At the heart of the suite is YuelDesign, which leverages advanced AI "diffusion models" to generate both the small molecule drug and the specific protein pocket structure simultaneously. This co-adaptive design ensures that the drug fits the target precisely, even as the protein flexes and shifts. During testing on CDK2, a well-known cancer-related protein, researchers demonstrated that YuelDesign was the only tool capable of capturing the critical structural changes that occur during binding. This level of precision is intended to overcome a major hurdle in pharmaceutical science: the 90% failure rate of new drugs in human testing, often caused by poor binding or unforeseen side effects.

Auxiliary Tools: YuelPocket and YuelBond

Complementing the design engine are two specialized auxiliary tools that streamline the discovery process. YuelPocket utilizes graph neural networks to identify the exact coordinates on a protein where a drug can attach, even when working with predicted structures from platforms like AlphaFold. Unlike previous tools that focused only on local structural features, YuelPocket integrates global interaction patterns to provide a more accurate map for binding. Meanwhile, YuelBond serves as a chemical validator, ensuring that the designed molecules maintain accurate chemical bonds and structural integrity, preventing "paper-only" successes that would fail in a laboratory setting.