UVA AI Tools Design Drugs for Flexible Proteins
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Nikolay V. Dokholyan and colleagues at UVA's Department of Neurology developed the YuelDesign suite (YuelDesign, YuelPocket, and YuelBond) — AI tools that use diffusion models to design drug molecules custom-fit to flexible protein targets.
- YuelDesign simultaneously generates the protein pocket structure and the small molecule that slots into it, letting both adapt to each other during design rather than treating proteins as rigid snapshots that miss the 'induced fit' phenomenon.
- YuelPocket uses graph neural networks to pinpoint where on a protein a drug should bind, and works on predicted protein structures generated by tools like AlphaFold.
- The approach targets a costly failure point: the average new drug costs an estimated $2.6 billion to develop, and roughly 90% of candidates fail in human testing, often because rigid models miss how proteins reshape when drugs bind.
- In testing on the cancer-related protein CDK2, only YuelDesign captured the critical structural changes that occur when a drug binds, according to researcher Dr. Jian Wang.
- Dokholyan's team published the work in three journals — Proceedings of the National Academy of Sciences, Journal of Chemical Information and Modeling, and Science Advances — and has made all three tools freely available to the global research community.
Why it matters: With the average new drug costing an estimated $2.6 billion and roughly 90% of candidates failing in human testing, even modest improvements in binding prediction could reshape the economics of drug development. By releasing YuelDesign, YuelPocket, and YuelBond as open-source tools, Dokholyan's team gives any researcher worldwide free access to flexible-protein modeling previously limited to well-funded labs, potentially lifting success rates for cancer, neurological, and other disease targets.
Ask SkimNews
