Nvidia and UW Release Open Protein Design Model RFD4-Proteina

RFD4-Proteina pairs protein generation with search and refinement, giving researchers a way to explore candidates against specific molecular targets and constraints. The open model, developed by the University of Washington School of Medicine’s Institute for Protein Design (IPD) with Nvidia, could help move promising designs toward lab testing in less time.

Built on Nvidia’s NVIDIA BioNeMo artificial intelligence (AI) platform, RFD4-Proteina uses biomolecular reasoning to generate proteins for defined functions. It refines a candidate’s sequence and structure alongside the structures of molecules it interacts with, allowing changes in one component to inform adjustments to the others. Researchers can then search and rank designs using molecular inputs, constraints and scoring objectives.

The developers report that the model produced candidates five to 10 times faster and achieved design success rates five to 10 times higher than earlier models. Nvidia says lab bench evaluations found higher computational design success and throughput than the tested baselines across four benchmark suites: protein binding, small-molecule binding, enzyme scaffolding and DNA binding. These results describe the reported evaluations; the practical value lies in helping teams prioritize which designs to test experimentally.

Applications from drug discovery to materials

Potential uses include proteins that interrupt disease processes, diagnostic proteins that detect signs of disease, and enzymes designed to break down pesticides or polymers. IPD director David Baker, a Nobel laureate and UW Medicine biochemistry professor, sees the model as part of a shift from finding naturally occurring proteins to designing new ones for specific tasks.

“RFD4-Proteina expands what scientists can create, from proteins that could advance drug discovery to enzymes that break down pesticides,” Baker said. He also pointed to possible applications in medicine, sustainability, materials and biotechnology, including improved medicines and new ways to make bioplastics.

Early projects at IPD offer examples of the model’s pace. Meg Lunn-Halbert, a UW doctoral candidate in biochemistry and biological physics, structure and design, said her team controlled which glycans—sugars involved in immune recognition—attached to proteins, completing the work in under a week rather than months.

UW bioengineering doctoral candidate and National Science Foundation fellow Seth Woodbury said he designed a pesticide-degrading enzyme in a few days that performed many times better than his previous best, which had taken months to develop.

Open access and research workflow integration

The model’s name connects two earlier systems: IPD’s RosettaFold Diffusion, a guided diffusion model for generating proteins, and Nvidia’s Proteina-Complexa, which designs protein binders for protein and small-molecule targets. Proteina-Complexa combines inference-time optimization with a pretrained flow-based generative model built on Nvidia’s La-Proteina atomistic protein design model.

RFD4-Proteina is released as open source under the BSD-3 license. It is available through Amazon Bio Discovery, Anthropic’s Claude Science, Benchling, Databricks, Google Cloud, Hugging Face, Nebius, OpenAI and Tamarind Bio, which provide routes for integrating it into agent, cloud and scientific computing workflows.

The project was led at IPD by UW PhD students Nathaniel Corley, Maxwell Kazman and Rafi Brent, along with visiting University of Oxford student Kieran Didi. Its funding came from the Howard Hughes Medical Institute, Gates Foundation (INV-092109), Grantham Foundation, Fund for Science and Technology, Schmidt Sciences, the OpenFold consortium (part of OMSF), Rosetta Commons and other IPD supporters. The Gates Foundation website lists a $25,748,619 commitment under a grant dated November 2025. The Jen-Hsun and Lori Huang Foundation also provided a computational grant that UW said was critical to development.

Baker shared the 2024 Nobel Prize in Chemistry for computational protein design with Sir Demis Hassabis and John Jumper. RFD4-Proteina’s open release gives research teams a new platform for testing whether faster computational design can translate into more useful proteins at the bench.

The post Nvidia, UW Institute for Protein Design Unveil Open-Source Protein Design AI Model appeared first on GEN - Genetic Engineering and Biotechnology News.