See all news
Image: UW Medicine
UW Medicine
AI

Proteins on demand! An open model from NVIDIA and UW designs molecules more precisely

The University of Washington and NVIDIA unveiled RFD4-Proteina, an open generative model that helps design proteins with specific functions. Its creators report gains in speed and design success over earlier methods, along with experimental tests of some results. The announcement, made on October 21, points to applications in medicine, diagnostics and materials, although computational proposals still need laboratory verification.

Designing a protein is not just a matter of writing a sequence of amino acids. How it folds, which molecules it interacts with and whether it ultimately performs the intended function also matter. UW Medicine's Institute for Protein Design and NVIDIA unveiled RFD4-Proteina to address these dimensions simultaneously. The model builds on RosettaFold Diffusion and Proteina-Complexa, and can devote more processing to especially difficult problems.

According to the university, computational tests showed design success rates five to ten times higher, with candidate generation also five to ten times faster than the comparison models. The researchers also reported laboratory assays that confirmed predicted behaviors in some designs. Among the examples shared are enzymes capable of breaking down pesticides and the control of sugar modifications on proteins.

The resource is distributed openly under a BSD-3 license, which makes it easier for other laboratories to study and adapt it. The advance is not equivalent to an approved drug, nor does it guarantee that every generated protein will work. Its significance lies in reducing early iterations and broadening the experimental exploration of new molecules.

At a glance

WhatUW Medicine and NVIDIA released RFD4-Proteina, an open model for generating and refining protein designs that could be used in biomedical and materials research.
WhenThe announcement was published on October 21, 2026; the computational comparisons and experimental assays described were carried out before the announcement.
WhereThe work comes from the Institute for Protein Design at the University of Washington, in the United States, together with NVIDIA researchers.
HowThe system jointly adjusts protein sequences and structures, takes into account molecules that interact with them, and uses additional processing when a problem requires more exploration.
WhyLaboratories need to explore enormous molecular design spaces; a faster, open tool can make it easier to test hypotheses and candidates.

Build your digital presence

Start today: write to us and we’ll reply within 24 hours.