TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Hugging Face has published a guide to using RFD4-Proteina, an open protein-design model co-developed by NVIDIA and the University of Washington’s Institute for Protein Design, through browser-based Spaces. NVIDIA is sponsoring free access to the dedicated GPU Space from Oct. 21 to Nov. 4, 2026; the generated designs are computational results and require experimental validation.
Hugging Face has published a browser-based guide for using RFD4-Proteina to generate protein sequences and structures, while NVIDIA is sponsoring free access to a dedicated version of the Space from Oct. 21 through Nov. 4, 2026. The model is an open BioNeMo project co-developed by NVIDIA and the University of Washington’s Institute for Protein Design, and the workflow lets users set molecular design constraints and inspect generated results without installing software.
RFD4-Proteina combines protein sequence and all-atom structure generation in one model. Hugging Face’s report says the approach is intended to reduce handoffs between separate models for backbone, sequence and detailed structure design. It can generate proteins in molecular contexts involving other proteins, small molecules, nucleic acids, glycans, metal ions and modified amino acids.
Users can try the workflow in two Hugging Face Spaces: a dedicated RFD4-Proteina Space using dedicated GPUs, and a ZeroGPU Space that runs on shared capacity. Both use AtomWorks Studio, a browser-based molecular viewer and annotation editor. A user begins with a target structure, marks a region and specific residues or atoms for the designed protein to engage, then generates and inspects a design in the viewer.
The report says outputs can be downloaded as annotated CIF files, including the target and binder-target complex, for further computational analysis. The guide demonstrates a PD-L1 example and explains how changing the random seed or the model’s low, medium or high “thinking” setting can produce different designs. Generated structures are computational predictions; the report says experimental testing is needed to confirm whether a proposed binder actually binds.
A Shorter Route From Target to Design
Putting sequence and all-atom structure generation into one model may make it easier for researchers to test constrained protein-design ideas without stitching together several model outputs. The browser interface also lowers the setup burden for an initial trial: users can supply a target, specify engagement points and inspect candidate structures before moving to local tools.
The practical value remains a matter for users to establish. A model-generated structure that appears to contact specified hotspots is not evidence of binding in a laboratory. Researchers still need to assess candidates with appropriate computational checks and, where warranted, experimental characterization. The sponsored access period provides a limited opportunity to try the dedicated GPU workflow, while the shared Space offers a continuing route subject to Hugging Face quotas.
protein structure modeling software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How RFD4-Proteina Builds Designs
The model pairs a flow-matching model with an autoencoder, according to Hugging Face’s report. The autoencoder represents amino acids, including modified residues, as well as sugars and DNA or RNA nucleotides, using an anchor-atom coordinate and a learned vector that encodes chemical identity and local geometry. The flow model generates coordinates and latent representations together; a decoder then expands them into sequences and all-atom structures.
This design allows the model to generate a protein around a supplied constraint, such as a target a binder should engage or a catalytic motif the protein should contain. In a benchmark described by Hugging Face, RFD4-Proteina at high thinking produced 76.40 structurally distinct designs per GPU-hour meeting the report’s stated computational criteria across 24 protein-binding targets. The report compared that result with 13.24 for Proteina-Complexa, the next-strongest method in that benchmark. These are benchmark results under specified computational filters, not evidence that the designs were experimentally successful.
As an affiliate, we earn on qualifying purchases.
Binding Still Needs Laboratory Testing
The report does not establish that the example designs, or designs generated by users, bind their intended targets in experiments. It describes computational filters and benchmark throughput, but the supplied material does not give experimental success rates, laboratory results for the PD-L1 example, or evidence that a particular generated candidate is ready for use.
Other practical details are also not specified in the report excerpt, including the amount of capacity available during the sponsored period and the precise quota limits for ZeroGPU users. Access to a generation interface should not be taken as a guarantee of a particular number of runs or of experimentally useful results.
As an affiliate, we earn on qualifying purchases.
Access After the Sponsored Window
NVIDIA’s sponsored free access to the dedicated RFD4-Proteina Space is scheduled to end on Nov. 4, 2026. After that date, Hugging Face says the dedicated Space will no longer be free, while the ZeroGPU Space will remain free subject to account quotas. The report does not give a price for continued dedicated access.
Users who generate candidates can download the annotated CIF files and compare runs by changing settings such as the seed and thinking level. Further computational assessment, followed by experimental characterization where appropriate, would be needed to determine whether any candidate binds or has useful properties.
computational protein modeling kit
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is RFD4-Proteina?
RFD4-Proteina is an open protein-design model co-developed by NVIDIA and the University of Washington’s Institute for Protein Design. Hugging Face says it generates protein sequences and all-atom structures together around specified molecular constraints.
Can I use it without installing software?
Yes. Hugging Face provides a dedicated Space and a ZeroGPU Space that run in a browser. Both use AtomWorks Studio for viewing and annotating molecular inputs.
When is the dedicated Space free?
NVIDIA is sponsoring free access from Oct. 21 through Nov. 4, 2026. The report says the ZeroGPU Space remains free after that period, subject to Hugging Face account quotas.
Do the generated designs prove that a protein binds its target?
No. The designs are computational results. The report says experimental validation is needed to confirm binding.
What does the thinking setting change?
The low, medium and high settings control how much the model refines each design during generation. Hugging Face says more refinement can increase the likelihood of passing in-silico filters, while requiring more compute per design.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
