How to Use RFdiffusion
Commercially Available Online Web Server
RFdiffusion guide to protein backbone generation, binder design, hotspots, motif scaffolding, partial diffusion, and sequence validation.
RFdiffusion generates protein backbones using a diffusion model. Instead of predicting the fold of an existing sequence, it proposes new structures that can be conditioned on a target protein, a structural motif, or a symmetry requirement. This makes it useful for de novo scaffold generation, binder-backbone design, motif scaffolding, and diversification around an existing structure.
On Neurosnap, task-specific settings cover binder design, motif scaffolding, partial diffusion, and unconditional generation. The Backbone Only option separates backbone generation from the accompanying ProteinMPNN sequence-design and AlphaFold2 prediction steps. Leaving it disabled allows those follow-up stages; a generated backbone alone is not a complete experimentally testable sequence design.
Choose a backbone-first design workflow
Use RFdiffusion when exploring geometry is a central part of the problem: where a binder could contact a surface, how a scaffold could hold a motif, or which oligomer arrangements could satisfy a symmetry. LigandMPNN addresses sequence design on an existing backbone, while BindCraft2 offers a different campaign-based binder workflow.
A diffusion-generated structure remains a design proposal. Foldability, binding, specificity, and expression need further evaluation, even if the backbone satisfies the requested geometry.
How RFdiffusion Works
How diffusion generates protein backbones
RFdiffusion learns to recover protein-like structures from noisy coordinates. During generation it progressively constructs a backbone while respecting the conditioning supplied for the design task. A fixed motif or target surface narrows the geometric search; sequence design and refolding then test whether an amino-acid sequence can plausibly realize that geometry. See the official RFdiffusion guide for the underlying method and design concepts.
Match the design mode to the question
| Design mode | Required context and intended output |
|---|---|
| Unconditional generation | Choose a length to explore new protein backbones without a supplied structural motif. |
| Binder design | Supply a target structure and chain, optionally a crop and hotspots, to generate a contacting binder backbone. |
| Motif scaffolding | Supply the structural motif and choose surrounding scaffold-length ranges. |
| Partial diffusion | Supply an existing structure and a selected chain or region to explore related geometries. |
| Symmetric generation | Choose the supported symmetry and order to explore oligomeric arrangements. |
Configure one coherent task. A custom contig is an alternative to the task-specific presets, not an extra constraint to combine casually with every mode. For an initial run, prefer the preset that expresses your actual experiment.
Prepare target chains, cropping, and hotspots
Upload a structure with clear chain identifiers and inspect the intended binding surface. This RFdiffusion service removes nucleic acids, heteroatoms, and waters, so do not assume a cofactor or glycan in the original upload remains part of the modeled environment.
In binder mode, specify Binder Input Chain and the desired binder-length range. Binding Pocket Residue Start and Binding Pocket Residue End can crop a large target around the site of interest; leaving both at 1 disables cropping. Preserve a record of the crop and its residue mapping.
Cropping can expose buried hydrophobic residues that attract designs to an artificial surface. Use the Hotspots selection to focus on biologically accessible residues and inspect every shortlisted binder against the full target. Include surrounding domains, oligomer partners, membrane context, and known glycosylation when assessing whether the proposed approach is physically accessible, even when they were absent from the design model.
For motif scaffolding, choose scaffold extensions that can plausibly support the motif rather than simply maximizing length. For partial diffusion, begin conservatively and evaluate how much the original geometry changes before increasing the noising level.
Timesteps, symmetry, and guiding potentials
Timesteps defaults to 50. Changing it adjusts diffusion generation, not the number of candidate sequences or a physical simulation time. Use Beta Model changes the structural design regime; compare resulting topology rather than assuming more beta structure is always desirable.
The symmetry choices exposed here include none, auto, cyclic, and dihedral. Choose Order deliberately and inspect the generated assembly. Symmetry is a geometric constraint and does not by itself prove an oligomer will assemble in solution.
Guiding potentials can bias compactness or contacts. Begin without them to establish a baseline, then add only potentials compatible with the selected design mode. A stronger contact objective is not a calculated binding free energy and may concentrate designs around unintended geometry.
Example: design a binder to an exposed protein domain
- Inspect the domain in the intact target and select an experimentally meaningful, accessible surface.
- Prepare the target chain and a conservative crop if necessary; map candidate hotspot residues to the uploaded structure.
- Choose binder-length bounds appropriate to the intended construct, retaining the default diffusion settings for the first comparison.
- Review backbone topology and target contacts before investing in a larger sequence-design campaign.
- Examine the sequence-design and refolding results when Backbone Only is disabled. If generating backbones only, carry suitable structures into LigandMPNN and an appropriate folding workflow.
- Compare the refolded binder with its designed backbone and inspect the predicted complex interface before selecting a diverse experimental set.
RFdiffusion troubleshooting and common questions
Why is the binder contacting the cut edge? The crop may have exposed a nonnative surface. Restore context, adjust hotspots, and inspect the complex against the full target.
Does a well-folded binder necessarily bind? No. A binder can fold correctly but adopt the wrong orientation or fail to interact experimentally. Evaluate fold recovery and interface confidence separately.
Are more contacts always better? Excessive or poorly placed contacts can reflect clashes, nonspecific hydrophobic packing, or an inaccessible surface. Structural inspection and experimental validation remain essential.
What is Neurosnap?
Neurosnap is the leading platform for bioinformatics and computational science focused on expanding access to powerful modeling and simulation tools. Because many state-of-the-art machine learning systems remain complex to install, configure, and scale, Neurosnap offers a clean, browser-based workspace that removes the burden of infrastructure management, dependency conflicts, and command-line tooling.
Built for biologists, chemists, and cross-disciplinary scientists, the platform enables advanced computational workflows without requiring expertise in software engineering or cloud architecture. Researchers can launch analyses through an intuitive interface, connect programmatically through a comprehensive API, and rely on automated resource management to scale workloads efficiently. By taking care of the underlying compute and operational complexity, Neurosnap allows teams to devote their energy to scientific progress and faster iteration. Security and data protection remain foundational principles, with clear safeguards outlined in our Terms of Use and Privacy Policy to ensure your work stays protected.
Advancing Discovery with RFdiffusion on Neurosnap
Using RFdiffusion on Neurosnap could drastically accelerate general-purpose protein diffusion design across binders, motifs, symmetries, and partial-diffusion exploration.
- One framework for many design problems: RFdiffusion covers unconditional generation, motif scaffolding, binder design, symmetry, and partial diffusion in one service.
- Scientifically meaningful controls: Target cropping, hotspots, motif ranges, symmetry, cyclic chains, and custom contigs let researchers tailor the design regime instead of accepting one generic protocol.
- Backbone-first campaign design: The model is especially useful for exploring structural hypotheses before committing to sequence and validation steps.
- No-code access to an advanced design stack: Neurosnap exposes a difficult diffusion workflow without requiring local hydra configs, model weights, or custom post-processing.
How to Use RFdiffusion on Neurosnap
To harness the capabilities of RFdiffusion, researchers can follow this streamlined workflow within Neurosnap:
- Access Neurosnap: Start by logging in to the Neurosnap website.
- Select Tool: From the list of available tools, choose RFdiffusion.
- Provide Inputs: Provide all the inputs specified within the submission panel and optionally configure the tool as desired.
- Run Tool: Submit the RFdiffusion job and Neurosnap will execute it in the cloud, automatically notifying you as soon as your results are ready.
- Review Output: Explore your results through rich visualizations, including figures, plots, and interactive views designed to help you analyze findings with clarity and confidence.
Citations
Please cite the original work when using RFdiffusion in publications or research outputs.
|
Watson, Joseph et al. “Broadly applicable and accurate protein design by integrating structure prediction networks and diffusion generative models” bioRxiv.Org, 10 December 2022, https://www.biorxiv.org/content/10.1101/2022.12.09.519842v1. |
|
If you specified Cyclic Chains please cite the following as well: Rettie, S.A., Juergens, D., Adebomi, V. et al. Accurate de novo design of high-affinity protein-binding macrocycles using deep learning. Nat Chem Biol (2025). https://doi.org/10.1038/s41589-025-01929-w |
|
Neurosnap Inc. (2022). Neurosnap: An online platform for computational biology and chemistry. Available at: https://neurosnap.ai/ |
Similar Services
Explore related tools that support similar research workflows:
Proudly supporting 50,000+ scientists worldwide, including 7,000+ leading biotech and global biopharma organizations.
Making Scientific Research
Faster & Easier
Register for free — upgrade anytime.
Interested in getting a license? Contact Sales.
Try Free