How to Use RFD4-Proteina
Commercially Available Online Web Server
RFD4-Proteina guide to all-atom protein generation, protein and DNA binders, ligand conditioning, motif scaffolding, and inverse folding.
RFD4-Proteina jointly generates protein sequences and all-atom structures using a flow-matching model. It supports de novo protein generation and structure-conditioned design around protein targets, small molecules, DNA, and retained motifs. It can also redesign an existing protein through inverse folding or partial diffusion.
Neurosnap exposes seven Design Mode choices through ordinary structure inputs and residue selections. The output consists of generated candidates and sequence tables for independent review. This workflow does not perform reward-guided search, automatic refolding, binding-affinity prediction, or automatic ranking.
Choose the design task before preparing inputs
Use De Novo Design to explore new proteins without a supplied target; binder modes to generate a protein around selected molecular context; Motif Scaffolding to retain a functional structural fragment; and structure-redesign modes to propose alternatives to an existing protein.
For search-based generation with integrated evaluation, compare Proteina-Complexa. For a different all-atom design workflow with explicit atom selections and origin controls, see RFdiffusion3. These services expose different conditions and outputs, so a successful submission to one does not imply its settings can be transferred unchanged to another.
How RFD4-Proteina Works
How joint sequence and structure generation works
The native RFD4-Proteina model uses flow matching to generate amino-acid identities and atomistic geometry under the conditions supplied for the task. Protein-target conditioning, ligand conditioning, motif retention, and backbone-only redesign constrain different parts of that problem. The official RFD4-Proteina documentation provides method background and native workflows.
Conditioning does not always mean copying every coordinate exactly. A protein target's side chains and global pose can change, and retained ligand or DNA geometry can shift slightly. Inspect the generated context as well as the newly designed protein before deciding whether the original hypothesis was preserved.
Match each mode to its required context
| Design Mode | Input and intended design |
|---|---|
| De Novo Design | Generate a new protein using the requested length range; no input structure is required. |
| Protein Binder | Retain selected protein target residues and generate a separate contacting protein chain. |
| Small Molecule Binder | Retain one selected non-polymer ligand and generate a protein pocket around it. |
| DNA Binder | Retain the selected duplex and generate a separate protein chain. |
| Motif Scaffolding | Retain selected protein motif residues and generate scaffold segments defined by a contig. |
| Inverse Folding | Keep protein backbone coordinates while generating new sequence and side chains. |
| Partial Diffusion | Generate variations on an existing protein sequence and structure. |
Fill only the sections relevant to the chosen mode. Inverse Folding and Partial Diffusion require protein-only inputs; they are not interchangeable with a ligand-conditioned binder setup.
Prepare complete structures and correct numbering
All conditioned modes require one Input Structure with one model. Selected protein residues need complete N, CA, C, and O backbone atoms. Remove alternate locations and insertion codes, verify finite coordinates, and use chain identifiers containing letters only.
For PDB inputs, selections use author residue numbering. For mmCIF polymers, use label chain IDs and label sequence numbers; non-polymer residues use author residue numbers. Check the actual identifier system in the uploaded file before constructing selectors. A chain named A in a molecular viewer may not be the label identifier expected from a particular mmCIF file.
Minimum Design Length / Maximum Design Length apply to de novo and binder modes; binder lengths exclude the retained target. Motif Scaffolding uses its contig, while structure redesign retains the input length. The selected context plus the maximum generated size must fit the 768-token service limit; non-protein atoms are conservatively counted individually. Length planning therefore needs the molecular context, not just the desired binder size.
Configure protein, small-molecule, and DNA binders
For Protein Binder, select contiguous Target Residues, such as A17-131, ensuring every residue in each range exists. Optional Hotspot Residues must lie inside that target selection. Retain enough context around the epitope to avoid creating an artificial cut-edge interface.
For Small Molecule Binder, Ligand Residue selects exactly one non-polymer molecule by chain and residue number, such as B1. The supported ligand has 2–100 heavy atoms and must have complete coordinates and intended chemistry. Prefer mmCIF with chemical-component bond information for custom ligands. Only the selected ligand is retained, so this mode should not be interpreted as preserving a surrounding protein pocket.
For DNA Binder, supply an already assembled complementary duplex and select both complete, contiguous strands, one range per strand—for example B1-10,C1-10. The input must already contain the intended pairing and geometry. Sparse nucleotide selections, RNA residues, or one isolated strand do not express this duplex-design task.
Scaffold motifs and control redesign strength
A Contig uses commas to separate generated lengths and retained motif ranges. For example:
30-50,A10-15,30-50
This generates two flanking segments around the six retained residues A10–15. Those generated lengths exclude the motif itself. Use /0 to separate independent chains, and do not repeat a retained motif residue. Motif conditioning retains supplied atom identities and geometry, while inverse folding constrains only the backbone and allows sequence redesign.
Partial Trajectory Fraction applies only to Partial Diffusion. Smaller values stay closer to the input; larger values allow greater change. It is a trajectory fraction, not a displacement in angstroms or a fraction of residues guaranteed to mutate.
Sampling budget and reproducibility
Sampling Budget selects Low, Medium, or High task-specific recycling schedules with 60 sampling steps. Medium is the recommended starting point. A higher budget increases runtime; it is not an independent validation step.
Record Random Seed, selected residues, construct preparation, and length bounds. Deterministic Sampling uses ODE sampling without added mid-trajectory noise, but initial noise still depends on the seed and exact GPU reproducibility is not guaranteed. Number of Designs counts generated candidates rather than experimentally accepted hits.
Interpret structures and sequence exports
Inspect the annotated mmCIF designs and the design and sequence tables. Sequence outputs include retained protein target chains as well as newly generated chains; identify the designed chain before selecting a sequence for synthesis. Do not assume the first sequence row is the binder.
Generated CIF B-factors are not calibrated confidence scores. There is no automatic affinity ranking or refolding result in this workflow. Independently assess candidate geometry, motif retention, clashes, and diversity, then use a suitable folding or complex-prediction method to evaluate shortlisted sequences. Experimental testing remains necessary.
RFD4-Proteina troubleshooting
Why is a selection invalid? Check PDB versus mmCIF numbering, missing residues, atom completeness, and whether every hotspot is inside Target Residues.
Why did retained context move? Conditioning can permit coordinate changes. Compare the actual output with the input rather than assuming exact coordinate preservation.
Which candidate is best? The native outputs are generated proposals, not an automatically validated ranking. Define independent selection criteria before expanding the campaign.
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 RFD4-Proteina on Neurosnap
Using RFD4-Proteina on Neurosnap could drastically accelerate joint protein sequence and all-atom structure generation.
- Seven generation tasks share one all-atom model.
- Structure-aware controls encode target, ligand, DNA, and motif conditions.
- Native structures and sequence tables support independent downstream assessment.
How to Use RFD4-Proteina on Neurosnap
To harness the capabilities of RFD4-Proteina, 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 RFD4-Proteina.
- Provide Inputs: Provide all the inputs specified within the submission panel and optionally configure the tool as desired.
- Run Tool: Submit the RFD4-Proteina 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 RFD4-Proteina in publications or research outputs.
|
RFD4-Proteina. Institute for Protein Design and NVIDIA. https://github.com/RosettaCommons/RFD4-Proteina |
|
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