RFD4-Proteina

RFD4-Proteina is a no-install online all-atom binder design tool you can run through tools.ranomics.com on a dedicated GPU. Sequence and all-atom structure come out of the same model, so a design arrives complete instead of as a backbone still waiting for a sequence step.

Example run — PD-L1, chain A, 115 residues, taken from the vendor's own example condition spec that ships inside the upstream image. See what came back →

Design inputs

Target

Target PDB

Upload the target structure as .pdb, .cif, or .mmcif.

Target chain

One chain (A), and one only. Name the chain carrying the face you want gripped. A,B is refused here rather than quietly designing against half of it (tools/rfd4/__init__.py::validate). For a target that only makes sense as two chains together, use Proteina instead.

Hotspot residues

Residues the binder should grip, in your file's own numbering. At least one is required: this run conditions the model on the patch you name, and there is nothing to design against without it. Plain numbers, read against the chain you named above (54,56,115). Up to 64. Click residues in the 3D viewer below to toggle them in and out of this list.

Not sure which residues? Score your target's surface with Epitope Scout first.

3D residue picker

Surface

Upload a target PDB above to enable click-to-pick. Typed entry in the Hotspot residues field above also works.

Design

Binder length (residues)

How long each new binder may be. The range you give here is passed to the generator as the binder length window. 60 to 120 is the default and a reasonable first try. Narrow it towards the short end for a compact single-domain binder on a flat patch, and widen it for a broad face or a groove a short binder cannot reach across. Longer binders are also harder to express and purify at the bench. 20 to 300 allowed.

Number of designs

How many binders to generate. Each is an independent sample and comes back as one structure file holding your target chain plus the binder. The table they land in carries no score column, so the count buys you variety to choose from and not a ranking.

First time on this target? Start with 2 to 8 designs to confirm the chain and hotspots you picked, then scale up once the small batch looks sane. Larger jobs take longer wall-clock, and you pay for the GPU time your run uses; cancellation does not refund time already used.

Runs against your own target. Upload a structure, mark the residues to grip, set the binder length. Each design comes back with its sequence and its all-atom structure out of the same model, so there is no separate sequence-design step afterwards. Results are emailed when the run completes on our GPUs.

Group

Group related runs on your jobs page. Leave blank to file under Uncategorized.

You pay for the GPU time your run uses. Cancelling does not refund GPU time already used.

About RFD4-Proteina

What it is

Designs de-novo protein binders against a patch of your target structure, and writes each binder's sequence and its full atomic structure in one model, so a design comes back complete rather than as a backbone waiting for a sequence step. RFD4-Proteina, Corley et al., preprint 2026.

When to use

You want a small de-novo binder against a protein target, not an antibody format.
You would rather have sequence and structure out of one model than run a separate sequence-design step.
You have your own way of judging a complex. This tool hands back structures and no score.

Prerequisites

Target structure (.pdb / .cif).
Chain ID of the target. One chain only.
At least one hotspot residue on that chain.

Input glossary

Hotspot residues. Residues on the target chain the binder should grip, in the uploaded file's own numbering. At least one is required.
Binder length (residues). The binder length range you give. Anywhere between 20 and 300 residues; the default range is 60 to 120.
Number of designs. How many binders to generate. Each one is an independent sample, and each is delivered as a structure file holding your target chain plus the binder.

One structure per design, each holding your target chain and the de-novo binder beside it, downloadable from a run of your own. There are no score columns: nothing in this run measures how well a binder holds, so take the structures into a predictor before choosing between them.

References

Trial run: 2 binders

Find out whether the face you picked can be gripped at all, before paying for a full sweep.

What you need

A structure file for your target (.pdb or .cif), the chain ID of the chain you want bound, and at least one residue on it.

Hotspot residues are the numbered residues on the patch of your target you want the binder to sit on, written in your PDB’s own numbering — this tool will not start without at least one. If you do not know yours yet, score your target’s surface with Epitope Scout first — it is free, and you copy the residues it picks into the field below.

What it costs

You pay for the GPU time your run uses.

Then what

Two binders tell you the run completes on your target and what shape of thing comes back. They cannot tell you which one binds: no number here ranks them. Open both, check the binder sits on the residues you named, then clone the run at a larger count and fold the complexes before reading anything into them.

Load these settings →

A past run we actually did

The target

PD-L1, chain A, 115 residues, taken from the vendor's own example condition spec that ships inside the upstream image. One hotspot residue, 56.

This run was ours but it was not a customer's: it is the validation run that proved the container works, driven straight from the binder condition spec that ships inside the upstream image rather than from the form above. So read it for the shape of what comes back, eight complete structures and no scores, and not as a result on a target of your own.

What we put in the form

Target PDB — PD-L1 chain A, 115 residues: Supplied to the container as the upstream spec's own annotated structure rather than as an upload, because this run predates the form. A run of your own uploads a .pdb or .cif here and the normalizer does the rest.
Target chain — A: One chain, and only one: the run designs against the chain you name here and refuses a request that names more than one.
Hotspot residues — 56: A single residue on the PD-1 face. One is enough to point the binder at a patch; more narrows it further.
Binder length (residues) — 60-120: The default range, passed to the generator as the binder length window. The eight structures that came back are not all the same size, and a binder can come back longer than the target, which is why the results page names the target chain instead of telling you to take the shorter one.
Number of designs — 8: Eight independent samples. Unlike the antibody tools there is no multiplier hiding here: eight designs is eight structures and eight rows.

What came back

Eight structures, all eight of the eight asked for. Each is one file holding chain A and the de-novo binder beside it. The table has no score columns, because there are no scores to put in them: every delivered design carries its position, its name and its structure, and nothing more.

Target chain A (115 aa) — each downloaded structure holds your target plus one de-novo binder chain, and the binder is the chain that is not A. Do not pick it by length; it can be longer than the target. Contig 60-120, /0, A1-115.

These designs carry no confidence score of their own — RFD4 generates backbones and sequences, it does not judge them. Scoring them means folding each one against your target, and these designs did not come back with the binder sequence that needs, so download the structures and fold them yourself.

Candidates (8) binder design

Returned 8 / 8

The # column is the order the designs came back in, not a ranking of how well anything binds.

#	3D	Structure
1	—	—

2	—	—

3	—	—

4	—	—

5	—	—

6	—	—

7	—	—

8	—	—

Validating in the lab is optional Designs from your own run are yours to download and take anywhere. If you do want them made and measured, Ranomics runs yeast display, mammalian display, BLI and DMS on the same data loop.

Test binders in the lab AI Binder Sprint

How to read that

The reading is structural. Open each file, find the chain that is not A, and check it actually sits against residue 56 rather than somewhere else on the surface. Eight samples at one setting will not all do that.

What happened next

For this run, nothing: it existed to prove the container, and it did. On a target of your own the next step is the one this page cannot do for you: fold each complex through a predictor, let that give you the confidence number, and only then decide which handful is worth a larger sweep or a bench. Eight unscored structures are a starting point, not a shortlist.

Recorded from that one past run. That is what it did, not a prediction for yours.

What good looks like

Each tool reports a subset of these depending on whether it does design, sequence recovery, or structure prediction.

ipTM: Predicted confidence in the contact between two chains, on a 0 to 1 scale. Higher is better: 0.65 to 0.75 depending on the tool. Individual tools set their own pass bar a little either side of that. On a multi-chain target the number may cover the target's own chain–chain interface as well as the binder's.
pLDDT: Per-residue confidence in the predicted fold. Higher means the model is more sure of that part of the structure.
i_pAE and pAE: Predicted alignment error, at the interface (i_pAE) or across the whole structure (pAE). Lower is better.
ProteinMPNN recovery: Fraction of native residues recovered when ProteinMPNN redesigns a known sequence on its native backbone — the folded shape on its own, before a sequence is chosen for it. Higher is better: > 0.4 usable; > 0.6 excellent.