Can I run BindCraft online without installing it?

Yes. Ranomics Tools runs the full BindCraft hallucination loop on a dedicated GPU through your browser. Upload a target PDB, pick hotspots and binder length, and candidates come back with AF2 ipTM and pLDDT on every hit.

BindCraft vs RFdiffusion: which should I pick?

Both design de novo binders against your target. BindCraft hallucinates sequences and backbones jointly inside AF2; RFdiffusion samples a backbone first then drops sequences in with ProteinMPNN. Run both on the same hub and compare ipTM before committing to wet-lab.

How long does a BindCraft trial run take?

Runtime depends on target size, how many designs you ask for, and how many candidates pass the internal ipTM filter. You pay for the GPU time your run uses.

BindCraft

BindCraft is a no-install online de novo binder design tool you can run through tools.ranomics.com on a dedicated GPU. Design and in-silico filtering run in one pass, so what comes back is already a shortlist rather than raw output you have to triage yourself.

Example run — Human PD-L1, the IgV domain, taken as chain A of PDB 4ZQK — the solved PD-1/PD-L1 complex. See what came back →

BindCraft runs on A100-80GB; larger design counts take proportionally longer. Results are emailed on completion; you can close this tab.

Design inputs

Target

Target PDB

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

Target chain

One chain (A). BindCraft's GPU image has not been run against an oligomeric target yet, so naming more than one chain is refused when you submit.

Hotspot residues

Comma-separated residue indices from the target chain that the binder should contact (e.g. 54,56,115). 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 (min) Binder length (max) Number of designs

How many final filtered designs to return. Each passes AF2 re-prediction with ipTM and pLDDT above the BindCraft default thresholds. No fixed ceiling: large requests become a wallet-bounded full-size run, split into pieces that run on our GPUs and bill as they finish.

First time on this target? Start with 4 designs to confirm your hotspots and binder length range, then scale to 100+ once the small batch looks reasonable. BindCraft optimises each design individually, so larger runs take considerably longer. You pay for the GPU time your run uses; cancellation does not refund time already used.

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 BindCraft

What it is

Designs brand-new mini-proteins that grip a patch of your target. It runs AlphaFold2 multimer backwards — pushing a random starting sequence toward one the model believes will bind the residues you named — then assigns a real sequence with ProteinMPNN and refolds every candidate to check the answer survives. What reaches you has already been filtered on interface confidence and fold quality. BindCraft, Pacesa et al., Nature 2025.

When to use

You have a target structure and at least one residue on its surface you want the binder to touch.
You want a small de novo protein of 50 to 150 residues, not an antibody.
You would rather see a filtered shortlist than every candidate the run generated.

Prerequisites

Target structure as .pdb, .cif, or .mmcif.
Chain ID of the target within that structure.
At least one hotspot residue index on the target chain.

Input glossary

Hotspot residues. Comma-separated target-chain residue indices the binder should contact (e.g. 54,56,115). These bias AF2 backpropagation toward the intended epitope. Click residues in the 3D viewer to toggle them.
Binder length (min/max). Residue-count window for the generated binder chain (50 to 150). Shorter binders are easier to validate in yeast display; longer ones can target larger interfaces.
Number of designs. How many final filtered designs to return (1 to 500). Each passes AF2 re-prediction with ipTM and pLDDT above the BindCraft default thresholds. Runtime scales with the count: start with a 4-trajectory batch to confirm your target and hotspots, then scale up.

Filtered candidate binders with ipTM, pLDDT, shape complementarity, and PDBs downloadable from a run of your own. An ipTM of 0.75 or more is a credible binder and 0.85 or more is a strong candidate. Hand off promising designs to the Ranomics yeast display CRO for in vitro validation.

References

Trial run: 2 trajectories

Check that your target parses, your hotspots resolve against it, and the pipeline returns scored designs — before committing to a run that hunts for hits.

What you need

A structure file for your target (.pdb, .cif or .mmcif), the chain ID it sits on, and at least one residue on the face you want the binder to touch.

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 its results hand the target and the residues back into this form.

What it costs

You pay for the GPU time your run uses.

Then what

Two trajectories tell you the setup works, not whether the target is bindable — BindCraft burns more GPU per design than anything else here, which is why the first step is this small. Once it comes back clean, clone the run and raise the count.

Load these settings →

A past run we actually did

The target

Human PD-L1, the IgV domain, taken as chain A of PDB 4ZQK — the solved PD-1/PD-L1 complex. Hotspots Ile54, Tyr56 and Met115.

The same target, chain and hotspots as the RFdiffusion and BoltzGen examples on this site, and that is the point — three different generators aimed at one epitope, so the three pages can be read against each other. Chain B of the same file is PD-1, the natural partner, so the hotspots are residues a real binding protein is known to cover rather than a guess.

What we put in the form

Target PDB — the 4ZQK file, uploaded whole: Both chains, exactly as it downloads from the PDB. No trimming and no renumbering, so chain A keeps its crystal numbering 18-132 — which is the numbering the hotspot field expects.
Target chain — A: This is what restricts the design to PD-L1. Chain B in the same file is PD-1, and the run ignores it — otherwise the model would be designing into an already occupied site.
Hotspot residues — 54, 56, 115: Ile54, Tyr56 and Met115 in the file's own numbering. Three is a normal number: enough to aim the binder at one patch, few enough that you have not drawn the interface yourself.
Binder length (min) — 50: The bottom of the default window. BindCraft picks a length inside the range rather than being told one.
Binder length (max) — 100: The top of the default window, left alone.
Number of designs — 2: Small even for a trial run, and that is this tool rather than impatience: BindCraft optimises each design individually instead of sampling a batch, so each design takes more GPU time than on the diffusion tools.

What came back

Two designs, landing in almost the same place: ipTM 0.75 and 0.76, pLDDT 81 for both, shape complementarity 0.64 and 0.60, a refolding RMSD of 3.04 and 2.96 Å, and surface hydrophobicity of 0.29 and 0.30.

Candidates (2)

#	ipTM ?	pLDDT ?	RMSD (Å) ?	Shape complementarity (SC) ?	Surface hydrophobicity ?	3D	Structure
1 Top	0.750	81.0	3.04	0.640	0.29	—	—

2	0.760	81.0	2.96	0.600	0.30	—	—

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

Read ipTM first: 0.75 is this tool's bar for a credible binder, so both designs sit on that bar rather than above it, and pLDDT 81 clears its own bar of 80 by about as little. Then read the two columns that stop you over-reading those. Shape complementarity wants 0.65 for an antibody-grade fit and both designs are under it; refolding RMSD wants to be inside 1.5 Å and both are close to 3. Put together: the interface is plausible, the fit is loose, and neither design rebuilds quite the backbone it was drawn as. That is a normal trial run result and it is worth more to you than a single headline number would be. Do not read the ranking as an ipTM sort. Design 2 has the marginally higher ipTM and still ranks second, because BindCraft ranks on a composite that includes the fit — and design 1 wins there, 0.64 against 0.60. The last column is not a result. Surface hydrophobicity 0.29 and 0.30 were always going to be under 0.35: BindCraft throws out anything above that line before a design reaches you, so clearing it is a precondition of being on this page rather than something these two achieved. Do not read them as low — on the 101 accepted designs FreeBindCraft publishes for this same target the median is 0.26 and the maximum 0.34, so roughly three-quarters of accepted designs sit below these two, nearer that line than most designs that pass. It is a property of the shortlist, and a poor way to choose between designs inside it.

What happened next

Treated this as a screen that came back amber rather than green. Two designs sitting exactly on the bar with a sub-par fit is a reason to run more of them, not a reason to order peptides. The next step is the same target and the same hotspots at a higher design count, then an independent re-fold on whatever clears the bar properly, and SPR or BLI only after that. A two-design trial run can tell you the target and hotspots are workable; it cannot tell you which binder to make.

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 — the guide for the tool you ran states its one. 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.

Related tools on Ranomics

If you are picking between BindCraft and a sibling algorithm, these run on the same hub against the same target.

Run RFdiffusion online: You have a target structure and a patch of its surface you want gripped, and you want brand-new binders of whatever shape works — the general-purpose starting point for de novo design. Every candidate comes back with a real AlphaFold2 confidence score against your own target. For antibody or nanobody formats use RFantibody or IgGM instead.
Run BoltzGen online: You have a target and have not settled on what shape the binder should be. One model here aims mini-proteins, nanobodies, antibodies or peptides at the same site, so you can compare formats instead of guessing. The only design tool here that handles sugars and modified residues on the target natively.
Run PXDesign online: You have a target and you want every single candidate to arrive with a real AlphaFold2 confidence score against that target, not a cheaper stand-in. This is the pipeline Ranomics runs for its own wet-lab projects. For design without that filtering step use BindCraft; for antibody formats use RFantibody or IgGM.

Questions people ask about BindCraft

Can I run BindCraft online without installing it?: Yes. Ranomics Tools runs the full BindCraft hallucination loop on a dedicated GPU through your browser. Upload a target PDB, pick hotspots and binder length, and candidates come back with AF2 ipTM and pLDDT on every hit.
BindCraft vs RFdiffusion: which should I pick?: Both design de novo binders against your target. BindCraft hallucinates sequences and backbones jointly inside AF2; RFdiffusion samples a backbone first then drops sequences in with ProteinMPNN. Run both on the same hub and compare ipTM before committing to wet-lab.
How long does a BindCraft trial run take?: Runtime depends on target size, how many designs you ask for, and how many candidates pass the internal ipTM filter. You pay for the GPU time your run uses.