Can I run RFdiffusion online without installing it locally?

Yes. Ranomics Tools runs the full RFdiffusion + ProteinMPNN + AF2 multimer pipeline on a dedicated GPU through your browser. You upload a target PDB, set hotspots and binder length, and candidates come back with real ipTM, pLDDT, and i_pAE scores.

How much does one RFdiffusion run cost?

Billing is by the second of dedicated GPU time. A pilot run (~25 minutes on an A100 for four designs) typically clears for under a few dollars from your wallet. New accounts start with a $15 balance, which covers a first small-target run.

Do I need to choose hotspots before running RFdiffusion?

Hotspots are strongly recommended. They tell RFdiffusion which residues on the target the binder should contact. If you do not have a hotspot guess, run Epitope Scout first to score the target surface, then hand off the picked residues into the RFdiffusion form.

RFdiffusion

RFdiffusion is a no-install online de novo binder design tool you can run through tools.ranomics.com on a dedicated GPU. The most widely used de novo binder generator, run end to end: every design it invents is re-folded with AlphaFold2 against your own target, so the confidence score you read is measured rather than the generator marking its own work.

Design inputs

Target PDB

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

Target chain

One chain (A), or several for an oligomeric target (A,B). Every named chain reaches the model; designing against one protomer of a dimer aims at half the epitope.

Hotspot residues

Residues the binder should contact, in original PDB numbering. A plain number is read as your first target chain; prefix the chain to name another (A296,B264). 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.

Binder length (residues)

How long the new binder should be, in residues. Every design draws its own length from this window, so a wide window buys variety and a narrow one buys consistency. 55 to 65 is a compact single-domain binder and the right first try on a small, fairly flat patch. Go to 90 to 150 when the face you are aiming at is broad, or sits in a groove a short binder cannot reach across. Longer binders are also harder to express and purify at the bench, so start short and lengthen only if nothing scores. 30 to 150 allowed.

Number of designs

How many candidates to sample. Each is independently scored by AF2 multimer (ipTM, pLDDT, i_pAE). No fixed ceiling: large requests fan out automatically as a wallet-bounded campaign across GPUs.

First time on this target? Start with 4 designs (~30 min) to confirm your hotspots and binder length, then scale to 100+ once the small batch looks reasonable. Larger jobs take longer wall-clock and are billed by actual GPU consumed; cancellation does not refund time already used.

Runs against your own target. Upload PDB, pick hotspots, set binder length. Results emailed when the run completes (A100-40GB).

Campaign

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

Estimated cost $0.00 based on parameters and historical compute time

Billed by actual GPU consumed, up to the estimate. Surplus returns to your wallet automatically. Cancellation does not refund GPU time already used.

About RFdiffusion

What it is

Generates brand-new protein backbones — 3D shapes with no sequence yet — by starting from noise and denoising toward something that fits the patch of your target you named. Ranomics then puts a sequence on each backbone with ProteinMPNN and refolds the pair with AlphaFold2 multimer, so every candidate that reaches you carries a confidence score measured against your own target rather than an idealised one. This is the general-purpose starting point for de novo binder design. RFdiffusion, Watson et al., Nature 2023.

When to use

You want general-purpose de novo binders, with confidence numbers that came from your actual target.
Your target is an ordinary protein epitope with no sugars or modified residues.
You want the binder's length and shape left open rather than locked to an antibody scaffold.

Prerequisites

Target structure (.pdb / .cif).
Chain ID of the target.
At least one hotspot residue.

Input glossary

Hotspot residues. Comma-separated target-chain residues the binder should contact during diffusion.
Binder length (min/max). How long the new binder should be. Each design draws its own length from this window. 55 to 65 is a compact single-domain binder and the right first try on a small, fairly flat patch; 90 to 150 suits a broad face or one sitting in a groove a short binder cannot reach across. Longer binders are harder to express at the bench, so lengthen only if nothing scores.
Number of designs. How many candidates to generate. Each passes ProteinMPNN sequence design and AF2 multimer scoring.

Runtime

Preset	Typical min
pilot	25 to 40 min (4 to 8 designs)

Ranked candidates with ipTM, pLDDT, i_pAE, and PDBs downloadable from a run of your own. Aim for at least 1 in 5 with ipTM ≥ 0.65 on a tractable target before committing to a full pilot.

References

Starter pilot: 8 binders

Find out whether your target and the face you picked are workable, before paying for a large run.

What you need

A structure file for your target (.pdb or .cif), 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 asks for 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

About $3.37, and about 25 to 40 min (4 to 8 designs) of GPU time. That is an estimate from these exact settings — you are billed for the compute the job actually uses, never more than the estimate.

Then what

Roughly 1 in 5 designs clears the confidence bar on a tractable target, so 8 is enough to tell. If one or two score well, clone the run and raise the design count. If none do, move the hotspots rather than scaling.

Load these settings →

A 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.

Chain B of the same file is PD-1, the natural partner, so the three hotspots are not a guess — they are residues a real binding protein is known to cover. It is also the target the BoltzGen worked example uses, so the two pages can be read against each other: the same epitope, two different generators.

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 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 (residues) — 55 to 65: A window rather than a fixed number — RFdiffusion samples a length inside it. The default window, left alone.
Number of designs — 8: A pilot-sized batch. Enough to tell whether the target and hotspots are workable before committing to 100+, which is what this preset is for.

What came back

Eight designs, two of them above the bar. The best scores ipTM 0.88 with an interface pAE of 3.65 Å and pLDDT 95.5; the second is ipTM 0.82. The other six fall away from 0.46 to 0.19.

Candidates (8) pilot

#	ipTM ?	pLDDT ?	i_pAE (Å) ?	vs. quality bar ?	3D	Structure
1 Top	0.880	95.5	3.65	Meets ipTM 0.65, pLDDT 80 and i_pAE 10 Å	—	—

2	0.820	94.6	4.25	Meets ipTM 0.65, pLDDT 80 and i_pAE 10 Å	—	—

3	0.460	63.3	13.19	ipTM 0.460, below 0.65; pLDDT 63.3, below 80; i_pAE 13.19 Å, above 10 Å	—	—

4	0.350	60.8	14.88	ipTM 0.350, below 0.65; pLDDT 60.8, below 80; i_pAE 14.88 Å, above 10 Å	—	—

5	0.290	59.3	15.51	ipTM 0.290, below 0.65; pLDDT 59.3, below 80; i_pAE 15.51 Å, above 10 Å	—	—

6	0.290	78.8	15.41	ipTM 0.290, below 0.65; pLDDT 78.8, below 80; i_pAE 15.41 Å, above 10 Å	—	—

7	0.230	75.3	16.99	ipTM 0.230, below 0.65; pLDDT 75.3, below 80; i_pAE 16.99 Å, above 10 Å	—	—

8	0.190	73.0	17.51	ipTM 0.190, below 0.65; pLDDT 73.0, below 80; i_pAE 17.51 Å, above 10 Å	—	—

GPU time: 2220 seconds.

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.

Binder Pilot AI Binder Sprint

How to read that

Sort by ipTM: 0.65 or more is a credible binder, 0.75 or more is strong. Then read i_pAE beside it, because the two agreeing is what makes either believable — here the two passing designs sit under 4.3 Å while everything below ipTM 0.5 is above 13 Å, and that gap is the real separation in this table. Look at the spread, not only the top row. A run whose eight designs all score within a whisker of each other is usually telling you something went wrong upstream rather than that you have eight equally good binders; a run that fans out from 0.88 to 0.19 is one where the scoring actually discriminated.

What happened next

Kept the two passing designs and dropped the other six. For those two the next step is an independent re-fold against the same target — a different model, so it is a real second opinion rather than the same one twice — and then SPR or BLI if they survive it. One pilot is a screen, not a result: two hits out of eight is roughly the yield this tool is scoped for, and the point of the pilot is to earn the bigger run.

Recorded from that one run, which cost $2.69, and took 37 minutes. That is what it did, not a quote for yours — the estimate on the form is the live number for your own inputs.

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 RFdiffusion and a sibling algorithm, these run on the same hub against the same target.

Run BindCraft online: You have a target structure and know roughly which patch of its surface you want gripped, and you want brand-new mini-proteins of 60 to 150 residues built to grip it. Every candidate is refolded and filtered before you see it, so what comes back is already a shortlist.
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 campaigns. For design without that filtering step use BindCraft; for antibody formats use RFantibody or IgGM.
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.

Questions people ask about RFdiffusion

Can I run RFdiffusion online without installing it locally?: Yes. Ranomics Tools runs the full RFdiffusion + ProteinMPNN + AF2 multimer pipeline on a dedicated GPU through your browser. You upload a target PDB, set hotspots and binder length, and candidates come back with real ipTM, pLDDT, and i_pAE scores.
How much does one RFdiffusion run cost?: Billing is by the second of dedicated GPU time. A pilot run (~25 minutes on an A100 for four designs) typically clears for under a few dollars from your wallet. New accounts start with a $15 balance, which covers a first small-target run.
Do I need to choose hotspots before running RFdiffusion?: Hotspots are strongly recommended. They tell RFdiffusion which residues on the target the binder should contact. If you do not have a hotspot guess, run Epitope Scout first to score the target surface, then hand off the picked residues into the RFdiffusion form.