RFantibody

RFantibody is a no-install online nanobody design tool you can run through tools.ranomics.com on a dedicated GPU. Nanobodies are small enough to reach into a cleft a full antibody cannot and simple enough to express, and this designs them straight onto the patch you name with no RoseTTAFold or Rosetta to install.

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 →

Design inputs

Target

Target PDB

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

Target chain 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

CDR lengths

CDR length spec (e.g. H1:8,H2:7,H3:10-16). Ranges allowed on H3.

Number of backbones (5 candidates each)

How many backbones to design. ProteinMPNN then puts five sequences on each one and every sequence is independently folded and scored by RoseTTAFold-2 (pLDDT, pAE, ipAE), so 4 backbones return 20 scored candidates. 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 that outputs look reasonable, then scale to 100+ once the small batch checks out. 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 PDB, pick hotspots. Results emailed when the run completes (A100-40GB).

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 RFantibody

What it is

Designs nanobodies — single-domain antibodies, the VHH format — against a patch of your target, then refolds each one with AlphaFold2 and reports how confident it is about the contact. Nanobodies are the format that carries most easily into yeast display, mammalian display and hybridoma workflows, which is the usual reason to pick this over a de novo mini-protein. RFantibody, Bennett et al., Nature 2026.

When to use

You want a nanobody scaffold rather than a de novo mini-protein.
Your downstream work is yeast display, mammalian display or a hybridoma workflow.
Your target is an ordinary protein epitope without heavy sugar coverage.

Prerequisites

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

Input glossary

Hotspot residues. Comma-separated target-chain residues defining the epitope the CDRs should target.
Number of backbones (5 candidates each). How many backbones to generate. Each gets five ProteinMPNN sequences, and every sequence is re-folded and filtered by RoseTTAFold-2 on pAE, ipAE and pLDDT -- so 4 here returns 20 candidates.

Ranked VHH candidates with pAE, pLDDT, ipAE, and PDBs downloadable from a run of your own. Filter at pAE ≤ 5 / ipAE ≤ 6 for downstream wet-lab work.

References

Trial run: 2 nanobodies

Check that a VHH scaffold can be placed on the face you picked at all, before scaling up.

What you need

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

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 designs answer whether the epitope is reachable by a VHH at all, not whether it is a good one. Roughly 1 in 5 clears the confidence bar on a tractable target, so once these come back scored, clone the run and raise the count to 10 or more before reading anything into the numbers.

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, BoltzGen and BindCraft examples on this site. That makes four generators aimed at one epitope, and this is the only one of the four that answers with an antibody — a VHH nanobody, framework fixed, with only the three heavy-chain CDR loops designed. So the comparison worth making across those four pages is not which tool scores higher. It is what shape of molecule you get back for the same request.

What we put in the form

Target PDB — the 4ZQK file, uploaded whole: Both chains, exactly as it downloads from the PDB. Chain A keeps its crystal numbering 18-132, which is the numbering the hotspot field expects.
Target chain — A: Restricts the design to PD-L1. Chain B in the same file is PD-1, the natural partner; leaving it in the upload is harmless because the run only reads the chain you name here.
Hotspot residues — 54, 56, 115: Ile54, Tyr56 and Met115 in the file's own numbering — the face PD-1 covers.
CDR lengths — H1:8, H2:7, H3:10-16: The three heavy-chain loops. H1 and H2 are pinned at their common lengths; H3 is given a range because it is the loop that does most of the binding, so its length is the thing worth sampling rather than fixing.
Number of backbones (5 candidates each) — 4: Four backbones — and this is the input to understand before you spend. The field sets the backbone count, not the candidate count: ProteinMPNN then designs five sequences on each one, so four here returned 20 scored candidates and twenty rows in the table below. Whatever you type, expect five times as many rows and roughly five times the work.

What came back

20 candidates. Five meet every bar; fifteen do not. Interface pAE runs 3.72 to 17.48 Å, global pAE 2.68 to 9.84 Å, and pLDDT 87 to 91.

Candidates (20)

#	ipAE (Å) ?	pLDDT ?	pAE (Å) ?	vs. quality bar ?	3D	Structure
1 Top	3.72	91.0	2.68	Meets pLDDT 80, ipAE 10 Å and pAE 5 Å	—	—

2	4.33	91.0	2.90	Meets pLDDT 80, ipAE 10 Å and pAE 5 Å	—	—

3	5.26	90.0	3.42	Meets pLDDT 80, ipAE 10 Å and pAE 5 Å	—	—

4	5.31	91.0	3.39	Meets pLDDT 80, ipAE 10 Å and pAE 5 Å	—	—

5	7.69	89.0	4.74	Meets pLDDT 80, ipAE 10 Å and pAE 5 Å	—	—

6	12.56	90.0	7.11	ipAE 12.56 Å, above 10 Å; pAE 7.11 Å, above 5 Å	—	—

7	12.76	89.0	7.31	ipAE 12.76 Å, above 10 Å; pAE 7.31 Å, above 5 Å	—	—

8	12.77	90.0	7.15	ipAE 12.77 Å, above 10 Å; pAE 7.15 Å, above 5 Å	—	—

9	12.90	90.0	7.25	ipAE 12.90 Å, above 10 Å; pAE 7.25 Å, above 5 Å	—	—

10	13.29	91.0	7.42	ipAE 13.29 Å, above 10 Å; pAE 7.42 Å, above 5 Å	—	—

11	13.40	91.0	7.45	ipAE 13.40 Å, above 10 Å; pAE 7.45 Å, above 5 Å	—	—

12	13.71	89.0	7.77	ipAE 13.71 Å, above 10 Å; pAE 7.77 Å, above 5 Å	—	—

13	14.06	90.0	7.85	ipAE 14.06 Å, above 10 Å; pAE 7.85 Å, above 5 Å	—	—

14	14.24	90.0	7.99	ipAE 14.24 Å, above 10 Å; pAE 7.99 Å, above 5 Å	—	—

15	14.26	91.0	7.87	ipAE 14.26 Å, above 10 Å; pAE 7.87 Å, above 5 Å	—	—

16	14.49	91.0	8.04	ipAE 14.49 Å, above 10 Å; pAE 8.04 Å, above 5 Å	—	—

17	14.85	89.0	8.36	ipAE 14.85 Å, above 10 Å; pAE 8.36 Å, above 5 Å	—	—

18	16.07	88.0	9.13	ipAE 16.07 Å, above 10 Å; pAE 9.13 Å, above 5 Å	—	—

19	16.46	87.0	9.30	ipAE 16.46 Å, above 10 Å; pAE 9.30 Å, above 5 Å	—	—

20	17.48	88.0	9.84	ipAE 17.48 Å, above 10 Å; pAE 9.84 Å, above 5 Å	—	—

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

Start with the column that does not help. pLDDT is 87 to 91 across all twenty, every failing candidate included, so all twenty clear the 80 bar and the column separates nothing at all. That is the normal case for an antibody run rather than a fault: the framework is a real nanobody scaffold, so of course it folds. Confidently folded is not bound. The column that separates is ipAE, and it splits cleanly instead of tailing off — five candidates between 3.72 and 7.69 Å, then nothing whatsoever until 12.56, then fifteen more out to 17.48. A gap that wide is the real signal here. It says the five are a different outcome from the fifteen, not merely the top of a continuum, which is a much safer thing to act on than a ranking. Global pAE agrees exactly: under 5 Å for all five, above 7 for every other one. The vs. quality bar column spells it out per row — every rejection names ipAE and pAE, and not one of them names pLDDT.

What happened next

Treated the five as the only rows carrying information, and the size of the ipAE gap as the reason to trust that split rather than hedging down the list. On a real target the next run is more backbones at these same settings — remembering the five-to-one again, so the bill climbs faster than the number you type — then an independent re-fold of the few that clear the bar, then yeast display, and SPR or BLI only after that. Twenty candidates off four backbones is a screen, 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.

Related tools on Ranomics

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

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 IgGM online: You have an antibody or nanobody and an antigen, and you want to redesign its binding loops, humanise its framework, raise its affinity, or just see how the two dock — one model does all of it, aimed at the epitope you name. For a nanobody from scratch use RFantibody; for a paired heavy and light antibody fragment use ESMFold2 design.
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.