Boltz-2

Boltz-2 is a no-install online cofold validation tool you can run through tools.ranomics.com on a dedicated GPU. The cheap second opinion on a shortlist: it works from sequence alone, so a whole batch of designs from another tool can be re-checked for seconds each before any of them reach the bench.

Example run — MDM2 — PDB 1YCR, chain A, 85 residues. See what came back →

Validation inputs

Preset

Single-sequence (default, fast) Folds from the sequences alone, with no search for related natural sequences. The right choice for anything you designed on this site — a designed binder has no natural relatives to find, so the search would come back empty anyway. With MSA (slower, natural sequences) Searches public databases for sequences related to yours and folds with that extra evidence, so it takes longer per design than the fast option above. Meant for natural or near-natural sequences, which do have relatives. On the one head-to-head we have run — three designed antibody binders — it did a WORSE job of telling the real binder apart from decoys than the fast option above, so reach for it only if your sequence genuinely has natural relatives.

Antigen PDB

Upload the target structure as .pdb, .cif, or .mmcif. The antigen folds as one chain; binders fold as separate chains.

Antigen chain Hotspot residues (optional)

The residues on your antigen you were hoping the binder would touch — comma separated, counted from 1 at the start of the antigen chain. These are scored, not steered: nothing here changes the fold, and the run reports how many of them the binder actually ends up contacting (any heavy atom within 5 Å) so you can tell whether it landed where you wanted. Leave blank to skip that count.

Binder sequences

One sequence per line, or FASTA with >name headers. Each sequence folds independently against the antigen. 20 to 400 aa per binder, and up to 50 binders per run on the single-sequence preset. With MSA the limit is 16, because each design takes longer: a larger batch would run past the run's time limit and the tail would be cut off.

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 Boltz-2

What it is

Folds a binder and its target together and tells you how confident it is that they touch. It was trained on antibody-antigen complexes and works from sequence alone, which makes it a genuinely independent second opinion next to AlphaFold2 multimer: when the two agree the complex is probably real, and when they disagree that is worth knowing before you order DNA. Returns the folded complex plus ipTM (confidence in the interface), pTM (confidence in the whole complex) and per-residue confidence for every design. Boltz-2, Passaro et al., bioRxiv 2025.

When to use

You designed binders with another tool here and need them scored against the antigen you actually care about.
You have natural or near-natural antibody fragments (scFv or Fab), nanobodies or peptides and want a fast independent fold of each.
AlphaFold2 interface confidence has topped out across your candidates and you want a second, differently trained opinion before committing to synthesis.

Prerequisites

Antigen PDB or mmCIF (single chain; the binder is added separately).
One or more binder sequences (scFv, nanobody, peptide, anything that folds as a single protein chain), 20 to 400 aa each.
Optional: a list of antigen hotspot residue numbers to count contacts against (1-indexed on the chosen antigen chain).

Input glossary

Antigen PDB. Upload the target structure as .pdb, .cif, or .mmcif. CIF inputs are converted to PDB server-side.
Antigen chain. Single chain ID (e.g. A) that Boltz-2 should treat as the antigen. The binder folds as a separate chain.
Hotspot residues. Optional. Comma-separated 1-indexed positions on the antigen chain (e.g. 55,56,57,71,72,73,74). The pipeline reports how many of these residues the binder contacts (heavy atom within 5 Å).
Binder sequences. Paste one sequence per line, or upload as FASTA (>name headers). Each sequence folds independently against the antigen. 20 to 400 aa per binder, and up to 50 binders per run on the single-sequence preset. With MSA the limit is 16: that preset is considerably slower, so a larger batch would run past the run ceiling and the tail would be cut off.
Preset. Single-sequence (default) folds in msa: empty mode, the right choice for designed sequences. With MSA fetches MSAs from the public ColabFold MMseqs2 endpoint. It is slower, and on the one comparison we have run — three designed antibody binders — it separated the real binder from decoys WORSE, so it is not the default. Whether it helps on genuinely natural sequences is untested here.

Per-design folded complex PDB + ipTM, pTM, complex_pLDDT, complex_iplddt, and hotspot contact count. Strict-pass classification (complex_pLDDT > 85, ipTM > 0.7, n_hotspot_contacts > 4) surfaces which designs are worth ordering.

References

A past run we actually did

The target

MDM2 — PDB 1YCR, chain A, 85 residues. The pocket p53 binds.

Because this one can be marked. MDM2 is the most heavily characterised protein-protein interface in drug discovery: the natural ligand is known, several tighter binders have been published, and every one of them has a measured affinity. So we can fold designs and known binders through the identical path and read the designs against a real scale rather than against a threshold somebody picked.

What we put in the form

Antigen PDB — 1YCR, chain A: MDM2's p53-binding domain. Chain B in that file is the p53 peptide itself and was removed — leaving it in would have let the model copy the answer.
Binder sequences — 12 designed peptides, 12 to 20 residues: Submitted in one batch. Boltz-2 designs nothing; it folds what you give it against the target and scores the interface, so a batch is just twelve independent verdicts.
Hotspot residues (optional) — 54, 57, 58, 61, 62, 67, 72, 75, 86, 91, 93, 96, 99, 100: The 14 residues lining the cleft — L54, L57, G58, I61, M62, Y67, Q72, V75, F86, F91, V93, H96, I99, Y100 — where p53's Phe19, Trp23 and Leu26 insert. Naming them makes the run report whether a design lands in the pocket, instead of only whether it scores well somewhere.
Preset — msa_server: Builds an alignment for the antigen before folding. Worth it on a target with many known relatives.

What came back

Twelve complexes, every one strict_pass. ipTM runs from 0.874 to 0.952, complex pLDDT from 91.3 to 97.1, and each design contacts 13 or 14 of the 14 cleft residues.

Boltz-2 cofold msa_server

Antigen length

85 aa

Hotspots requested

54, 57, 58, 61, 62, 67, 72, 75, 86, 91, 93, 96, 99, 100

Boltz-2 is antibody-trained and uses a calibrated confidence head. Strict-pass designs (complex_pLDDT ≥ 85, ipTM ≥ 0.7, at least 4 hotspot hits) are the closest equivalent to AF2-multimer "high confidence". Use the agreement / disagreement with your AF2 score as the orthogonal signal.

Candidates (12) with MSA

Returned 12 / 12

#	ipTM ?	pTM ?	pLDDT ?	Hotspot hits ?	vs. quality bar ?	3D	Structure
1 Top	0.952	0.973	96.1	13	Meets ipTM 0.7, pLDDT 80 and Hotspot hits 4	—	—

2	0.950	0.965	96.6	13	Meets ipTM 0.7, pLDDT 80 and Hotspot hits 4	—	—

3	0.950	0.968	97.1	13	Meets ipTM 0.7, pLDDT 80 and Hotspot hits 4	—	—

4	0.942	0.963	95.3	13	Meets ipTM 0.7, pLDDT 80 and Hotspot hits 4	—	—

5	0.940	0.963	96.7	13	Meets ipTM 0.7, pLDDT 80 and Hotspot hits 4	—	—

6	0.939	0.959	95.1	13	Meets ipTM 0.7, pLDDT 80 and Hotspot hits 4	—	—

7	0.936	0.963	95.7	13	Meets ipTM 0.7, pLDDT 80 and Hotspot hits 4	—	—

8	0.935	0.960	96.4	13	Meets ipTM 0.7, pLDDT 80 and Hotspot hits 4	—	—

9	0.910	0.948	95.3	14	Meets ipTM 0.7, pLDDT 80 and Hotspot hits 4	—	—

10	0.909	0.942	96.2	13	Meets ipTM 0.7, pLDDT 80 and Hotspot hits 4	—	—

11	0.886	0.958	91.3	13	Meets ipTM 0.7, pLDDT 80 and Hotspot hits 4	—	—

12	0.874	0.941	94.8	13	Meets ipTM 0.7, pLDDT 80 and Hotspot hits 4	—	—

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

Hotspot contacts (top 12 designs by ipTM)

Design	54	57	58	61	62	67	72	75	86	91	93	96	99	100	Hits
design 01															13 / 14
design 02															13 / 14
design 03															13 / 14
design 04															13 / 14
design 05															13 / 14
design 06															13 / 14
design 07															13 / 14
design 08															13 / 14
design 09															14 / 14
design 10															13 / 14
design 11															13 / 14
design 12															13 / 14

How to read that

Read the numbers against the scale, not against zero. Folded through the identical path, three published MDM2 binders scored: the native p53 peptide 0.933–0.941, PDI 0.930–0.935, and PMI 0.905–0.912, each across three seeds. The designs sit inside that band and the better half sit above it — which is the useful reading of 0.94, rather than “0.94 sounds high”. Now the part worth carrying away: PMI is the tightest of the three at roughly 3 nM, an order of magnitude better than the native peptide, and it scored the LOWEST of the three. ipTM tells you the model is confident these two things form the complex you asked about. It does not rank affinity, and nothing on this page does. Treat a good score as a reason to make the molecule, never as a predicted K₁.

What happened next

Took the panel into a developability screen — several of the twelve were then rejected on protease liability, which the interface score says nothing about — and carried what survived toward synthesis. On your own candidates the same shape works: fold the batch, keep what clears the bar, and let the next filter be the one this score cannot see.

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 Boltz-2 and a sibling algorithm, these run on the same hub against the same target.

Run AlphaFold2 online: You have a sequence and want the most trusted 3D prediction of it, with per-residue confidence you can act on. It searches for related natural sequences first, which is where the accuracy comes from and where the time goes. For a faster answer use ColabFold or ESMFold; to score a binder against its target use Boltz-2.
Run ColabFold online: You have a sequence and want its 3D shape quickly, trading a little accuracy for speed. It skips the search for related natural sequences that full AlphaFold2 runs. Use it to triage a batch; use AlphaFold2 when the answer has to be right.
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.