Can I run AlphaFold2 multimer online without a local GPU?

Yes. Ranomics Tools runs AlphaFold2 (via the ColabFold implementation) on a dedicated GPU through your browser, with full MSA search and template support. Results land on a job page with ipTM, pLDDT, and pAE plots.

How is this different from running ColabFold yourself?

Same underlying weights and pipeline, but you skip CUDA setup, MMseqs2 round-trip on your laptop, and the wait for the public Colab queue. You also get a persistent job page you can share or hand off into ProteinMPNN or BindCraft.

How much does an AlphaFold2 multimer run cost?

You pay for the GPU time your run uses, from your wallet. You can top up from $5, which is enough to start a run.

AlphaFold2

AlphaFold2 is a no-install online structure prediction tool you can run through tools.ranomics.com on a dedicated GPU. The reference-standard fold, with the homolog search and templates that make it accurate on natural proteins, and a multimer mode that predicts several chains together rather than one at a time.

Example run — Ten designed variants of a microbial rhodopsin — the seven-transmembrane-helix light-driven channel family — at 333 residues each, folded in one submission. See what came back →

Prediction inputs

Preset

Standalone, one FASTA Paste or upload FASTA. Multi-record FASTA is treated as one multimer fold (chains joined). Up to 1500 AA total. Runs on an A100-80GB. Batch for many fold targets Fold up to 50 independent targets in one job. Each record can be a monomer or a multimer (use : inside a record to break chains). Slowest of the three structure-prediction tools.

FASTA

Paste single-chain or multimer FASTA. For multimer, use one >header line per chain. Up to 1500 AA total across chains on the atomic tier.

Or upload a .fasta file:

Fold targets (FASTA or one per line)

Up to 50 independent fold targets per submit. Each record is one fold; up to 1500 AA total per record (across chains). For a multimer record, separate chains with : on the sequence line (the ColabFold-multimer convention). Lines without a > header are auto-named. Detected records: 0.

Number of recycles

How many AF2 recycles to run (1 to 5). More recycles use more GPU time but converge marginal cases. Default 3 matches the AF2 paper and is a good balance.

Use PDB templates

Fetch homologous structures from PDB as templates for the prediction. Turn off for orphan sequences or when you want a pure sequence-only fold.

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 AlphaFold2

What it is

Predicts the 3D structure of a protein from its sequence — one chain, or several chains folded together as a complex. It searches public databases for related natural sequences first, which is where most of its accuracy comes from and most of its runtime goes, then returns a structure with per-residue confidence (pLDDT) and an estimate of the error between any two residues (PAE). Both are calibrated, meaning the numbers mean what they say. AlphaFold2, Jumper et al., Nature 2021, packaged via ColabFold (Mirdita et al., Nature Methods 2022).

When to use

You have a sequence and want the most trusted structure prediction available, with confidence numbers you can act on.
Your target is a single chain, or a small complex of two to four chains.
You can wait for the homolog search and three refinement passes each run makes.

Prerequisites

Single-letter FASTA sequence(s). Multimers separated by : or pasted as multi-record FASTA.
A stable target topology. AF2 underperforms on intrinsically disordered or flexible regions.

Input glossary

Sequence. Paste FASTA. Use : as a chain separator for multimers (e.g. SEQ_A:SEQ_B).
Recycles. Number of model recycles. 3 is the AF2 default; lower is faster but trades a small amount of accuracy.

Predicted PDB with per-residue pLDDT, pairwise PAE, and pTM (or ipTM on a multimer). An ipTM of 0.6 or more is a plausible interface and 0.75 or more is strong. On a run of your own the structures download from the results page. The PAE matrix downloads only from a standalone run — a batch returns no PAE file.

References

A past run we actually did

The target

Ten designed variants of a microbial rhodopsin — the seven-transmembrane-helix light-driven channel family — at 333 residues each, folded in one submission. They came off a generative model as ten samples from one design round, each 72% to 79% identical to the first.

A membrane protein is a hard case for a structure predictor, and ten near-siblings is the shape a real design round actually arrives in. The interesting part is not whether they fold — it is what you are entitled to conclude from a table of ten scores, which is less than the table's own layout suggests.

What we put in the form

Preset — Batch for many fold targets: Ten sequences in one submission rather than ten submissions. Same cost per sequence, one job to watch.
Fold targets (FASTA or one per line) — 10 records, one chain each: Whether a record folds as a monomer or a multimer is not a setting — it follows from the record itself, and joining chains with a colon is what makes one a multimer. These are single chains, so all ten folded as monomers, which is why the ipTM column stays blank throughout: there is no second chain, so there is no interface to score. Blank is not a low score.
Number of recycles — 3: The default. Each pass lets the model refine its own answer; three is the standard setting behind published AF2 numbers.
Use PDB templates — on: Known structures are allowed as scaffolding. The rhodopsin fold is well represented in the PDB, so there is real signal to use.

What came back

All ten folded, no failures. Mean pLDDT runs from 74.16 to 75.93 and pTM from 0.71 to 0.73. The table sorts by mean pLDDT and highlights the top row — but the top row beats the bottom row by 1.77 pLDDT points, and the whole pTM column spans 0.02. Every one of the ten landed in the same place.

AF2 batch batch

AlphaFold2 + ColabFold MMseqs2 MSA per record. Sorted by mean pLDDT — values above 70 are confidently folded; multimer records additionally surface ipTM (interface confidence) where applicable.

Candidates (10) batch

Returned 10 / 10

#	mean_pLDDT ?	ipTM ?	pTM ?	AA	chains	3D	Structure
1 Top	75.93	—	0.730	333.00	1.00	—	—

2	75.71	—	0.730	333.00	1.00	—	—

3	75.37	—	0.730	333.00	1.00	—	—

4	75.17	—	0.730	333.00	1.00	—	—

5	75.16	—	0.730	333.00	1.00	—	—

6	75.04	—	0.710	333.00	1.00	—	—

7	74.88	—	0.730	333.00	1.00	—	—

8	74.57	—	0.720	333.00	1.00	—	—

9	74.50	—	0.710	333.00	1.00	—	—

10	74.16	—	0.720	333.00	1.00	—	—

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 highlight on the top row is a sort order, not a verdict. A ranked table implies the first row is better than the last, and across a spread this narrow it simply is not: re-run the same ten sequences with a different seed and the order would reshuffle. Nothing here distinguishes these designs, and picking the top one to carry forward is picking at random while feeling informed. The result that is real is the agreement. Ten variants differing at a fifth to a little over a quarter of their residues all folding to the same confidence says the scaffold tolerates that much variation — which is a genuine finding about the design round, and the opposite of what you would conclude by reading row one. For scale: 70 is the usual line for a confidently folded structure, and all ten clear it. Compare with the ESMFold example on this site, where a genuinely disordered protein has no residue reaching 70 anywhere in the chain.

What happened next

A batch this tight does not narrow itself, so the choice has to come from somewhere other than these scores — a property the table does not carry, or an experiment. The runs worth doing next are the ones that can separate the ten: fold them against the partner or ligand that matters if there is one, which turns a blank ipTM column into a real interface number, or take several forward together rather than betting on row one. When a screen comes back this uniform, that is the screen telling you it has finished its job.

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 AlphaFold2 and a sibling algorithm, these run on the same hub against the same target.

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 ESMFold online: You have one protein sequence and want its 3D shape, fast. No search for relatives, so it works on designed or orphan sequences that have no natural family to align against. One chain only — for complexes use ColabFold or AlphaFold2.
Run ProteinMPNN online: You have a backbone — a 3D shape with no sequence decided yet — and need amino-acid sequences that will fold into it. You get ranked candidates to pick from. To generate the backbone in the first place, run a binder design tool and feed its PDB in here.

Questions people ask about AlphaFold2

Can I run AlphaFold2 multimer online without a local GPU?: Yes. Ranomics Tools runs AlphaFold2 (via the ColabFold implementation) on a dedicated GPU through your browser, with full MSA search and template support. Results land on a job page with ipTM, pLDDT, and pAE plots.
How is this different from running ColabFold yourself?: Same underlying weights and pipeline, but you skip CUDA setup, MMseqs2 round-trip on your laptop, and the wait for the public Colab queue. You also get a persistent job page you can share or hand off into ProteinMPNN or BindCraft.
How much does an AlphaFold2 multimer run cost?: You pay for the GPU time your run uses, from your wallet. You can top up from $5, which is enough to start a run.