Protein Structure · biohub
ESMFold2-Fast
Predicts all-atom 3D protein structures rapidly from single amino acid sequences for high-throughput screening.
What ESMFold2-Fast is
ESMFold2-Fast is an inference-optimised protein structure prediction model developed by biohub. It is an accelerated variant of ESMFold2 designed for high-throughput structure prediction across large sets of proteins while maintaining strong accuracy, generating full atomic coordinates directly from primary amino acid sequences.
Unlike the full ESMFold2 model, which can use multiple sequence alignments (MSAs) to aid prediction on difficult targets, ESMFold2-Fast operates solely on single sequences without MSA conditioning. This architecture avoids the computational overhead of searching and aligning sequence databases, enabling rapid structural characterisation at scale.
The model outputs complete all-atom 3D coordinates, encompassing both backbone and side-chain atoms, alongside detailed reliability metrics including residue-level confidence scores (pLDDT) and predicted alignment error metrics (pTM).
How it works
ESMFold2-Fast executes single-sequence structure prediction without MSA inputs. The model computes 3D coordinates directly from the input sequence through iterative sampling loops and inference steps, generating full-atom backbone and side-chain geometries.
The model was trained on structural data from the Protein Data Bank (PDB) and the AlphaFold DB (AFDB), with a training data cutoff of September 2021.
Key facts
- Task
- Single-sequence protein structure prediction
- MSA conditioning
- None (single sequence only)
- Output format
- All-atom coordinates (PDB format), pLDDT, and pTM
- Training data
- Protein Data Bank (PDB) and AlphaFold DB (AFDB)
- Training data cutoff
- September 2021
- GPU memory
- 16 GB
What you can use it for
- High-throughput proteome folding
- Screen large batches of uncharacterised protein sequences to predict all-atom 3D models without generating multiple sequence alignments.
- Structure-based protein design
- Rapidly evaluate the structural conformations of de novo designed protein sequences and assess model confidence via pLDDT and pTM scores.
- Residue confidence assessment
- Analyse local structural reliability and disorder across a target chain using predicted per-residue pLDDT metrics.
Run it here
No installation and no GPU of your own: fill in the inputs and it runs on our servers. Results stay in your browser unless you sign in.
Runs on On-demand GPU
Runs on our servers with on-demand compute. A first run needs time to load the model; active capacity can be reused and scales down when idle.Run this model
ESMFold2-Fast
Strengths and limits
Good at
- Generates all-atom 3D structures from single amino acid sequences without requiring multiple sequence alignments.
- Optimised for fast inference, facilitating high-throughput structure prediction across large collections of proteins.
- Provides confidence estimations alongside atomic coordinates, including residue-level pLDDT and overall pTM scores.
Watch out for
- Does not support multiple sequence alignment conditioning, which can limit accuracy on challenging targets.
- May inherit biases from the PDB and AFDB training sets, leading to variable performance on underrepresented structural classes, disordered regions, or post-translational modifications.
- Predictions represent computational hypotheses and cannot substitute for experimental structure determination techniques such as cryo-EM, X-ray crystallography, or NMR.
Inputs and outputs
You provide
Protein sequence
One-letter amino-acid codes.
You get back
All-atom coordinates in PDB format, viewable and downloadable.
Residue count, mean pLDDT (0–100) and pTM.
Licence
Check the licence before commercial use · See model card
The See model card licence carries terms worth reading before you depend on it.
Questions
Can I use ESMFold2-Fast commercially?
Check the licence before commercial use. The See model card licence carries terms worth reading before you depend on it.
How do I run ESMFold2-Fast without installing anything?
Use the form under “Run it here” on this page: fill in the inputs and press Run model. It runs on our servers on CPU only, typically in 12s, and a run costs about 28 credits. Visitors without an account get 1,000 credits a month.
Does ESMFold2-Fast require a multiple sequence alignment (MSA)?
No. ESMFold2-Fast is an inference-optimised variant designed specifically for single-sequence prediction and does not accept or condition on MSAs.
What confidence metrics are produced with the structure?
The model outputs residue-level confidence via pLDDT scores (0 to 100) and overall structure confidence via predicted TM-score (pTM).
What hardware memory is needed to run the model?
The model requires 16 GB of GPU memory.
What training data was used for ESMFold2-Fast?
It was trained on protein structures from the Protein Data Bank (PDB) and the AlphaFold DB (AFDB) up to a September 2021 cutoff date.
How to cite it
Candido, Salvatore and Hayes, Thomas and Derry, Alexander and Rao, Roshan and Lin, Zeming and Verkuil, Robert and Wu, Bryan and Lee, Jin Sub and Bruguera, Elise S. and Keval, Jehan A. and Kopylov, Mykhailo and Pak, John E. and Wu, Wesley and Thomas, Neil and Mataraso, Samson and Hsu, Alvin and Trotman-Grant, Ashton C. and Fatras, Kilian and dos Santos Costa, Allan and Badkundri, Rohil and Akın, Halil and Oktay, Deniz and Deaton, Jonathan and Montabana, Elizabeth and Sitwala, Hrishita and Yu, Yue and Wiggert, Marius and Carlin, Dylan Alexander and Goering, Anthony W. and Blazejewski, Tomasz and Sandora, McCullen and Hla, Michael and Jia, Tina Z. and Kloker, Leon H. and Sofroniew, Nicholas J. and Uehara, Masatoshi and Pannu, Jassi and Bachas, Sharrol and Liu, Daniel S. and Sercu, Tom and Rives, Alexander. Language Modeling Materializes a World Model of Protein Biology. bioRxiv, 2026. https://www.biorxiv.org/content/10.64898/2026.06.03.729735.
Written from ESMFold2-Fast's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.