Protein Structure · biohub
ESMFold2-Experimental
Predicts all-atom protein structures and confidence metrics from sequences for experimental evaluation.
What ESMFold2-Experimental is
ESMFold2-Experimental is an experimental protein structure prediction model developed by Biohub. It was released to support the reproducibility of research exploring modified training recipes and architectural adjustments to the ESMFold2 framework.
The model accepts an amino acid sequence and produces predicted all-atom three-dimensional coordinates alongside structure confidence estimates, including per-residue predicted local distance difference test (pLDDT) scores and predicted template modelling (pTM) scores. It provides an avenue to evaluate experimental updates to single-sequence and multiple sequence alignment (MSA) conditioned structure prediction.
How it works
The model is a variant of the ESMFold2 architecture that investigates altered model design and training protocols for protein structure generation, sequence design, and confidence scoring. Depending on the ESMFold2 formulation, prediction can operate directly on single sequences or be conditioned on multiple sequence alignments (MSAs) to handle difficult structural targets.
Key facts
- Domain
- Protein structure prediction
- Input modality
- Protein sequence
- Output formats
- PDB coordinates, residue count, mean pLDDT, and pTM
- Memory requirement
- 16 GB GPU memory
What you can use it for
- All-atom structure generation
- Generating 3D all-atom structural coordinates in PDB format directly from raw protein sequences.
- Structural confidence estimation
- Assessing predicted structure quality through mean pLDDT (on a 0–100 scale) and overall pTM metrics.
- Methodological reproducibility
- Benchmarking and reproducing study outcomes evaluating modified training recipes and architecture changes against standard ESMFold2.
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-Experimental
Strengths and limits
Good at
- Provides all-atom structural predictions together with standard confidence metrics including pLDDT and pTM.
- Enables direct reproduction and evaluation of experimental architecture changes and training recipes.
- Operates within a modest 16 GB GPU memory footprint.
Watch out for
- Intended primarily for reproducibility of published research rather than routine research modelling, which is directed to the standard ESMFold2 model.
- Outputs and behaviours may deviate from standard pipeline specifications due to its experimental status.
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-Experimental 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-Experimental 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.
What inputs does ESMFold2-Experimental accept?
The model takes an amino acid sequence as a raw text string.
What outputs are generated by the model?
It generates all-atom atomic coordinates in PDB format, accompanied by the total residue count, mean pLDDT score (0–100), and pTM score.
Should this model be used for general research structure prediction?
No. The standard ESMFold2 model is recommended for primary research tasks, as this experimental variant was released specifically to support reproducibility of experimental architectural and training investigations.
Can the model handle multiple sequence alignments?
ESMFold2 supports both single-sequence prediction and MSA-conditioned prediction to improve accuracy on difficult targets.
How to cite it
Written from ESMFold2-Experimental'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.