Protein Structure · AlQuraishi Lab / Columbia
OpenFold
Predicts three-dimensional protein structures from amino acid sequences using a retrained PyTorch implementation.
What OpenFold is
OpenFold is a trainable PyTorch reproduction of AlphaFold 2 developed by the AlQuraishi Lab at Columbia University. It addresses computational protein structure prediction, generating spatial coordinates for all atoms in a protein directly from its primary sequence.
The system matches AlphaFold 2 prediction accuracy following a clean-room retraining. In addition to inference, it provides structural biologists and computational researchers with an environment to study protein folding mechanisms and train models at scale alongside resources like OpenProteinSet.
How it works
OpenFold implements the AlphaFold 2 architecture within the PyTorch framework. The model was trained from scratch using a clean-room retraining methodology and pairs with datasets such as OpenProteinSet to learn three-dimensional structural representations from sequence data.
Key facts
- Version
- 2.0
- Architecture framework
- PyTorch
- Input modality
- Protein sequence
- Output modalities
- All-atom PDB coordinates, per-residue pLDDT scores
- GPU memory requirement
- 40 GB
What you can use it for
- Protein Structure Prediction
- Generates all-atom coordinate files in PDB format from single protein amino acid sequences.
- Local Prediction Quality Assessment
- Produces per-residue predicted Local Distance Difference Test (pLDDT) confidence scores between 0 and 100 to pinpoint well-ordered and reliable structural regions.
- Folding Mechanism Research and Retraining
- Allows researchers to train structure prediction pipelines from scratch or fine-tune models to examine how neural networks learn protein geometries.
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.
Not runnable yet
This model isn't available to run right now.Run this model
OpenFold
Strengths and limits
Good at
- Reaches structural accuracy matching AlphaFold 2 through an independent retraining process.
- Provides a fully trainable PyTorch architecture rather than an inference-only pipeline.
- Provides per-residue pLDDT confidence scores to highlight reliable regions across predicted structures.
Watch out for
- Requires high-specification hardware with at least 40 GB of GPU memory.
- Structural predictions are less reliable in regions where per-residue pLDDT scores fall below 70.
Inputs and outputs
You provide
Protein sequence
Single-letter amino acid code. Whitespace and FASTA headers are stripped automatically.
You get back
All-atom coordinates in PDB format, viewable and downloadable.
pLDDT score per residue (0–100). Above 70 is generally reliable.
Licence
Commercial use permitted · Apache-2.0
The Apache-2.0 licence allows use in commercial research and products.
Questions
Can I use OpenFold commercially?
Commercial use permitted. The Apache-2.0 licence allows use in commercial research and products.
How do I run OpenFold 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 1× A100 40GB, typically in 2.0min, and a run costs about 336 credits. Visitors without an account get 1,000 credits a month.
What inputs does OpenFold take?
OpenFold takes a primary protein amino acid sequence as its input.
What files and metrics does OpenFold output?
It generates all-atom three-dimensional atomic coordinates in standard PDB format along with a per-residue pLDDT confidence score vector.
How should pLDDT scores be interpreted?
pLDDT is a local confidence score ranging from 0 to 100. Scores above 70 are generally considered reliable.
Is OpenFold limited strictly to running inference?
No, it is built as a trainable PyTorch implementation designed for both inference and training from scratch.
How to cite it
Ahdritz, Gustaf, Nazim Bouatta, Christina Floristean, Sachin Kadyan, Qinghui Xia, William Gerecke, Timothy J. O'Donnell, Daniel Berenberg, Ian Fisk, Niccolò Zanichelli, Bo Zhang, Arkadiusz Nowaczynski, Bei Wang, Marta M. Stepniewska-Dziubinska, Shang Zhang, Adegoke Ojewole, Murat Efe Guney, Stella Biderman, Andrew M. Watkins, Stephen Ra, Pablo Ribalta Lorenzo, Lucas Nivon, Brian Weitzner, Yih-En Andrew Ban, Peter K. Sorger, Emad Mostaque, Zhao Zhang, Richard Bonneau, and Mohammed AlQuraishi. "OpenFold: Retraining AlphaFold2 yields new insights into its learning mechanisms and capacity for generalization." bioRxiv (2022): 2022.11.20.517210. doi:10.1101/2022.11.20.517210.
Written from OpenFold's own documentation (code). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.