Antibody & Biologics · Ruffolo et al., Johns Hopkins
IgFold
Predicts antibody 3D structures from heavy and light chains in seconds.
What IgFold is
IgFold is an antibody structure prediction model developed by Ruffolo et al. at Johns Hopkins. It takes heavy and light chain sequences to produce all-atom coordinate structures in PDB format alongside per-residue confidence metrics.
Antibodies typically share a conserved framework structure, and the difficult part of the prediction is the six hypervariable loops. IgFold pairs a language model pre-trained on 558 million natural antibody sequences with graph networks that predict backbone coordinates directly. Its paper reports structures of similar or better quality than alternative methods, AlphaFold included, in under 25 seconds.
How it works
A language model pre-trained on 558 million natural antibody sequences feeds graph networks that directly predict backbone atom coordinates.
Key facts
- Version
- 0.4
- Task
- Structure prediction
- Input modalities
- Heavy chain and light chain protein sequences
- Output modalities
- All-atom PDB structure and per-residue pLDDT vector
- GPU memory
- 8 GB
What you can use it for
- Rapid antibody structure modelling
- Generates antibody coordinate files from paired heavy and light chain amino acid sequences in seconds.
- Preparation for molecular docking
- Produces accurate antibody structures that can be fed into downstream computational docking studies.
- Developability screening
- Supplies structural models to assist with the evaluation of antibody developability characteristics.
Run it here
We list this model but cannot run it yet: its licence is non-commercial and commercial use needs a separate licence from Johns Hopkins.
For the same job, run ImmuneBuilder here instead. Everything else on this page still applies.
Strengths and limits
Good at
- Predicts structures of similar or better quality than alternative methods, AlphaFold included, in under 25 seconds, according to its paper.
- Provides per-residue pLDDT confidence scores to indicate local model reliability.
- Outputs complete all-atom coordinates in PDB format suitable for immediate visualization and downstream workflows.
Watch out for
- Specialised for antibodies and dependent on the presence of conserved antibody framework regions.
- Requires paired heavy and light chain sequence inputs rather than single-chain or non-antibody targets.
Inputs and outputs
You provide
Heavy chain
Single-letter amino acid code. Whitespace and FASTA headers are stripped automatically.
Light chainoptional
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
Academic and non-commercial use only · JHU Academic Software License (non-commercial)
Read 2026-11-04: Graylab/IgFold is under the JHU Academic Software License Agreement, which grants use "for non-commercial purposes" only; "commercial use of the Software requires a separately executed written license agreement". The README says the code and pre-trained models are available for non-commercial use, including at commercial entities, and points commercial inquiries to Johns Hopkins Technology Ventures.
Questions
Can I use IgFold commercially?
Academic and non-commercial use only. Read 2026-11-04: Graylab/IgFold is under the JHU Academic Software License Agreement, which grants use "for non-commercial purposes" only; "commercial use of the Software requires a separately executed written license agreement". The README says the code and pre-trained models are available for non-commercial use, including at commercial entities, and points commercial inquiries to Johns Hopkins Technology Ventures.
What inputs are required to run IgFold?
IgFold requires two sequence inputs: the antibody heavy chain sequence and the antibody light chain sequence.
What structural outputs does the model generate?
It provides all-atom coordinates in standard PDB format as well as a per-residue pLDDT confidence vector scored from 0 to 100.
How are the pLDDT confidence scores interpreted?
Scores range from 0 to 100, with values above 70 generally considered reliable.
How fast is IgFold compared to general-purpose structure prediction models?
Its paper reports structures of similar or better quality than alternative methods, including AlphaFold, in under 25 seconds.
Written from IgFold's own documentation (code, paper). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.