Antibody & Biologics · Hummer et al., DTU
AntiFold
An antibody inverse folding model that designs CDR sequences tailored to antibody backbone frameworks.
What AntiFold is
AntiFold is an inverse folding model designed for antibody engineering, developed by Hummer and colleagues at the Technical University of Denmark (DTU). It focuses specifically on redesigning antibody backbones and complementarity-determining region (CDR) loops, which are critical loops responsible for antigen recognition.
Standard inverse folding algorithms such as ProteinMPNN often struggle when applied to antibody CDR loops because these loops display atypical structural conformations compared to typical globular proteins. AntiFold addresses this structural divergence by training specifically on antibody structures, allowing it to propose sequence variants that conform to the canonical framework of the antibody.
How it works
The model is trained specifically on antibody structural data to capture the distinct geometry of antibody backbones and CDR loops. Given an input antibody coordinate file and a specification of the regions targeted for modification, AntiFold generates novel sequence variants while scoring each position with log-odds metrics.
Key facts
- Version
- 1.0
- Domain
- Antibody
- Supported tasks
- Inverse folding, CDR design
- Input modalities
- Protein structure (PDB)
- Hardware requirements
- 8 GB GPU memory
What you can use it for
- CDR loop redesign
- Generating tailored amino acid sequences for specified CDR regions to fit an antibody's existing structural scaffold.
- Antibody backbone inverse folding
- Generating candidate sequence variants based on fixed antibody backbone structures.
- Variant sequence scoring
- Evaluating candidate residues within redesigned regions using per-position log-odds values.
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
AntiFold
Strengths and limits
Good at
- Trained specifically on antibody structures to handle the structurally atypical conformations of CDR loops.
- Generates sequence designs that respect the antibody's canonical framework.
- Outputs per-position log-odds metrics alongside generated sequences to assist variant evaluation.
Watch out for
- Requires pre-existing three-dimensional coordinates in PDB format.
- Specialised for antibody architectures rather than general non-antibody protein folds.
- Requires hardware with at least 8 GB of GPU memory.
Inputs and outputs
You provide
Antibody structure (PDB)
Regions to redesignoptional
Comma-separated. CDR1, CDR2, CDR3 or FR1–FR4; CDRH3 and CDRL3 are understood too.
How many designsoptional
You get back
Redesigned regions with per-position log-odds.
Licence
Commercial use permitted · BSD-3-Clause
The BSD-3-Clause licence allows use in commercial research and products.
Questions
Can I use AntiFold commercially?
Commercial use permitted. The BSD-3-Clause licence allows use in commercial research and products.
How do I run AntiFold 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 16GB, typically in 5.0s, and a run costs about 28 credits. Visitors without an account get 1,000 credits a month.
What inputs does AntiFold require?
It takes an antibody structure in PDB format, the textual specification of regions to redesign, and the requested number of sequence candidates.
Why use AntiFold rather than a general tool like ProteinMPNN?
General inverse folding models like ProteinMPNN struggle with CDR loops because they are structurally atypical; AntiFold is trained specifically on antibody structures to produce sequences that match canonical frameworks.
What information is provided in the model outputs?
The model outputs a table of redesigned sequence regions containing the designed sequences alongside per-position log-odds values.
Written from AntiFold's own documentation. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.