Antibody & Biologics · Oxford Protein Informatics Group
AbLang2
An antibody language model for generating paired-chain sequence embeddings and guiding CDR design.
What AbLang2 is
AbLang2 is an antibody-specific language model developed by the Oxford Protein Informatics Group. General-purpose protein language models frequently underperform when applied to antibodies because the common immunoglobulin fold dominates their learned signals. AbLang2 is trained on the Observed Antibody Space (OAS) database to specifically capture antibody sequence variation.
The model natively handles paired heavy and light chains to yield joint sequence representations. These paired embeddings provide features tailored for downstream antibody engineering tasks, such as complementarity-determining region (CDR) design and fitness prediction.
How it works
AbLang2 is trained on antibody sequence data from the Observed Antibody Space (OAS). It accepts both heavy and light chain protein sequences alongside a user-selected operational mode, processing them together to generate a vector embedding of the paired-chain antibody.
Key facts
- Version
- 2.0
- Domain
- Antibody
- Input modalities
- Protein sequence (heavy and light chains)
- Training data
- Observed Antibody Space (OAS)
- GPU memory
- 8 GB
What you can use it for
- Paired-chain sequence embedding
- Generate vector representations of antibodies directly from paired heavy and light chain sequences.
- CDR design
- Provide antibody-specific representations to guide the design and engineering of complementarity-determining regions.
- Antibody fitness prediction
- Supply sequence embeddings to downstream models evaluating antibody mutational fitness and stability.
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
AbLang2
Strengths and limits
Good at
- Trained specifically on the Observed Antibody Space to overcome the immunoglobulin fold bias found in general protein language models.
- Natively supports paired heavy and light chain sequence inputs rather than single-chain sequences alone.
- Operates with a modest GPU memory requirement of 8 GB.
Watch out for
- Restricted to antibody sequence data and not intended for general, non-immunoglobulin proteins.
- Requires paired heavy and light chain sequence inputs to produce paired-chain vector embeddings.
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.
What to computeoptional
You get back
Paired-chain antibody representation.
Licence
Commercial use permitted · BSD-3-Clause
The BSD-3-Clause licence allows use in commercial research and products.
Questions
Can I use AbLang2 commercially?
Commercial use permitted. The BSD-3-Clause licence allows use in commercial research and products.
How do I run AbLang2 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 1.2s, and a run costs about 14 credits. Visitors without an account get 1,000 credits a month.
What inputs does AbLang2 accept?
AbLang2 requires a heavy chain sequence, a light chain sequence, and a mode parameter specifying what computation to perform.
What does the model output?
It outputs a vector embedding representing the paired heavy and light chain antibody.
Why use an antibody-specific model over a general protein language model?
General protein models are often dominated by the conserved immunoglobulin fold signal, whereas AbLang2 is trained directly on antibody diversity from the Observed Antibody Space.
What hardware is needed to run AbLang2?
The model specifies a requirement of 8 GB of GPU memory.
Written from AbLang2's own documentation. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.