Antibody structure prediction

What is antibody structure prediction?

Antibody structure prediction is the computational process of building three-dimensional antibody variable domains from heavy- and light-chain sequences. It gives specialized attention to conserved frameworks and diverse complementarity-determining-region loops, with CDR-H3 commonly remaining the most uncertain loop.

Antibody-specialized models use the conserved immunoglobulin fold while learning variable loop geometries that shape antigen recognition. Correct chain pairing and complete variable-domain boundaries are foundational requirements.

Read confidence at residue and loop level. A strong framework does not guarantee an accurate CDR-H3 conformation, and an unbound model can rearrange upon antigen binding.

When to use antibody structure prediction

Use antibody models when paired variable-domain geometry matters:

• VH and VL sequences are available for one antibody.
• CDR mutations need a structural context.
• You want to compare uncertain loops before docking to an antigen.

Strengths and limitations

Antibody-specialized predictors use the conserved framework to focus on variable-domain packing and CDR geometry. Comparing two model families can identify loops whose interpretation depends on the predictor.

The conserved framework may look reliable while CDR-H3 remains uncertain. Incorrect heavy–light pairing invalidates the interface, and antigen binding may change loop conformations. A predicted unbound paratope does not establish recognition.

Antibody-model interpretation

ABodyBuilder3 and ImmuneBuilder are trained for immune-receptor geometry rather than general proteins. Compare framework alignment and each CDR separately; numbering with ANARCI can map equivalent positions, while surface tools contextualize rather than validate coordinates.

Applications of antibody structure prediction

A variable-domain model can help map CDR mutations, inspect framework packing or prepare a structural hypothesis for antigen-binding studies. Paired-chain input matters because heavy–light orientation shapes the paratope.

Comparing antibody-specialized models is particularly useful at CDR-H3, where sequence diversity and loop flexibility make confident-looking predictions easy to overinterpret.

How to run antibody structure prediction online

1. Confirm chain pairing. Provide the correct VH and VL sequences and boundaries as separate FASTA inputs.
2. Run antibody models. Generate a paired variable-domain model with ABodyBuilder3.
3. Compare the second branch. Review ImmuneBuilder with its configured antibody model and keep confidence outputs by method.
4. Inspect loops. Compare each CDR and framework region separately, noting uncertain H3 conformations and chain orientation.
5. Export for validation. Retain paired sequences, numbering context, structures and confidence for antigen-bound or experimental follow-up.

How to interpret antibody structure prediction results

Framework confidence and loop confidence must be read separately. Two models can agree on the conserved fold while disagreeing at the residues that govern recognition. Neither result establishes the bound paratope conformation.

Use CDR differences to prioritize structural or mutational tests. An unbound antibody model does not confirm antigen affinity, specificity or function.

How antibody structure prediction works

Send paired VH and VL sequences to ABodyBuilder3 and ImmuneBuilder and retain both structures and confidence estimates.

1. Verify chain identity. Confirm the heavy and light sequences are correctly paired and cover the intended variable domains.
2. Trim variable domains. Provide VH and VL as separate workflow inputs.
3. Run antibody models. Run ABodyBuilder3 and ImmuneBuilder with their documented model settings.
4. Inspect CDR confidence. Compare framework superposition, CDR-loop geometry, and residue-level confidence.
5. Prepare downstream validation. Export both structures and use antigen-bound data, mutagenesis, or experimental structures for validation.

Inputs and outputs

Check formats before running, then inspect and download the result from every workflow step.

Inputs

Research input

FASTAPDBmmCIF

Paired VH and VL amino-acid sequences with correct variable-domain boundaries.

Outputs

Prediction and review outputs

PDBmmCIFCSVJSON

Predicted antibody PDB structures with model-native per-residue confidence or error estimates.