Unlock the Secrets of Antibody-Antigen Interactions with AI
We are excited to announce the SEPIQ VHH-Epitope-Prediction Challenge, scheduled to begin in June 2026. This challenge invites researchers, bioinformaticians, and AI practitioners to tackle a critical problem in structural biology: the blind prediction of antigen epitopes based on VHH antibody sequences.
Challenge Overview:
- Goal: Predict epitopes for 7 specific VHH antibodies.
- Data Provided: Extensive sequence data with metadata for hundreds of thousands of VHH clones (both binders and non-binders) targeting the Her2 antigen. This dataset also includes competition data with known antibodies, which can be useful for epitope prediction.
- Evaluation: Binary assessment of amino acid residues in the antigen-antibody interface Track A (residue-level resolution; SEPIQ), where participants submit per-residue epitope/paratope probabilities, and Track B (atomic-level resolution; CAPRI), where participants submit predicted complex structures.
Why Participate?
- Contribute to cutting-edge research in antibody design.
- Test your models against high-quality experimental ground truth.
- Results are planned to be submitted to NeurIPS and Protein Science.
Organized by: Epitope Prediction Challenge Committee, SEPIQ (London)
Co-organized by: CAPRI, OpenFold, AMED-BINDS and COGNANO, Inc.
Sponsors: To Be Confirmed
Join us in advancing the frontier of AI-driven drug discovery!
Challenge | Blind prediction of antigen epitopes based on the amino sequences of seven given VHH antibodies. |
Provided Data | Sequence data(with metadata) for over a million VHH clones — both binders and non-binders — against the Her2 antigen. This dataset also includes competition data with known antibodies, which can be useful for epitope prediction. |
Evaluation Metrics | Based on CAPRI standards; structural accuracy evaluated for both backbone and side chains using metrics such as DockQ. |
Ground Truth | Cryo-EM structural analysis of Her2-VHH complexes, currently in progress. |
Organizer | Epitope Prediction Challenge Committee, SEPIQ (London) |
Co-organizer | CAPRI, OpenFold, AMED-BINDS and COGNANO, Inc. |
Judging Panel | Multiple experts in structural biology and AI (under coordination) |
Supporters / Endorsers | To Be Confirmed |
Publication Venues | Planned submission to NeurIPS, Protein Science(tentative) |
Sponsors | To Be Confirmed |
Foundation | London |
Timeline
FAQ
- What exactly is the prediction task?
It involves predicting the 3D structure of the antigen-antibody complex (docking), not just identifying epitope residues. Evaluation will be based on structural accuracy (DockQ, etc.).
- What data will be provided?
We will provide sequence data for VHH clones. Details on whether antigen structure (PDB) is provided or must be predicted will be released with the full guidelines.
- How will the ground truth be determined?
Ground truth is based on Cryo-EM structures (target resolution 3.0 Å).
- Can we use tools like AlphaFold?
Yes, existing tools are permitted. Specific rules on code submission will be provided.
Co-Organizers
Celebrate SEPIQ challenge
The performance by Markus Placci in SEPIQ Challenge Opening Ceremony in Kyoto on 28th March 2026
Sponsors
Gold
Coming soon
Silver
Coming soon
Bronze
Coming soon
In-kind
What is SEPIQ?
SEPIQ (Sequence-based Epitope Prediction Intelligent Query) is a London-based public organization dedicated to leading structure-based biopharmaceutical discovery.
Binary assessment of amino acid residues in the antigen-antibody interface
Track A (residue-level resolution; SEPIQ), where participants submit per-residue epitope/paratope probabilities, and Track B (atomic-level resolution; CAPRI), where participants submit predicted complex structures.