PolyXpert
Sequence-based antibody polyreactivity prediction using a fine-tuned ESM-2 model.
Developer: Zhou et al.; PolyXpert researchers
Overview
- Developer
- Zhou et al.; PolyXpert researchers
- Category
- developability and stability
- Commercial rights
- Unverified
- License fee
- Unverified
- Open weights
- Unverified
- Last verified
- 2026-10-18
- Added
- 2026-10-18
Capabilities
Uses antibody VH and VL sequences to predict polyreactivity risk without a 3D structure. Authors report machine-learning evaluation; a prediction does not establish experimental polyspecificity or clinical developability.
When to use it
Flag candidate IgGs with possible off-target/polyreactivity liabilities before experimental polyspecificity panels; use orthogonal assays to confirm.
Compute
Compute category: unverified. See original installation instructions or platform terms for specific hardware costs and runtime.
Example use cases
- Flag candidate IgGs with possible off-target/polyreactivity liabilities before experimental polyspecificity panels; use orthogonal assays to confirm.
Source-based workflow description. No experimental performance ranking is implied.
Known limitations
- An MIT GitHub project does not by itself establish commercial rights in pretrained upstream ESM-2, datasets, downloaded weights or downstream clinical use.
Check the original linked sources and the terms of every model dependency.
Commercial licensing
Commercial permission — Unverified
MIT source repository; finetuned weights, training data and model dependencies not fully licensed here.
An MIT GitHub project does not by itself establish commercial rights in pretrained upstream ESM-2, datasets, downloaded weights or downstream clinical use.
Commercial fee — Unverified
No complete model + checkpoint + cloud compute + service charges have been verified. Public research software or downloadable weights do not imply a free commercial deployment.
Official references
Validate your designed binder.
Seq2Study coordinates expression, affinity, stability and developability testing for AI-designed antibodies and proteins.