SAbDab2: the structural antibody database in the age of machine learning.

SAbDab2 re-engineers the Structural Antibody Database to support new antibody formats and provide ML-grade, standardized, versioned train/test splits for antibody and complex structures.

• Why it matters: Antibody research has expanded into many formats, while machine learning has increased the need for consistent, high-quality structure datasets that are easy to retrieve and compare.
• What they did: The team added broad format support and streamlined retrieval/comparison of all known structures for a given antibody, plus ML-grade antibody and antibody–antigen-complex access with versioned train/test splits updated every 6 months.
• The result: SAbDab2 is updated weekly and is freely available, with ML splits released via Zenodo (https://zenodo.org/records/20083994), enabling reproducible model training and evaluation.

AI summary of the abstract · may contain errors · not medical advice · read the paper.