ABodyBuilder2: Predicted Antibody Structures from Observed Antibody Space

University of Oxford researchers used the ABodyBuilder2 model to predict approximately 1.5 million paired antibody structures from sequences in the Observed Antibody Space database. The dataset was created by Alexander Greenshields‐Watson. Structures are organized in folders corresponding to their original OAS source files and studies.

Use Cases

• Benchmarking antibody structure prediction models based on a large-scale set of predicted structures.
• Training machine learning models for antibody design based on predicted structural features.
• Analyzing structural diversity and properties of antibodies based on the predicted point cloud data.

Strengths

• Contains approximately 1.5 million predicted antibody structures, providing a large-scale resource.
• Structures are derived from the widely-used Observed Antibody Space (OAS) sequence database.
• Data organization mirrors the source OAS file and study structure, aiding traceability.

Limitations

• Row count for the final structured data is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

University of Oxford

Collection Method

Computational modeling using ABodyBuilder2 on paired antibody sequences from Observed Antibody Space.