RosettaAntibody Decoy Structures for 49 Antibody Fv Domains

Approximately 2800 computationally generated decoy structures exist for each of 49 antibody Fv domains, created using the RosettaAntibody protocol with stringent template exclusion criteria. This dataset provides a benchmark for evaluating scoring functions by their ability to identify near-native conformations from decoys based on provided H3-loop RMSD values. The structures originate from a curated set first compiled by Marze et al. in 2016 and were used in prior studies of CDR-H3 loop modeling.

Use Cases

• Benchmarking scoring functions based on the provided Rosetta ref2015 scores and H3-loop RMSD values.
• Comparing the performance of different antibody structure prediction programs against a common set of decoys.
• Training machine learning models for antibody structure quality assessment using the decoy pool and associated scores.

Strengths

• Contains a large number of decoys, with about 2800 models generated for each of 49 distinct antibody targets.
• Includes precise quantitative annotations for each model: the Rosetta ref2015 score and the H3-loop RMSD from the experimental crystal structure.
• The modeling protocol used stringent homolog exclusion settings (over 95% identity for CDR templates, over 90% for FR templates), enhancing its value for method evaluation.

Limitations

• Key metadata such as exact row count, file size, and last update date are not provided across any source.
• The associated research study is described as unpublished, which may limit the interpretability of the dataset without final methodological details.
• The dataset appears to be structured for a specific computational pipeline (Rosetta), which may require adaptation for other software.

Provenance

Source

Johns Hopkins University (author Jeliazko R. Jeliazkov), generated using RosettaAntibody.

Collection Method

Computational homology modeling and H3-loop modeling based on the protocol from Weitzner, Jeliazkov, Lyskov et al. (Nature Protocols, 2017).

Time Range

The dataset file name suggests a generation date of March 23, 2020 (decoys_rosettaantibody_20200323.tar.gz).