RosettaAntibody generated models for a dataset of 49 antibody-Fv structures

Structures of antibody-Fv domains computationally generated by RosettaAntibody based on the protocol of Weitzner, Jeliazkov, Lyskov et al. (Nature Protocols 12, 401–416, 2017). There are 49 antibody targets, with about 2800 decoy structures provided per antibody. A table is also provided with the H3-loop rmsd of each structure from the experimental crystal structure.

These structures can be used to evaluate whether a score function can identify the near-native structures from the pool of decoys. These structures can also be used for comparison with other sets of structures generated by other antibody structure prediction programs.

The research study on this set is unpublished and a manuscript is under preparation. Please cite Jeliazkov, Frick, Zhou & Gray, “Robustification of RosettaAntibody and Rosetta SnugDock,” in preparation, 2020.

The homology modeling stage of RosettaAntibody was run with stringent homolog exclusion settings, excluding CDR templates of over 95% identity and FR templates of over 90% identity. The H3 modeling stage was run as described by Weitzner, Jeliazkov, Lyskov et al. (Nature Protocols 12, 401–416, 2017). 

The set of 49 antibody-Fv domains was originally compiled by Marze et al. (Protein Eng. Des. Sel. 29(10), 409-418, 2016). The dataset was first used to evaluate CDR-H3 loop rmsds in Weitzner and Gray (J. Immunology 198(1):505-515, 2017).

The dataset can be extracted on a linux interface using: 

                 tar -xvzf decoys_rosettaantibody_20200323.tar.gz 

When extracted the output comprises RosettaAntibody generated models for 49 antibodies in the format:

/ model-.relaxed_.pdb.gz

The Rosetta "ref2015" scores (ref2015_score) and rmsd of the H3 loop (h3_rmsd) for every model (model) is provided in model_scores_and_rmsds.txt . RMSDs are calculated with respect to the corresponding crystal structure (pdb id) over all heavy atoms in the h3 loop (93–102 in Chothia numbering) after superposition of the framework residues. ID 1 comes from the homology model source (lower is better). ID 2 indicates the loop model.

The decoy PDB files contain additional metrics following the ATOM records such as VH–VL relative orientation metrics (from Marze et al.) and per-residue Rosetta scores. Caveat: the “RMS” values reported within the decoy files were calculated against the input homology model and not the crystal structures (whereas the model_rmsds.txt file contains the H3 rmsds w.r.t. crystal).