Relax

The Relax API runs the Rosetta FastRelax protocol. Quoting from the RosettaCommons documentation: “This finds low-energy backbone and side-chain conformations near a starting conformations by performing many rounds of packing and minimizing, with the repulsive weight in the scoring function gradually increased from a very low value to the normal value from one round to the next.” This is useful for taking a structure not generated by Rosetta and “relaxing” it into the Rosetta scorefunction preparatory to further modeling, or for generating a backbone ensemble for a further modeling experiment.

Inputs

  • Input PDB file
    • CLI argument: –pdb-file input.pdb
    • Python submit() argument: pdb-file=”input.pdb”
    • Do not include nonprotein residues.
    • Do not include multimodel (NMR-sourced) PDBs.
  • Optional – Number of Repeats. The number of different relaxed conformations the loop modeling API will return.
    • CLI argument: –repeats 5
    • Python submit() argument: repeats=5

Outputs

Output file descriptions

  • Results can be downloaded if and only if a job has succeeded – DONE state, see the list command below.
  • models — A directory of PDB files with relaxed structures. The models will be in the form input_####.pdb, where #### is an index (0001, increasing).
    • Note that Rosetta will renumber the residues monotonically starting from 1, contiguous across chains.
  • score.sc — A text file containing the Rosetta scores for each generated model.

Output data interpretation

The model numbering is irrelevant to quality.

Model quality can be assessed via scores. Broad documentation on interpreting scores can be found here. The score.sc file is a space-delimited data table, padded for easier reading. It can be parsed with pandas dataframes, excel, or your tool of choice. In the score.sc file, pay particular attention to the total_score column. The total_score is Rosetta’s overall grade for a model. The total_score should have a negative value for good models, although this will depend enormously on the quality of the input model.

Usually you should sort your models by total_score, look at the lowest-scoring 5 or so, pick the one your biophysical intuition says is best, and proceed with it for whatever your further experiment is.

Updated: