Fig. 1.
Overview of CS-Rosetta. BrokerMain marks the start of any CS-Rosetta protocol. Abrelax is triggered from the AbrelaxMover via BrokerMain and contains two representative steps, (i) a setup step that is used to initialize topology broker, fragment sampler, and loop closure protocols, and (ii) an apply step, which calls the ab initio fragment sampler (violet) to obtain low-resolution models that undergo loop closure and high-resolution structure refinement using relax. A fragment sampler (and assembly) runs through four stages (violet): (i) stage I: large fragment insertion moves are performed and any steric clashes are prevented using score weights provided by score0, (iii) stage II: fragment insertion moves are combined with score weights (score1) to account for residue-pair interactions, (iii) stage III: fragment insertion moves are carried out to promote local (stage IIIa) and nonlocal (stage IIIb) β-strand pairings with the help of weights defined in score2 and score5, and (iv) stage IV: small fragment insertion moves are performed primarily to close loops. In addition, score weight for radius of gyration is increased (in score3). RASREC is an iterative method that performs structure calculation across six resampling stages (tan). In stage I, β-strand pairs are exhaustively enumerated. During stage II, β-strand pairs are enumerated again and β-topologies are resampled. Stage III is used to carry out only β-topology resampling. During stage IV, folds are intensified (also known as protofold resampling) and fragments are resampled from low-resolution models. Loops are closed and the resulting structures are converted from centroid to full atom in stage V. In the last stage (stage VI), the final structural models are refined in the Rosetta force field. Archive in Rosetta is used for the purpose of communication between parallelizable unit (MPI-based job distributor) and RASREC algorithm. Based on the status of the job distributor, archive can read decoys, setup batches of runs or check if the calculation is complete. AutoNOE iteratively assigns NOEs and performs structure calculation using RASREC (green). It interacts with RASREC across all the stages of resampling. AutoNOE updates NOE assignment parameters at three distinct phases (green, right). Phase I is designed to be more permissive meaning; more ambiguous NOE restraints are assigned, and no restraints are eliminated. Phase II is less permissive, where unambiguous NOE restraints are favored and elimination of violated restraints is allowed in the highly converged regions. Phase III is the most rigid, where unambiguous NOE restraints are generated and violated restraints are eliminated. The assigned NOE list generated at the start of each resampling phase is filtered by eliminating all the diagonal cross-peaks. The NOEs are further scored and refined using chemical shift compatibility, covalent compliance, agreement with decoys (obtained from RASREC), symmetric assignments, and network anchoring. The intensities of cross-peaks of assigned NOEs are calibrated to obtain upper distance bounds for restraints followed by the removal of spurious peaks.