Computational Approaches in Peptide and Protein Design: An Overview - Peptide and Protein Design for Biopharmaceutical Applications - page 28

PDB for each short fragment (usually, three to nine residues long) of the
target sequence. (In this regard, the Rosetta algorithm is not an ab initio
physics-based approach, but is based on heuristics.) The backbone con-
formations of each fragment are then selected by Monte Carlo sampling,
starting with the most represented, and are inserted into the entire
sequence, which is initially put into the extended 3D structure. In parallel,
fragments of the regular structures, such as a-helices and b-strands, which
are predicted by statistical methods, are also inserted into the structure. An
ensemble of various structures is generated by including various backbone
conformations for each fragment segment, and by various perturbations
made in the dihedral angles of the peptide backbone of the entire chain and
in the space of the Cartesian coordinates. The generated rough structures
are ranged by calculation of specifically developed scoring function, and
the most probable are refined by an all-atom approximation with insertion
of various rotamers of the side chains and by performing fine Monte Carlo
sampling in the space of dihedral angles, followed by energy minimization
with MM (Monte Carlo plus minimization procedure [78]). The loop
fragments, which may not be represented in the PDB, have to be modelled
separately and usually with less accuracy [79].
Over the last decade, the Rosetta approach has repeatedly shown good
performance in the community-wide critical assessment of structure pre-
dictions (CASP), where researchers are asked to predict 3D structures of
proteins with known sequences with experimentally determined 3D
structures withheld until after the predictions are made [80-84]. The
Rosetta algorithm has allowed numerous applications, including the
design of novel proteins with a desired 3D structure [85]; design of
protein sequences with new folds [86]; design of protein sequences
aimed at avoiding specific 3D folds ('negative design') [87]; as well as
applications for protein-protein docking [88-90]. Though it would be
premature to evaluate the overall success of the Rosetta approach, it can
be concluded that Rosetta provides a powerful tool for the computational
design of peptides and proteins.
Sampling of side-chain rotamers Many procedures for sampling the
conformational space of peptides and proteins, as well as for sampling
the orientations of the ligands, are developed primarily considering back-
bone conformers. At the same time, all-atom models should include low-
energy combinations of the side-chain rotamers, which must pack with one
another and with the backbone without steric clashes. While the resulting
3D shape of the entire structure is not significantly affected by selecting a
particular side-chain rotamer, the number of combinations of side-chain