forward. NMA methods can access many collective motions and has
been used in approaches which look for correlated displacements
between pockets [138] or modes which can transmit perturbations
made in predicted pockets [139]. NMA approaches have been
shown to capture long timescale dynamics of proteins but not the
less populated states when compared to the millisecond trajectories
of DESRES [140]. However, the lightly populated states were
missed by NMA; motions with a low degree of collectively can
also be missed.
Evolutionary constraint offers another rapid method for under-
standing correlated dynamics which may be able to access longer
timescale phenomena. Studies to date have applied statistical cou-
pling analysis (SCA) [141] which derives a statistical energy from
the probability of a particular amino acid appearing at a given site. A
conservation-weighted covariance matrix built with SCA leads to a
“correlation entropy” which describes the allosteric coupling
between groups of highly coupled residues. This coevolution
derived analysis of allostery has been applied to serine proteases,
hemoglobin, GPCRs, PDZ domains, PAS domains, SH2 and SH3,
identifying functionally important residue sectors and allosteric
couplings between them [142–144]. Subsequently, SCA correla-
tion analysis combined with a small molecule pocket search and
high-throughput docking has discovered an active compound for
cathepsin K which binds at a novel allosteric binding site [145].
If the required sequence data is available, SCA offers detail of
allosteric connections at any timescale with minimal computational
costs. Although, coevolution approaches cannot account for post-
translational modification or ligand binding which can be a prob-
lem for many drug targets.
The concepts of local instability relating to functionally impor-
tant regions and minimal frustration have been described since the
early discussions of protein folding [146, 147]. The group of Peter
Wolynes evaluates residue frustration mutations or alternate con-
figurations through coarse-grained comparison of energies after
mutations and threading of the new sequence to available crystal
structures [148, 149]. If the mutated energies are high compared
to the wild-type then this position is considered minimally frus-
trated and vice versa. This approach has been packaged in an
algorithm called the frustratometer, which can highlight regions
which are likely to be functionally important [150, 151]. While
based on approximate methods, prediction of dynamics by the
frustratometer has been favorably compared to NMR data for the
catabolite activator protein [152]. Another approach called CON-
TACT models alternate residue configurations into crystallography
electron density, scoring potential van der Waals clashes and deriv-
ing a score similar to the frustratometer [153]. Pathways of residues
which frustrate each other have been show to coincide with NMR
data and extensive mutational studies [154].
Computational Study of Protein Conformational Transitions 353