These atomistic MM models of proteins are, with modern GPU
hardware, sufficiently fast to access biologically relevant timescales.
They are able to reasonably accurately reproduce many experimen-
tal observables including NMR measurements. In addition they are
transferable between almost any protein system/model. For these
reasons the atomistic MM resolution is by far the most popular and
important resolution for protein simulations.
At the other end of the spectrum of protein model resolution,
coarse-grained (CG) models offer very large time steps and time-
scales at the expense of reduced accuracy and detail. Protein con-
formational change is complex and therefore it is very difficult to
produce a transferable CG model which is predictive for conforma-
tional change (seeNote 6). While examples of transferable CG
models for conformational change exist none are widely used or
applicable to large complex systems [15–17].3.2 Sampling Large
Motions
with Collective
Motions
An interesting protein sampling strategy is to make large collective
motions based on Eigen-function calculations or prior knowledge
about structure. In general approaches which make large collective
motions must use an implicit solvation model owing to the poten-
tial clashes with the solvent as a result of the motions introduced.
Also these approaches do not generally offer properly weighted
dynamic ensembles, but are generally designed to be very efficient
for large motions, like hinge bending, within a low dimensional
subspace or energy basin.
Normal mode analysis (NMA) is a well-known approach for
recovering the motions of a protein using an approximation which
assumes the potential energy is harmonic. A normal mode is a
“pattern of motion in which all parts of the system move sinusoi-
dally with the same frequency, with a fixed phase relation and is
independent from other normal modes.” Hence, NMA has been
mainly used for understanding the slowest or lowest frequency
motion of a protein. This generally means motions whose start
and end points are within the same low energy basin or
low-dimensional subspace. NMA with a full atomistic molecular
mechanics model is considered computationally expensive as it
requires a large amount of computer memory (seeNote 7).
Reduced NMA models, namely elastic network models
(ENMs), have become popular owing to their simplicity, computa-
tional lightness, and remarkable accuracy compared to experiment
and more expensive computational approaches. ENMs use simpli-
fied potentials which often utilize harmonics with a single force
constant between all alpha-carbons [18, 19]. Improved ENMs are
common and often focus on generalized plans for adjusting the
connections and strengths of the harmonics [20–22](seeNote 8).
Protein energy landscape exploration (PELE) is a minimized
Monte-Carlo basin-hopping method which combines advanced
ligand moves with an ENM and side-chain sampling. PELE isComputational Study of Protein Conformational Transitions 343