Two-dimensional infrared spectroscopy (2DIR) is another
interesting technique able to probe protein secondary structure
elements without protein labeling. While the structural resolution
of 2DIR is not particularly high compared to NMR it is able to
cover a large array of timescales as the underlying time resolution is
in the ps range. Labeling with azides or NO can offer residue level
resolution. Thus far combinations with simulation have been at the
level of cross-validation to help with assignment of spectra and
understanding the timescales of ion-channel and [108, 109]. The
use of temperature-jump 2DIR offers more detailed timescale data
which can help validate information-rich simulations like MSMs
[110]. The emergence of reliable methods for calculating 2DIR
spectra from simulation data suggests that a more integrated com-
bination with MD is a possibility [111].
The dream of integrative structural biology is to bring all of
these different sources of structural data together in one model,
which can be used to guide drug design. Platforms for the integra-
tion of any form of structural data into MD simulations have started
to emerge and the favored approaches for this are Bayesian infer-
ence and the maximum entropy principle. The optimal combina-
tion and arrangement of these approaches is not completely clear
yet and reviews and perspectives are starting to appear
[112–114]. MELD uses a parameter describing the fraction of an
experimental data set which is reliable combined with Bayesian
inference picking out the most appropriate sets of restraints for
each frame in a simulation. MELD is a well-developed platform
with published examples of difficult problems like structure
prediction but its application is limited to small systems [115].
Metainference is another advanced integrated simulation method
implemented in Plumed, the well-known metadynamics plugin.
Metainference is able to produce accurate simulations even with
reduced models, such as implicit solvation, when combined with
NMR chemical shift data [116].
A common use case when searching for hidden conformations
is the non-direct use of experimental structural data. Any type of
insight or hypothesis can be used to manually build variables for
biasing potentials or adaptive sampling; but even with rich struc-
tural data, producing the conformational transitions often requires
variables which are not obvious. The capture of a hidden confor-
mation with Fab domains and X-ray crystal can suggest the exis-
tence of a whole new region of conformational space and be a great
aid to MD-based conformational exploration. An interesting exam-
ple is the antibody captured linear conformation of IgE; a large
metadynamics simulation showed that this conformation is likely
not stable in solution without the stabilizing Fab domains. Other
conformations predicted to be metastable by the simulation were
captured by the therapeutic antibody omalizumab [117, 118].350 Benjamin P. Cossins et al.