As a quality control measure, the resulting simulation systems were individually inspected.
Problems such as minimization or preparation failures (e.g. failure of the total energy to
decrease during minimization, typically a result of omission of a disulfide bond, etc.), and
large structural changes in the protein were identified and corrected. To ensure that all systems
were correct, each simulation’s starting configuration was examined by hand.
After simulation, we aim to identify any structure that undergoes a considerable rearrangement or
lacks a well-defined conformation. For native state simulations, no such
rearrangements or lack of structure should occur. Simultaneously, we must keep
in mind that small conformational changes from the experimentally determined
starting structure are to be expected and small fluctuations may be necessary
for structural stability. To quantify the stability of (native state)
simulations, we calculated root-mean square deviation (RMSD) values based on
only the Cα positions of residues that make up the ‘core’ of the protein: all
residues that have a solvent-accessible surface of less than < 40 Å2.
This prevents large movements of tails or surface loops in our simulations from
significantly affecting RMSD values. We further defined a ‘median structure’ as
the structure from the simulation that has the lowest RMSD value to all other
structures. A median root-mean square fluctuation (mRMSF) can now be defined as
the root of the average RMSD from every structure to the median structure.
Additionally, the median root-mean square deviation (mRMSD) was defined as the
RMSD between the starting structure and the median structure. For all simulated
targets, these two values are correlated (R=0.74), with the mRMSD being
approximately twice as large as the mRMSF. Clear outliers were simulations with
mRMSD values above 4 Å or mRMSF values above 2 Å (Figure 1). Inspection of these
simulations revealed several with large structural rearrangements and exposure
of hydrophobic cores, whereas this was not observed in simulations with smaller
mRMSD and mRMSF values.
Where NMR data are available from the PDB or the BioMagResBank (BMRB)  in
formats that could be directly employed for analyses or readily converted by the
DOCR tools , trajectories were analyzed in the context of those data. For the
purposes of comparison to nuclear Overhauser effects (NOE), an experimental NOE
was considered satisfied if the r-6 weighted distance of the closest pairs of
protons specified by the constraint was less than 5.0 Å or the experimental
upper-bound. Order parameters (S2 ), primarily in the form of main-chain amide,
were calculated from MD trajectories as described in  and an R to the
experimental values was used as a measure of agreement. Similarly, chemical
shifts predicted from MD trajectories were compared to the experimental values
by the measure of R. Residual dipolar couplings (RDC) from MD trajectories were
calculated by the algorithm presented in  and compared with experimental
data. All analyses were performed in ilmm with the exception of chemical shifts,
which were calculated with SHIFTS .