- With the AMBER suite, Umbrella Sampling calculation gen-
erates a dump file, in which all the values of the restrained
variableζalong the simulation of the unbinding process are
stored. A preprocess step is needed to split the large file into
multiple files, one for each window. These files can be viewed
and analyzed with, for example, xmgrace (http://plasma-gate.
weizmann.ac.il/Grace/), by converting each file into a histo-
gram plot. This analysis is essential to verify the shape of the
distribution (each window must have a Gaussian-type distribu-
tion, according to Boltzmann law) and an overlap of each
window with their neighborhoods all along the unbinding
process. In our case, one can see on Fig.5 that the overlap
between all windows during the simulation is reasonable. All
distributions of each window have a Gaussian-like shape mean-
ing that the system has reached an equilibrium point and they
overlap with their neighborhoods. This protocol is important
for the relevance of the potential of mean force (PMF), created
in the next step.
From these files, the WHAM (Weighted Histogram Analy-
sis Method) algorithm is used to obtain the potential of mean
force (PMF) associated to the evolution of the bias. Informa-
tion, description of the methodology, and software sources are
available at (http://membrane.urmc.rochester.edu/content/
wham). Briefly, the WHAM protocol allows for the conversion
of sampling distribution ofζinto free energy, by comparison of
the distribution center to the expected valueζthfor a single
window. This conversion is made for each window in order to
build the PMF profile. If the system is equilibrated enough in
each window, the PMF profile could be associated to the free
4
3 3
2 2
1 1
0 0
05
ς(Å) ς(Å)
10 20 30 40 5.5 6 6.5 7
Fig. 5Left: Sampling distribution, for all windows, of the restrained variableζalong the whole simulation of
the unbinding process. One distribution is associated to one window. Right: Focus on the first four distribution
of the simulation. The overlap between windows is quite satisfying, and the distribution shape is the one
expected in all cases
420 Sonia Ziada et al.