Biology Reference
In-Depth Information
energy calculations, since the solvent effect is described according to a
PBSA or GBSA implicit solvent model. More recently, some studies also
performed the MD simulations using implicit solvent models.
52,53
The
normal modes are usually calculated on a smaller number of frames, due
to the central processing unit (CPU) requirement of such calculations.
Short 0.5-1-ns trajectories are generally performed, yielding to con-
verged energy terms. Longer simulations have been tested, up to 10 ns
in length,
50
but they were not found to provide better results, most prob-
ably because long simulations emphasize force field errors and limita-
tions. Indeed, it has been found that MM-PBSA yields better results with
MD simulations restrained around the X-ray structure, compared to
unrestrained simulations.
54
Two possibilities arise concerning the number of MD simulations to
perform. In principle, one should make three trajectories, one for the
complex and each of the isolated partners, and calculate the energy
terms using the adequate simulation. However, a popular alternative
consists of performing only one MD simulation for the complex. In this
variant, the terms relative to one isolated partner are calculated after
removing the atoms of the other partner in the frames extracted from
the MD simulation of the complex. As a consequence, the reorganiza-
tion energy of the molecules upon association is neglected (
0);
however, this variant is less CPU-demanding and leads to increased con-
vergence due to cancellation of errors, reduction of noise arising from
flexible remote regions relative to the binding site, and conformational
restraints imposed by the complex geometry. Thus, this one-simulation
variant is attractive when
∆
H
intra
=
H
intra
may be reasonably neglected.
Comparisons between one- and three-trajectory results can be found in
the literature.
36,45,54
MM-PB(GB)SA is expected to estimate absolute
∆
G
bind
without
adjustable parameters. Although several studies were able to reproduce
experimental
∆
G
bind
for protein-protein association with an error lower
than 2 kcal mol
−1
,
45,50
these results are open to discussion. Indeed, the
approach contains several “hidden” parameters, such as the force field
used, the choice of the PB or GB variant and that of the nonpolar solva-
tion model, the use of one or three trajectories, and the different terms that
can be included or neglected. As a consequence, it is sometimes possible
∆
