Essentials of Computational Chemistry

420 11 IMPLICIT MODELS FOR CONDENSED PHASES
while GB models are very robust for the prediction of solvation free energies, they are
less successful in the generation of potentials of mean force (Rankin, Sulea, and Purisima
2003). Lack of high-quality data makes it difficult to evaluate this possibility at present,
although ongoing comparisons between different theoretical models are helping to further
illuminate the issue (see, for example, Jayaram, Liu, and Beveridge 1998 and Gohlke and
Case 2004).
Note that one feature of Figure 11.12 is a solvent-separated minimum for the X–Y pair.
Insofar as solvent-separated minima involve intervening solvent molecules that typically
differ significantly in their behavior from normal bulk solvent as a consequence of being
isolated between the two solutes, such situations are unlikely to be handled accurately by
continuum models in general.
It is sometimes the case that the structure of the first shell (or shells) of solvent is a
property of primary interest for a given modeling study. It is perhaps stating the obvious to
note that in such an instance, continuum models cannot be used, since by construction they
ignore the molecular nature of the solvent and assume a homogeneous surrounding medium.
Of course, if one is interested only in the free-energy well associated with full complexation,
many technical aspects of the calculation are simplified. The tremendous speed of
continuum solvent models has made them attractive tools in evaluating solvation effects on
docking, especially insofar as they permit more extensive sampling of varying target-receptor
geometries to be carried out in an efficient manner (see, for example, Gouda et al. 2003;
Taylor, Jewsbury, and Essex 2003; and Zoete, Michielin, and Karplus 2003).
11.4.5 Molecular Dynamics with Implicit Solvent
A large fraction of the expense of a typical MD simulation involving a solute in solution
(discussed in much more detail in the next chapter) is associated with the hundreds or
thousands of solvent molecules that are explicitly represented in the full simulation cell.
However, when the fine details of the solvation process are not of primary interest, it can be
about an order of magnitude more efficient to propagate a trajectory for the solute within the
context of continuum solvation. The methodology that has been most extensively explored for
this process to date has tended to involve GB solvation models developed for biomolecular
force fields (although PB models have also seen substantial use). To maximize speed, Born
radii are computed either from a PD algorithm or are set to constant values determined from
initial PB calculations (Onufriev, Case, and Bashford 2002). For truly enormous systems,
additional algorithms allowing certain portions of the solute to be held frozen while others
are dynamical have been described (Banavali, Im, and Roux 2002; Guvench et al. 2002).
A particular advantage of MD with implicit solvation is that solvent friction is not an issue
with respect to the solute being able to explore phase space. That is, no solvent molecules
need to be pushed out of the way in order for otherwise energetically accessible largescale
motions to take place. As long as the energy landscape for the solute is as accurately
predicted with the continuum solvent as with an explicit solvent, this feature leads to much
more rapid achievement of converged sampling (Okur et al. 2003) especially when LES is
used (Cheng, Hornak, and Simmerling 2004). This behavior has been successfully exploited