Re: CCL:Switching functions and double cutoffs
On Fri, 13 Feb 1998, Nick Glover wrote:
On Fri, 13 Feb 1998, Nick Glover wrote:
> My recent post regarding salt-bridges and h-bonds has generated some
> replies concerned with my use (or lack thereof) of switching functions.
> Having just completed a search of the CCL archives, I note that the
> arguments for and against the use of switching functions in simulations
> of enzyme systems has received some debate.
I should note that solvent accessible salt-bridges are also the subject of
debate; some people argue, based on NMR data, that they are crystal artefacts
and not that significant in solution. The work on the leucine zipper GCN4-p1,
a 33 AA peptide which forms a stable dimer in solution, is one instance.
> Having already received some useful replies (thanks!) I would be very
> interested to hear the community's opinion regarding the use of double
> or single cutoffs and the use, or not, of switching functions. Opinions
> regarding the values to be assigned to these functions would also be
> appreciated. The system is peptide bound to enzyme in a solvated
> sphere. CVFF/Discover_95 is the computational source.
I try to avoid switching functions for electrostatic potentials; there's a
well known problem related to the forces (derivative of the potential) at the
inflection point of the sigmoidal curve. The effect can be minimized by using
longer switching intervals, e.g. more than 4 A or so. Better still (if you
must use cutoffs) are force based cutoff methods, where the forces (instead of
the potential) are brought to zero gracefully.
Ewald summation methods seem to be both gaining favor and efficiency; the
current dilemma is that while there are strong theoretical reasons for using
Ewald methods, most molecular mechanics parameters were developed with
explicit cutoffs instead. The other major factor is that Ewald methods
require the use of periodic boundary conditions, which implies a rather
specific concentration of the solute (e.g. a protein). Ewald methods do
increase the complexity of the setup for nonbond calculations; there are quite
a few additonal parameters, some of which are best "tuned" for the
specific
system being simulated. Periodic systems do allow a rigorous treament of
pressure, however, so that the statistical mechanics are well defined, and one
can reasonably calculate thermodynamic properties.
The same cannot be said for the infamous spherical bag of waters; a simulation
of that system represents a solute in an isolated droplet of a vacuum aerosol,
and not a protein in solution. The pressure is not exactly determined, and
there are electrostatic artefacts induced by the dielectric boundary
(water/vacuum); charges behave as if there were an image of the charge on the
the vacuum side of the boundary. While I don't have the page number, I've
been told the reference for the "image charge" phenomenon is a paper
by
Onsager in Vol. 1, No. 1 of either J. Chem. Phys. or J. Phys. Chem.; he also
published on reaction field methods to deal with dielectric discontinuities
over 60 years ago.
Basically, unless a reaction field or similar technique was employed, the
sphere of water may be as problematic as the use of switching functions,
depending on the molecular properties to be estimated. It's especially true
if parts of the protein come within 1/2 the nonbond cutoff of the sphere
surface. On the other hand, if the catalytic site is somewhat buried, and you
aren't concerned with either bulk water or protein surface sidechain dynamics,
the sphere of water is not such a terrible approximation, as long as one is
aware of the limitations of the technique.
--
Rick Venable =====\ |=| "Eschew Obfuscation"
FDA/CBER Biophysics Lab |____/ |=|
Bethesda, MD U.S.A. | \ / |=| ( Not an official statement or
rvenable %! at !% deimos.cber.nih.gov | \ / |=| position of the FDA;
for that,
http://nmr1.cber.nih.gov/ \/ |=| see http://www.fda.gov )
http://www.erols.com/rvenable