Re: shake and potential functions
Kim Sharp writes:
1) Missing terms in potential functions. Dave Pearlman's comments
regarding the energy term implicit in the SHAKE constraint applies
more generally in any constraint we add to potential functions.
Another example is polarization: If we parameterize our force field
to give a certain energy for some process where the polarization is in
reality changing, but which our potential functions keep fixed, then
we are implicitly putting energy in to maintain our constraint, but
not including it in our final analysis of enegy changes...
But polarization is not the only other place where this is a consideration.
For example, atomic charge couples to bond length, bond angles, and dihedral
angles (esp. in cases such as an amide bond where rotation couples to the
hybridization).
2) If all SHAKE is giving you is a factor of 2, my personal feeling
is that it is not worth the uncertainty involved if one is doing
energy calculations. If it gave a factor of 10 maybe.
In effect you are argueing that the uncertainties (errors?) introduced by SHAKE
are the leading error in the potential function, but it is not at all clear to
me that this is so.
3) Time and effort is probably better spent on treating electrostatic
interactions in dynamics, since this is probably the most problematic
area in dynamics. Maybe this is another area that will generate
discussion.
Again, agreed. Compared to the assumptions made in treating electrostatics,
SHAKE seems pretty benign.
Some issues:
a) Cutoffs
In effect aren't we doing very high salt strength simulations with a
cutoff/Debye Huckle screening of long range effects?
b) empirical dielectric functions: since there is no explicit h-bond
potential, h-bonds are subsumed into the LJ and electrostatic terms.
What happens when any other dielectric constant than e=1 is used
(h-bonds are carefully parameterized in DISCOVER using a constant
dielectric of one). Although DISCOVER allows one to use other constant
and distant dependent dielectrics, do these have any meaning?
Why do we uniformly neglect electronic polarization in calculating the
dielectric constant for condensed phase simulations? In vacuo unity is
appropriate, but in most condensed organic or bio-organic systems the index of
refraction would suggest that electronic polarization (on a time scale fast
compared to molecular vibration) is contributing a dielectric effect that is
screening Coulombic interactions by a factor of two. If you assume that the
numbers assigned to atomic charges bear any relationship to physical charge,
then it would seem that this screening should be considered.
Of course, if you actually do alter the dielectric constant, you will have to
completely reparametrize all of the hydrogen bonding and many of the van der
Waals interactions. Given the degree of coupling between these terms, one has
to be very cautious in attaching physical meaning to them. After all, an
empirical solvent model is in some sense just mathematical engine tuned to
exhibit a set of behaviors we desire.
c) neutralizing charged groups. this is standard procedure in NOE
refinement. So long as we have enough NOE's/degree of freedom
the system is determined enough that this seems to be o.k. Similarly,
in nucleic acid simulations, it is common to scale the phosphate
charges down to -0.2 to -0.3. It seems to "work", but is that
justification enough. These approaches are pretty black box. what are
people's experiences?
I seem to recall that the long range rigidity of DNA can be predicted reasonably
well based on polyelectrolyte theory alone. This would suggest that if you muck
with the electrostatics and still see appropriate long range rigidity, then you
have introduced a compensating error somewhere else.
kim sharp, columbia u, dept of molecular biophysics
David States
National Center for Biotechnology Information/National Library of Medicine
----- End Included Message -----