> I still feel that my best chances are through divcon. I gratefully accept your
> suggestion to go to divcon directly, chapter 7. Checking against QuantumBio web
> pages, it seems that the version of divcon in Amber9 is the latest, 4.5.
I would like to re-emphasize Professor Case's point regarding the
sensibility of charge-fitting for large molecules... (below).
> > You are really taking the charge derivation part of antchamber where it was
> > not designed to go. Even it things don't crash, you would need to examine
> > whether or not the charge model one gets for very large molecules is a good
> > one.
> >
> > The "am1-bcc" charge philosophy is to perform a geometry minimization on the
> > input molecule before assessing the charges. This makes sense for small
> > molecules, where geometry optimization is fast. For large, floppy molecules,
> > it probably makes much less sense. Again, this is a scientific question that
> > will require patience and research. If you wish to use divcon for this
It is not clear to me whether or not the charges will make *any* sense at
all for a large molecule with AM1-BCC or RESP charges. This is a research
question and is unknown. I would be highly skeptical regardless of how
much computer power you throw at this with DIVCON, that is unless you
explicitly consider conformational dependence (and exploring
conformational dependence in a large molecule with QM quickly becomes
prohibitive). Even then, I would have to be convinced that the resulting
charges are better than either random, or ad hoc assigment (i.e. carbon
slightly negative i.e. -0.2, hydrogen slightly positive +0.1,
electronegative atoms more negative), or simpler fragment based
approaches.
An easier way to think about this is to consider the RESP charge fitting
that AM1-BCC is intended to approximate. Essentially, you perform QM
calculations to estimate an electrostatic potential and then find point
charges on the atoms that best fit that potential. If the molecule is
large (and/or floppy), the fit is necessarily noisy and will not be as
well determined. [Bayly/Cieplak/Kollman saw this (sort of) with the
original ESP fitting on small molecules that evolved into RESP to prevent
buried charges from becoming large during the fit; the charges became
large since their value did not have considerable influence on the ESP,
i.e. they were not well determined; see RESP papers by Bayly/Cieplak.]
Moreover, if a molecule is large, the resulting charges will depend
heavily on the conformation chosen. This is why we tend to consider small
molecules for the charge fitting in vacuum, ideally looking at explicitly
at the conformational dependence (or choosing particular subsets of
accessible conformations, such as extended backbones in peptides). It is
not simply a matter of simplicity, but that the model makes sense with
small molecules and may start to break down with larger ones.
What I would recommend doing is breaking up a larger molecule into
"pieces" and fitting each independently first (perhaps even with a RESP
approach). Then you could spend more time sampling in molecular mechanics
space the conformations (which is presumably what you want to do) rather
then spending weeks generating charges for a very large molecule that may
be unrealistic.
If you do not want to do this, then I would suggest calculating the
charges on multiple different representative conformations of the large
molecule and average them (probably I would use R.E.D.). How many
conformations depends on how flexible the molecule is, and averaging would
be best in principle using some kind of Boltzmann weighting (and this is
all a research question). Then I would compare the results to random
charges and ad hoc / rule based charges...
-- tec3 at utah
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Sun Aug 19 2007 - 06:07:15 PDT
