AMBER Archive (2006)

Subject: Re: AMBER: partial charge glycam

From: kkirschn_at_hamilton.edu
Date: Thu Jan 26 2006 - 08:49:35 CST

        Apologies for the delayed response. I agree with the Francois'
responses. The two strategies (R.E.D. and Glycam) are not too
different, as pointed out in his first paragraph below. What we feel
that is unique to Glycam and our charge determination is that we
incorporate the idea of an ensemble charge set. The method for
fitting charges to fairly rigid molecules has been work out well by
many researcher, as in Francois' R.E.D.. The question that one
always tackles is what conformation to use that best represent the
molecule in solution at room temperature (here I am assuming that the
force field will be used for an solution-phase MD simulation at
298K). I would say that for rigid molecules the gas-phase optimized
geometry will approximate the solution-phase geometry pretty well,
but this is an assumption and the accuracy of your model ultimately
lie in how good your assumption are. With very flexible molecules
(ie. carbohydrates) the molecule can adopt several conformations that
may or may not be well represented by one or even several gas-phase
optimized conformations. Thus, we adopted the approach of performing
an MD simulation in solution to obtain an ensemble of solution-phase
conformations, which we then proceed to fit charge as outlined
before. Obviously, there are still assumptions being made in our
procedure (ie. the MD simulation is adequately sampling the
conformational space of our molecule). One neat thing we get out of
our ensemble method is that we can assign to each partial charge a
standard deviation. This standard deviation allows us to see which
atoms have the least and the greatest partial charge fluctuation as a
function of the molecule's conformation.

In response to your specific question Kevin:
          In my experience, small changes in the partial atomic charge of
the glycosidic oxygen does not alter the MD simulation results too
much. However, I feel that in any MD simulation the results one
obtains must be considered in regard to the approximations that were
made to generate them (ie. what force field was used, what charge set
was used, what are the non-bonded cut-offs values, what type of
solvation was used, whether scaling factors were used for the non-
bonded and electrostatic, etc.). So, alter the partial atomic charge
enough and you could alter the MD results.

These have been good discussions :)

Karl

On Jan 25, 2006, at 4:23 AM, FyD wrote:
> Quoting Kevin Murphy <k.murphy_at_hud.ac.uk> (full email below):
>
>> This strategy is different to the one you've out lined below, I
>> was wondering
>> if you've tried it? is the charge on the glycosidic oxygen and
>> surrounding
>> atoms different? and does it matter? i.e. does it affect the
>> dynamics of an
>> oligosacchride chain during an MD simulation.
>
> The GLYCAM developers can correct me, but I am not sure the two
> strategies
> reported are so different than that, or in any cases they present
> many common
> points: CHELPG algo., HF/6-31G* theory level, single RESP stage with a
> hyperbolic restraint qwt=.01, C-H charges = zero. This has to be
> opposed for
> instance by what is used in AMBER FF development (Connolloy surface
> algo.,
> HF/6-31G* theory level, two RESP stages with different hyperbolic
> restraints
> (qwt=.0005/.001). Now, the two strategies might differ in the
> minimum selection
> scheme and/or in the fitting procedure.
>
> For instance, if we look @ the RNA project F-55: The project
> summary is
> available in R.E.DD.B. @
> http://www.u-picardie.fr/labo/lbpd/REDDB/up/F-55/info.html:
> - Concerning the minimum selection, we simply selected a single
> conformation
> (C3'endo since this is RNA) avoiding hydrogen bonds (HB) between
> the C'2 & C3'
> hydroxyls (avoiding also the TS structures generated by the use of QM
> constraints) and controlling mol. orientations using a rigid-body
> re-orientation algo. HB affect charge values and the polarization
> obtained
> using HF/6-31G* and, HB in sugars are problematic since a hydroxyl
> (O or OH)
> can either be an acceptor or a donor of HB...
> - Concerning the fitting procedure, we used charge constraints for
> C-H hydrogens
> to set their charge values to zero. Other strategies are always
> possible.
> Personnally, what I learnt is that fitting is neither averaging nor
> suming and
> looking at the RRMS is a/one rigorous attempt for checking the
> quality of the
> fit (a minimum number of charge constraints being always used since
> constraints
> slightly increase the RRMS); The charge values reflecting only the
> MEP...
> - Concerning the connections betwen sugars, once again by looking
> at the RRMS
> and comparing the values obtained with and without intra or inter-
> molecular
> charge constraints (possibly needed for the connections) is _one_
> rigorous way
> for answering 'YES' or 'NO' the constraints break or not your
> system. I
> personnally prefer this type of approach compared for instance to
> some 'manual
> manipulations'...
> - Other differences ? May-be, I do not know ;-)
>
> These emails demonstrates that different strategies are used in the
> charge
> derivation even for a same FF. My opinion about answering to the
> question
> "which one is the best" is not easy since charge values are
> affected by many
> parameters and consequently many different options can be chosen.
> It is why we
> developed the RESP ESP charge DDatabase where highly reproducible
> RESP and ESP
> charges are available. R.E.DD.B. stores charge values, but also the
> _computational conditions_ [mol. conformation (Cart. coordinates) and
> orientation, basis set, MEP algo, fitting procedure, human errors,
> etc...] used
> in the charge derivation procedure. Thus every modeler can use the
> charge values
> available but also re-calculate, compare, criticize the values
> available; The
> goal being to improve the procedures reported and try to answer
> which one(s)
> is(are) the more suitable for protein, nucleic acid and/or sugars.
>
> Exemple of comparisons are already available @
> http://www.u-picardie.fr/labo/lbpd/REDDB/up/F-51/info.html
> http://www.u-picardie.fr/labo/lbpd/REDDB/up/F-52/info.html
> http://www.u-picardie.fr/labo/lbpd/REDDB/up/F-53/info.html
> http://www.u-picardie.fr/labo/lbpd/REDDB/up/F-54/info.html
> http://www.u-picardie.fr/labo/lbpd/REDDB/up/F-55/info.html
> http://www.u-picardie.fr/labo/lbpd/REDDB/up/F-56/info.html for RNA
>
> http://www.u-picardie.fr/labo/lbpd/REDDB/up/F-45/info.html
> up to http://www.u-picardie.fr/labo/lbpd/REDDB/up/F-50/info.html
> for DNA
>
> http://www.u-picardie.fr/labo/lbpd/REDDB/up/F-1/info.html
> up to http://www.u-picardie.fr/labo/lbpd/REDDB/up/F-22/info.html
> for amino-acid
> fragments.
>
> http://www.u-picardie.fr/labo/lbpd/REDDB/up/W-46/info.html
> up to http://www.u-picardie.fr/labo/lbpd/REDDB/up/W-50/info.html
> for solvent
> molecules.
>
> We hope the community will help to develop R.E.DD.B. and criticize
> what is
> already available...
>
> Regards, Francois

On Jan 24, 2006, at 7:18 AM, Kevin Murphy wrote:
> Karl,
> Thanks for your helpful reply. I don't know if you read what
> Francois posted so I've cut out the most relevant paragraph
>
> Now concerning, the connections of your new monosaccharide unit
> with others
> belonging to GLYCAM: you simply have to use some linking fragments
> where you
> will fix the total charge value for this fragment to a target value
> (usually
> zero, but you can imagine any values...), i.e. a charge value which
> makes your
> new unit compatible with other units (In this case just check the
> RRMS to see
> if your constraint does not increase it too much). Concerning the
> linking
> fragment itself I would first try the groupmnent OH or OMe... This
> type of
> strategy is the more or less same for DNA, peptide or sugar; except
> that some
> force fields use different 'tricks' and/or MEP algorithms...
>
> This strategy is different to the one you've out lined below, I was
> wondering if you've tried it? is the charge on the glycosidic
> oxygen and surrounding atoms different? and does it matter? i.e.
> does it affect the dynamics of an oligosacchride chain during an MD
> simulation.
>
> Thanks
> Kevin
____________________________________
Karl N. Kirschner, Ph.D.
Visiting Assistant Professor of Chemistry
Hamilton College, Clinton NY 13323
____________________________________

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu