I missed your point about the two REsP runs giving different
results--indeed, you should get the same result if there are two
conformations, one given 100% of the weight, or if there is just the one
conformation. I don't know what could be the problem here. However, I
still do not recommend simply averaging the results from multiple REsP
runs--fit all conformations simultaneously.
On the subject of what weight to assign, I would recommend that you give
all conformations an equal weight in lieu of strong evidence that one is
dominant. Reason being, it is hard to know what the occupancy of any one
conformation will be from its vacuum phase characteristics--the solvation
energy can drastically change what is favorable or not. So, if all of the
conformations experience similar levels of bond, angle, and dihedral
strain, I'd have a hard time telling which is more or less favorable when
the thing is placed in water. Also, the consequences of the weighting may
not be as significant as you think--the nature of the least squares problem
tends to be buffered against problems of over- or under-representation.
If you're truly interested in the theory behind the problem, I'd suggest
that a much more influential factor on the outcome is the way you select
the fitting points. If you take Gaussian, it makes a surface (or perhaps
several) over the molecule in order to choose points at which to calculate
the electrostatics due to the wavefunction and then serve as inputs to the
charge fitting problem. This can go wrong if you have too high a
concentration of points around one atom (like a small polar hydrogen) or
too few around another (like a carboxylate). It's the same problem of
biased representation, only within a particular conformation rather than
from lack of an alternative conformation. If your program is approximating
the surface by making a sphere of points and then stretching it to fit the
size of each atom, this will put more points around hydrogens and less
around heavy atoms. Again, REsP is naturally buffered against a good deal
of this--if monopoles are in fact a good model for what's going on, then
you can get the electrostatics correct for 1000 points around an atom just
as well as you can for only 100, it doesn't really matter how much you
emphasize one end of the molecule or the other because there is actually a
good answer and that will be found. The problem is when monopoles don't
really do a good job--then REsP will start making compromises, by its very
construction. When that starts happening, it begins to matter how much
data came from this hydroxyl versus that phenyl ring.
Dave
On Wed, Nov 23, 2016 at 12:29 PM, Marion, Antoine <antoine.marion.tum.de>
wrote:
> Hi Dave,
>
> Thanks for the answer and the interesting discussion.
> The choice of charge model (polarizable or not) is indeed a crucial point.
> However, at the moment I am more focused on the theory behind RESP.
> I will make such choices later.
>
> I might clarify my questions:
>
> - How the weight of each conformation is accounted for during RESP fitting?
> - Why do the following two settings of RESP give different results?
> a) single conformation (confA)
> b) two conformations (confA and confB) with the weight of confA and confB
> set to 1.0 and 0.0, respectively.
> - What is your opinion regarding the use of a simple average of confA and
> confB charges instead of multiconformation RESP?
>
> Thanks again for your help.
>
> Best regards,
> Antoine
>
> ________________________________________
> From: David Cerutti [dscerutti.gmail.com]
> Sent: Wednesday, November 23, 2016 5:27 PM
> To: AMBER Mailing List
> Subject: Re: [AMBER] About multiconformations RESP fitting
>
> I don't think that a simple averaging is quite what you'd get if you were
> to fit both conformations simultaneously, equally weighted. But, it's also
> not surprising that you get very different results by fitting to both as
> opposed to just the one. REsP is more stable than its boisterous cousin
> ESP, but there is still a high dependence on the input conformations. When
> I fitted charges for amino acids, I took twenty conformations of each of
> them (500 conformations across twenty amino acids plus other protonated
> forms--it was a big matrix problem). Much of this depends on how flexible
> and polarizable your system is. If there is a high degree of electronic
> polarization, then there could be a problem making a fixed charge model at
> all, but in general if the molecule is flexible you're going to expose
> different surfaces with different conformations, which will implicitly
> weight the importance of particular charges differently, so even if there
> is no polarization a flexible molecule will have conformational dependence
> in REsP fitting which can only be removed by sampling.
>
> Dave
>
>
> On Wed, Nov 23, 2016 at 10:39 AM, Marion, Antoine <antoine.marion.tum.de>
> wrote:
>
> > Dear Amber community,
> >
> > I am trying to reach a better understanding of the RESP procedure.
> > In particular, I am interested in multiconformations RESP fitting.
> >
> > I have followed the explanations of the tutorial (
> > http://ambermd.org/tutorials/advanced/tutorial1/section1.htm)
> > and adapted it to my problem (modified alanine dipeptide).
> > The programs are running well and after a few corrections of inputs, I
> get
> > results that seem very reasonable.
> >
> > My concern is about the theory behind.
> > I understood (by trying) that the way in which the different
> conformations
> > are taken into account is not a simple average.
> > Then, I tried to vary the weight of each conformation (say confA and
> > confB).
> > To my surprise, when setting the weight of confB to zero, I get a (very)
> > different result than when I do a RESP fitting on confA only.
> > I might have miss one equation, but I could not find where the weighting
> > factors are used.
> >
> > I would greatly appreciate some directions to understand this matter.
> >
> > Another question:
> > Would a simple average of the atomic charges (weighted or not) be also
> > acceptable to account for both conformations?
> >
> > Thanks in advance.
> >
> > Best regards,
> > Antoine MARION
> > _______________________________________________
> > AMBER mailing list
> > AMBER.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber
> >
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Nov 23 2016 - 11:00:02 PST
