In GAFF, the bond stretching and bond angle bending force constants come
from protein/nucleic acid force fields (FF94, FF99) whenever possible. For
those not having any counterparts in FF94/FF99, we used a set of empirical
formulae (please refer to the GAFF paper) to calculate the force
constants. The parameters of those formulae were derived to reproduce the
existing force constants in FF94/FF99.
However, in GAFF2, we adopted a different philosophy to derive those
parameters (26 in total). We optimized those parameters to minimize the
differences between the calculated MM frequencies and the ab initio ones
(the training set has 572 small molecules and 22407 frequencies in total).
We also tested the best parameter set with a large test set (711 molecules,
29000 frequencies). The ab initio frequencies were obtained at B3LYP/6-31G*
level. We could reduce the RMS errors from 120 cm-1 for GAFF to 60 cm-1 for
both the training and test sets.
We are now doing a set of evaluation for GAFF2, which will be a part of the
GAFF2 paper. I agree with you that GAFF is more consistent with the current
protein/nucleic acid force fields. And the difference between GAFF and
GAFF2 for c3-c3 is very big.
Hopefully this can help you to choose a proper version of GAFF force field
in your study.
Junmei
On Tue, Jan 24, 2017 at 7:24 PM, Marek Maly <marek.maly.ujep.cz> wrote:
> Thank you !
>
> BTW that difference in c3-c3 force constant GAFF vs GAFF2 is really big
> 300.9 vs 232.52 I would say. To be frank I am a bit confused with such a
> big change moreover in such basic/common atomtype bonding term.
> This indicates that methodology and/or set of molecular fragments which
> were used for fitting in GAFF2 case had to be quite different from
> methodology or mol. fragments used in GAFF case.
>
> I am also a bit confused with the fact that the same term in actually
> recommended protein force field ff14SB has still force constant close to
> old GAFF value:
>
> CT-CT 310.0 1.526 JCC,7,(1986),230; AA, SUGARS
>
> So old GAFF seems to me to be a more compatible with ff14SB from this
> point of view than with the new GAFF2, which is a bit strange or not ?
>
> BTW the same trend holds if we compare here GAFF/GAFF2 with LIPID14 ff.
>
> cA-cA 303.1 1.5350 Lipid11 v1.0 (GAFF c3-c3)
> cD-cD 303.1 1.5350 Lipid14 v2.0 (GAFF c3-c3)
>
> again we are much closer here to GAFF values.
>
> The same trend holds for vdW terms:
>
> GAFF
> c3 1.9080 0.1094 OPLS
> GAFF2
> c3 1.9069 0.1078
> FF14SB (parm10)
> CT 1.9080 0.1094 Spellmeyer
> LIPID14
> cA 1.9080 0.1094 OPLS
> cD 1.9080 0.1094 OPLS
>
>
> So it seems that if one is going to simulate some heterogenous system
> (e.g. protein + drug) it should be more suitable to use still old GAFF
> (not the new GAFF2) as GAFF is clearly more compatible
> with bio-force-fields than GAFF2 or am I wrong here ?
>
>
> Best wishes,
>
> Marek
>
>
>
> Dne Tue, 24 Jan 2017 19:44:34 +0100 David Case <david.case.rutgers.edu>
> napsal/-a:
>
> > On Tue, Jan 24, 2017, Marek Maly wrote:
> >>
> >> I have technical question regarding QM methods used in GAFF2
> >> development.
> >
> > cc-ing to Junmei Wang, who doesn't always follow the mailing list. Note
> > that
> > you may need to wait for the paper on GAFF2 to be published...
> >
> > ...dac
> >
> >>
> >> In Amber16 manual, page 286, there is written:
> >>
> >> "We have performed B3LYP/6-31G* optimization for 15 thousands
> >> marketed or
> >> experimental ..."
> >>
> >>
> >> So B3LYP/6-31G* is here QM method used to obtain optimized molecular
> >> structures (low energy
> >> conformations) from which equilibrium geometrical terms (eq. bond
> >> lengths,
> >> eq. angles) were obtained
> >> and so in some cases old GAFF values were updated.
> >>
> >>
> >> But in GAFF2, comparing to GAFF, changed clearly also some force
> >> constants.
> >> As an example I can mention c3-c3 bond:
> >>
> >>
> >> In GAFF
> >> c3-c3 300.9 1.5375 SOURCE1_SOURCE5 88072 0.0058
> >>
> >> In GAFF2
> >> c3-c3 232.52 1.538 SOURCE1_SOURCE5 88072 0.0058
> >>
> >>
> >> My question is, which QM methodology was used to derive (perhaps using
> >> paramfit ?) new GAFF2 bond, angle or dihedral force constants ?
> >>
> >> I assume, that since here we need just to calculate for each
> >> conformation
> >> only it's
> >> energy (no optimization) and since we also need to obtain reliable
> >> force
> >> constants (hence we need accurate QM energies) some more accurate method
> >> was probably used like MP2/6-31G** , MP3/6-31G** etc. to derive QM
> >> energies. Am I right ?
> >>
> >> So thank you in advance for QM methodology specification used in GAFF2
> >> for
> >> force constants
> >> derivation.
> >>
> >> Best wishes,
> >>
> >> Marek
> >>
> >
> > _______________________________________________
> > AMBER mailing list
> > AMBER.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber
> >
> > __________ Informace od ESET NOD32 Antivirus, verze virove databaze
> > 14822 (20170124) __________
> >
> > Tuto zpravu proveril ESET NOD32 Antivirus.
> >
> > http://www.eset.cz
> >
> >
> >
>
>
> --
> Tato zpráva byla vytvořena převratným poštovním klientem Opery:
> http://www.opera.com/mail/
>
> _______________________________________________
> 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 Thu Jan 26 2017 - 10:30:02 PST
