Re: [AMBER] optimizing GAFF2 torsional parameters

From: Karl Kirschner <>
Date: Thu, 12 Sep 2019 09:06:56 +0200

Hello Christos,

  To answer your second question, there is no general preference for
directing the FF error towards a false minima versus a barrier height. It
depends on what you are willing to deal (i.e. error) with when looking at
the resulting MD data. On the one hand, you would likely be populating a
false minmum/minima (and taking population away from true minima), or you
will decrease/increase the sampling rate between conformations (increasing
the rate might be beneficial). The upside is that you have a good
understanding of what is to be expected in the resulting data since you now
know the sources of possible errors and can better gauge the data's

  The first question takes much longer to answer. A couple of thoughts to
consider is a) do you trust your QM data, b) have you checked to make sure
that the MM and QM structure are the same at each conformation (i.e. you
don't suddenly create an intramolecular hydrogen bond at -30 degrees that
is not present in the QM conformation), c) have you used all three
multiplicities at the same time to fit the QM data (or just doing each one
at a time), and d) have you considered the other torsion terms that are
coupled to the the main one you are looking at (eg. hc-x-x-hc) - warning:
this can lead to opening a "can-of-worms." I think your approach of keeping
the phase angles 0 or 180 is good, especially if you want to keep the force
field transferable. However, if you want to really focus on this one
torsion and don't care about transferibility, then you assign the phase
angle to other values. If this torsion is used elsewhere in the model with
different chemical environments (ie. other coupled dihedral terms) then you
might want to create a new atom type and isolate it from the rest of the


On Wed, Sep 11, 2019 at 8:23 PM Christos Deligkaris <>

> dear all,
> I am trying to improve the GAFF2 dihedral parameters for a torsion on a
> small molecule. My small molecule has 5 torsions, only one is problematic,
> I have been able to improve the dihedral parameters for the other 4
> torsions by adjusting the force constants of the GAFF2 dihedral terms.
> In the figure shown here (PDF file):
> red is QM, green is GAFF2, blue is one of my attempts to improve the
> results. The somewhat improved energy maximum at 0 degrees comes with an
> artificial minimum at around -30 degrees. This minimum seems to also be
> present at the default GAFF2 torsional profile but it is more shallow.
> I can increase the weights of the points around the artificial minimum to
> make them more consistent with QM data, but of course that comes at the
> expense of less agreement with QM data at around -180- -120 degrees. I have
> also tried optimizing the phase angles, without restricting them to 0 or
> 180, but that introduces an additional artificial energy minimum at around
> -150 degrees. I have also tried all possible multiplicities, 1,2,3 without
> any luck. I am using Mathematica to do the optimization.
> What strategies have people used in the past to improve agreement of
> dihedral parameters with QM data? Are there reasons to prefer no artificial
> energy minima versus better agreement on energy barriers?
> Best wishes,
> Christos Deligkaris, PhD
> Assistant Professor of Physics, University of Southern Indiana
> SC2220, 8600 University Blvd, Evansville IN, 47712
> Office Phone: (812) 228-5056
> .DeligkarisGroup
> _______________________________________________
> AMBER mailing list
AMBER mailing list
Received on Thu Sep 12 2019 - 00:30:02 PDT
Custom Search