Re: [AMBER] Missing (ATTN, need revision) and poor parameters (high penalty score) for small molecule parameters and hard time with paramfit

From: David Cerutti <>
Date: Sun, 23 Aug 2020 18:06:46 -0400

Allan, if you have the conformations and QM energies, put the energies into
a file (one per line) and make a trajectory out of the conformations. You
can then jump into the mdgx tutorial at a fairly late stage and get your
parameters fitted:

If you have good sampling and the bond shows harmonic characteristics, you
should be able to get a decent answer out of this.


On Sat, Aug 22, 2020 at 5:00 PM Állan Ferrari <> wrote:

> Dear Amber user community,
> I am interested in simulating a protein - small molecule inhibitor
> interaction.
> I optimized the small molecule geometry with Gaussian and derived RESP
> charges.
> Also, with parmchk2 I derived frcmod files containing small molecule
> parameters to prepare the system for simulation using both parm10 and gaff.
> However, there are still some missing parameters or parameters with high
> penalty score totalizing 12 parameters among one bond, couple angles and
> dihedrals.
> I have tried for days to use paramfit following this tutorial (
> ) but with no success.
> In my first attempt, I have tried to generate a large number of conformers
> (> 1000) with random values for the angles and dihedrals I am interested in
> parametrizing. Then, I runned Gaussian to evaluate its energy (constrained,
> single point energy calculation). When I tried to derive K (the
> difference between Amber energy and quantum energy, in Hartree) using
> paramfit the fit was horrible (R2 was negative!).
> In a second attempt, I tried to generate only the r_eq and bond force
> constant for one bond. I generaeed 50 conformers with random distances
> between the two atoms of interest (0.5 to 3.5 A), runned gaussian single
> energy point calculation for each and when it came to derive K again an
> horrible fit (r2 < 0.05)! I tried to continue the tutorial regardless of
> the poor fit but it didn't work, as expected. It complained I had
> insufficient bond information in sample structures. Definitely not the
> case.
> Finally, I was hoping that I could derive manually the parameters, once I
> have the energies as a function of the distance between the atoms.
> I fit k*(r-r0) function to the derived (E, r) data points. The k, however,
> is equal to 4.5 while the k for bonds already found from parm10 and gaff
> files are in the 200-600 value range.
> Questions:
> 1) Is this last attempt a valid approach, that is, vary the degree of
> freedom I am interested in parametrize and then manually fit to obtain the
> missing parameters?
> 2) Is there a multiplying factor for this parameter (10E-2?) that would
> justify the difference I am observing (4.5 to 200-600)?
> If (1) is not an option, I would really appreciate a better
> tutorial/documentation for paramfit or other source for small molecule
> amber parametrization...
> Kind regards,
> Állan
> _______________________________________________
> AMBER mailing list
AMBER mailing list
Received on Sun Aug 23 2020 - 15:30:03 PDT
Custom Search