Re: [AMBER] Parametrization: consistency with gaff force field

From: David A Case <>
Date: Mon, 14 May 2018 10:25:43 -0400

On Tue, May 08, 2018, Alechania Misturini wrote:

> a question has arisen about the
> consistency of generated parameters with the force field employed (GAFF, in
> my case).

You raise difficult questions, for which there are often no clear
answers. One rule seems clear to me: if you want charges that are
consistent with GAFF (and with GAFF2 and Amber force fields like ff14SB
and our nucleic acid force fields,) you should use RESP charges with
the HF/6-31G* basis. [AM1/BCC should be viewed as a fast but imperfect
approximation to RESP/HF/6-31G*, to be used in cases where convienence
and speed are more important than strict "consistency".]

For energies, basis sets and methods have tended to improve as
computers and codes have become faster. Here I personally see less
reason to worry about "consistency" and more reason to be concerned with

> *Due the silicon atom present in the fragment, I need at least a triple
> zeta basis set to better describe it, so, I did a benchmark with some
> functionals and basis set (MP2, B3LYP and PBE0, with 6-31G*, 6-311G*,
> def2-TZVP and cc-pVTZ). Fortunately I could find a similar x ray structure
> to help me validate these results, and pick the best level of theory for my
> case.*

The above sounds OK to me: if one needs bigger basis sets for heavier
atoms, that's what is needed.

> *The same doubt rise about the structures used to parametrize with paramfit
> or mdgx. Different QM level of theory than the employed in GAFF force field
> can result in inconsistent parameters, looking the whole picture? (cause
> just bonded parameters will be parametrized, LJ values will come from GAFF
> and parm99 data, for Si)*

Whether one should transfer Lennard-Jones parameters from GAFF/GAFF2 to
new systems, or try to reparameterize them, it one of the difficult
questions. Consistency would argue in favor of the former. If you are
going to try to be consistent, I'd recommend using GAFF2 (not GAFF 1.x)
as your "standard".

Others may well have different opinions. Remember that GAFF(2) is
indended to provide initial estimates of force fields that are often
"good enough" for purposes like docking. If you have the time and
knowledge to tweak the GAFF estimates for individual systems, you may
well get better results.

...good luck....dac

AMBER mailing list
Received on Mon May 14 2018 - 07:30:02 PDT
Custom Search