AMBER Archive (2009)

Subject: Re: [AMBER] Nonstandard molecules simulated with parm99EP ?

From: case (
Date: Tue Jun 30 2009 - 11:19:00 CDT

On Tue, Jun 30, 2009, FyD wrote:

> >"How it is possible to force the antechamber to use some different
> >forcefield from GAFF ( like for example parm99EP
> >for parametrisation of the relevant residui (generating of
> >PREPIN/FRCMOD files) ?"

This is not possible in general. You can use the "-at amber" flag to
get parm99 atom types assigned, but that won't work well for arbitrary
molecules, since many common ligand environments are not present in
proteins and nucleic acids.

> Personnally, I always use parm99 FF atom types for organic
> molecules/protein/nucleic acid and add FF atom types manually; pick up
> only _classical_ force field parameters from GAFF (bond & angle force
> constants, aromatic dihedrals) and adapt them to parm99 way. Get
> equilibrium values from high ab initio geometry optimization or X-ray
> structures (not from GAFF). Key dihedrals are always refitted (not
> obtained from GAFF).
> What you said above summarize exactly the problem "so some inacuracy
> regarding to specific molecules is "penalty" for it's
> transferability/universality". When you use GAFF automatically all is
> fine if your molecule has been well parametrized but you do not know
> if it was indeed well parametrized...

I want to second Francois' comments here: The gaff parameters can be
considered as an (often good) starting point for molecular mechanics
parameterization, but the procedure outlined above should generally give
better results -- especially the advice about refitting key dihedrals.
Be especially critical of gaff recommendations where your ligand has extensive
conjugation or bonds with partial double-bond character. On the other hand,
gaff should be pretty good for saturated compounds (like, say, fatty acids or
many lipids), or where you have "simple" aromatic groups where it is easy for
antechamber to assess the extent of electron delocalization.

Of course, for projects like virtual screening of large databases, one must
generally accept the compromises inherent in gaff, since it is not feasible to
perform hand-optimization on thousands of ligands.


AMBER mailing list