Hi Igor,
As David said, whether or not to use relaxed structures is a question
without a clear answer. For the purposes of the tutorial, we use unrelaxed
structures and then weight the unreasonable ones less importantly in the
fit. This helps users get calculations quickly in order to become familiar
with Paramfit.
However, in the lipid paper we did indeed do QM optimization on the hexane
and octane. It's not clear which approach is better-- if you're concerned,
you may want to try both approaches and see if the resulting parameters
differ significantly.
Paramfit can fit bonds and angles just fine. Most of the paper and its
usage focuses on dihedrals since they are more complicated and often
several are fit simultaneously. With a bond or angle it is more simple to
do a QM scan and then fit to that, which can usually be accomplished
without any sophisticated minimization algorithms. Paramfit will work just
fine in these cases as well.
Hope this helps,
Robin
On Wed, Jul 27, 2016 at 10:02 AM, David Cerutti <dscerutti.gmail.com> wrote:
> The use of MM versus QM relaxed structures is a quasi-religious issue.
> Carlos and I have had a number of discussions about this and one can make a
> case either way. One thing you cannot get away with is taking QM relaxed
> structures and using those same coordinates to compute MM energies as the
> basis for new parameters. I've done fine with QM energies of MM relaxed
> structures and then following a 1:1 mapping of coordinates and energies,
> which is what the tutorial is doing.
>
> I'm working on something in mdgx that may eventually resolve the issue.
> It'd be best to use QM relaxed structures all the way, but the reason this
> won't work, as I mentioned above, is that the MM just can't dance to that
> tune (the fine footwork of bonds and angles isn't what we'd need). There's
> a way to fix this and not require fundamental changes to PMEMD.cuda, I
> think, but it's a ways off.
>
> Dave
> On Jul 27, 2016 12:36 PM, "Igor Marques" <igor.dragon88.gmail.com> wrote:
>
> > Dear AMBER users,
> >
> >
> > I have some questions concerning the Paramfit tutorial and paper.
> >
> > On the tutorial, after the structures generation and validity check, we
> get
> > to the quantum calculations. From the available header, we will be
> > performing SP calculations.
> >
> > *%RWF=job.rwf
> > %Chk=job.chk
> > %NoSave
> > %Mem=2GB
> > **#SP** b3lyp/6-31G* geom=connectivity*
> >
> > On the other hand, in the paper the authors say that the:
> >
> > *The CH₂–CH₂–CH₂–CH₂ alkane torsion potential was fit with Paramfit to
> the
> > energy of structures from torsion scans performed on hexane and octane
> > molecules evaluated using the hybrid method for interaction energies
> > (HM-IE).[28] Initial structures were generated from a 15° torsion scan of
> > hexane and octane and then optimized at the MP2/cc-pVDZ level before
> > performing the single-point energy calculation (Fig. 9).*
> >
> > So... what's the correct way to go about this? From the paper, it looks
> > likes a (Gaussian?) relaxed scan every 15°, with optimization of each
> > conformation. But this seems contradictory to the tutorial, where, for
> > instance, the MM energy of every valid structure is matched to its QM
> > energy determined with a single point. Is someone able to clarify this?
> >
> >
> > On the other hand, both the tutorial and the paper focus on the
> > parametrization of dihedral angles. Does anyone have any experience
> > concerning the parametrization of bonds or angles? How reliable is
> Paramfit
> > for those?
> >
> >
> > Many thanks,
> >
> > Best wishes,
> >
> > Igor
> >
> > Igor Marques, PhD Student
> > MSc in Pharmaceutical Biomedicine
> > BSc in Biomedical Sciences
> >
> > Molecular Modeling Group – University of
> > Aveirohttp://molecular-modeling.dq.ua.pt/
> >
> > Mobile phone: +351 918 567 294
> > Landline: +351 234 247 254
> > Through PBX: +351 234 370 200 (ext. 22160)
> >
> > Universidade de Aveiro
> > Campus Universitario de Santiago
> > Agra do Crasto – Edificio 30
> > 3810-193 Aveiro - Portugal
> > _______________________________________________
> > 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
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Jul 29 2016 - 14:00:02 PDT
