Hi Francois, Matthew, and the rest of the listserve,
I have one small correction to point out. Much of Glycam06's energetic parametrization on neutral molecules was done using B3LYP/6-31++G(2d,2p) single-point calculations, performed on HF/6-31G(d) optimized geometries - not B3LYP/6-31++G**. This use of this theory and larger basis set comes from N.L. Allinger's study of carbohydrate energies (Lii et al J. Comput. Chem, 1999, 20, 1593), and is pointed out in the 2007 Glycam06 paper. The use of the basis set is important, since hybrid DFT results are very dependent upon the basis used.
Cheers,
Karl
----- Original Message -----
From: "Matthew Tessier" <matthew.tessier.gmail.com>
To: "AMBER Mailing List" <amber.ambermd.org>
Sent: Thursday, September 22, 2011 8:25:12 PM
Subject: Re: [AMBER] about Glycam_06g.dat and thiol-glycosidic linkage
Francois,
>> I think QM theory levels are different when one wishes to target a
geometry, an energy, and a MEP.
True, but for the development of GLYCAM06 only HF/6-31G* and HF/6-31++G**
were employed for optimizations. Using MP2 for optimization would be very
expensive for little gain. For parameter fitting, single point energies
were calculated with B3LYP/6-31G* & B3LYP/6-31++G** depending on the
molecule. The DFT method was found to give reasonable agreement for little
cost when applied to sugars. Using higher levels of theory may get better
gas phase results but when it comes to agreeing with solution rotamer
populations, DFT is more cost effective and gives reasonable agreement. If
someone knows of a QM/DFT study that shows a better way then we'll certainly
explore that.
>> Thus, is it really useful to add diffusion functions in geometry
optimization in these conditions?
It actually does make a difference in the cases of anionic systems. We've
compared sulfate charges developed with HF/6-31G* and HF/6-31++G** and found
that free energies of binding agree better with the 6-31++G** basis set.
-Matt
-----Original Message-----
From: FyD [mailto:fyd.q4md-forcefieldtools.org]
Sent: Thursday, September 22, 2011 2:46 AM
To: AMBER Mailing List
Subject: Re: [AMBER] about Glycam_06g.dat and thiol-glycosidic linkage
Matthew, Yun,
I think QM theory levels are different when one wishes to target a
geometry, an energy, and a MEP.
- For MEPs and for a non-polarizable/fixed charge model, HF/6-31G* is
used to induced an implicit polarization
see
http://ambermd.org/doc6/html/AMBER-sh-19.4.html#sh-19.4
- To get representative energy values and representative dE values
between conformations high basis sets (triple zeta with multiple
polarization functions and diffusion functions; and MP2 or even MP4
(if one can afford it)) are recommended.
- For _bio-organic_ molecules, to get an optimized geometry HF/6-31G**
is quite basic, Yes, and using diffusion functions should lead to a
better accuracy, Yes. However, there is _obviously_ a reason for this
choice... Geometry optimization is carried out in gas phase, while MD
simulations are done in condensed phase. Thus, the conditions in
geometry optimization will never fit to these of the molecules in
their experimental environment (i.e. partially or totally solvated;
docked in a protein, etc...). Thus, a basic theory level is used in
geometry optimization with several empirical rules (i) a big molecule
is split into elementary building blocks to fully define to different
conformations to be involved in charge derivation (ii) canonical
intra-molecular hydrogen bonds are generally avoided (hydrogen bonds
induce over-polarization). To avoid intra-molecular hydrogen bonds and
overpolarization effect partial geometry optimization can be used (the
'ModRedundant' mode when using Gaussian as reported by Matthew below).
Thus, is it really useful to add diffusion functions in geometry
optimization in these conditions? I am not sure... The main idea here
is to get an empirical/canonical conformation representative of what
one chooses and computes charges for these conformations.
This is the approach followed by Cieplak et al. in 1995 (HF/6-31G*)
and by Duan et al. (HF/6-31G**) in geometry optimization almost 10
years later.
If you look at optimized geometry of monosaccharide units in R.E.DD.B.
intra-molecular hydrogen bonds are generally avoided:
See for instance (two conformations might be present in these PDB files):
http://q4md-forcefieldtools.org/REDDB/projects/F-85/
http://q4md-forcefieldtools.org/REDDB/projects/F-85/mol1.pdb
http://q4md-forcefieldtools.org/REDDB/projects/F-85/mol8.pdb
http://q4md-forcefieldtools.org/REDDB/projects/F-72/
http://q4md-forcefieldtools.org/REDDB/projects/F-85/mol1.pdb
http://q4md-forcefieldtools.org/REDDB/projects/F-85/mol2.pdb
I think you can find information about hydrogen bonds, MEP
computation, geometry optimization, energy values and dE in:
http://www3.interscience.wiley.com/cgi-bin/abstract/109583237/ABSTRACT
http://onlinelibrary.wiley.com/doi/10.1002/jcc.10349/abstract
http://pubs.rsc.org/en/Content/ArticleLanding/2010/CP/c0cp00111b
regards, Francois
> As Francois noted, we use the MD ensemble to weight the charges towards
> solution conformations. My only concern is your QM levels are a bit off.
> For neutral sugars, we optimize and develop the ESP with a 6-31G* basis
set
> whereas for anionic systems 6-31++G** is used for optimization and ESP
> calculations. The RESP weight is the same (0.01) either way and the rest
of
> the options you mentioned are correct. I would use a crystal geometry for
> your pre-ensemble averaged charges (if available) or an optimized QM
> structure.
_______________________________________________
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 Sep 23 2011 - 03:30:03 PDT
