AMBER Archive (2003)

Subject: RE: Valdiation of Latest Amber parameters

From: Yong Duan (yduan_at_udel.edu)
Date: Thu Jul 10 2003 - 00:25:33 CDT

Dear Jiri,

Obviously, we have different opinion on this.

The so-called gas-phase charges in AMBER was based on HF/6-31G* which is
really somewhere between gas-phase and condensed-phase.

CHARMM relies on interaction energies which indeed takes into account
part of the condensed-phase properties (so, is not gas-phase). Since the
interaction energies are calculated by dE=E12-E1-E2 and the later two
terms were calculated in gas-phase, dE actually has the potential to
exagerate condensed-phase properties because of compensation.

OPLS is tuned using condensed-phase enthalpy data. So, it is not
gas-phase.

I hope we do not have to completely dismiss the gas-phase energies. I do
not think anybody can anyway. But, as far as ff is concerned, we have to
be considerably better than 0.5kcal/mole. We really have to talk within
this limit. Actually, 0.5 kcal/mole in key areas is horrible for ff
parameters.

It is true some of the energy differences are similar. For instance,
when you look at the energy profiles calculated using gas-phase and
condensed-phase, they do look similar. The question remains just how
similar. Different people have different standard. But, are they within
0.5 kcal/mole limit? The answer is "maybe". Is it sufficient to meet the
"considerably better than 0.5kcal/mole" criterion? The answer is very
likely not.

Well, we can apply the same criteria to those existing ff. We should not
forget those fine differences between different ff in those key areas
are actually within this 0.5kcal/mole mark. In other words, as far as
comparison to the QM data is concerned, all ff are equal because
anything below 0.5 kcal/mole accuracy is not attainable from that set of
QM data. So, we are bettern served to dismiss that particular
comparison.

Within the 10 conformers, one of the useful conformers were the fully
extended beta- structure. They also did a calculation on an
alpha-helical conformer. These two are really important. But the
extended and alpha-helical are also very different in terms of solvent
exposure, particularly the main-chain which is much more "burried" in
alpha- conformation than in beta- . The whole idea of protein ff is to
model the main-chain correctly because errors in main-chain are
cumulative. Until sombody can demonstrate to me that the gas-phase and
condensed-phase energies are really very similar, I would have to guess
they are quite different.

yong

-----Original Message-----
From: Jiri Sponer [mailto:sponer_at_ncbr.chemi.muni.cz]
Sent: Wednesday, July 09, 2003 1:04 PM
To: Yong Duan
Cc: amber_at_heimdal.compchem.ucsf.edu
Subject: Re: Valdiation of Latest Amber parameters

Dear Yong:

   going back to the discussion of the "validation",
   I would like to make just few points (although
   the discussion has moderated already).

   I do not thing it is a good idea to dismiss accurate
   gas phase data. Although the aqueous solution affects
   the electronic structure of segments of biopolymers,
   the effects are not so dramatic. Major electronic features
   of the species are not changed. AMBER anyway relies
   on "gas phase" HF charges. CHARMM relies on gas phase
   water ... something charges, this still is a gas phase system.

   What is substantially affected by solvent is the order
   of various minima, primarily due to complex specific hydration and
   solvent screening effetcs. These may even invert the gas phase
   (free) energy trends. This however concerns primarily
   the long-range electrostatic
   contributions while the intrinsic electronic
   properties are an order of magnitude less sensitive
   to the solvent effects.
   (In order to really change the electronic structure
   you need to bind for example a transition metal.)
     Thus the main problem to use Beachy et al for
     a comparison is not the gas phase nature of the
     calculations but the selection of the structures.

   QM calculations of microhydrated clusters could well
   be used for comparison with force fields.

   It is obvious why the ff are parametrized for
   condensed phase biopolymer simulations. However, in an optimal
   case the ff should also be reproduce the gas phase
   data and certainly should not be drastically different.
   Of course, due to simplicity of the ff this often is
   not possible and one needs to calibrate the force field
   for biologically relevant geometries. Then, however, we
   rely on the compensation of errors and that could sometimes fail.
   In any case, substantial problems to reproduce
   gas phase data always are a warning sign and should be
   investigated.

   Best wishes,

   Jiri

-------------------------------------------------------
 Jiri Sponer
Institute of Biophysics
Academy of Sciences of the Czech Republic
Kralovopolska 135
CZ-61265 Brno
Czech Republic
e-mail: sponer_at_ncbr.chemi.muni.cz
fax: 420 5412 12179
phone: 420 5415 17133
http://www.ibp.cz/labs/LSDNA/
Wellcome Trust Senior Research Fellow for Biomedical Research
--------------------------------------------------------