Comments regarding Antechamber.
1. I did not notice any symmetrization of charges on atoms that are equivalent like (H-atoms of methyl groups, benzene molecule etc.)
2. Geometry optimization is not working properly in standard Amber10 installation.
I do not know if I am the only one experiencing these problems or not, and I will expand a bit more on problem 2 below:
I was somewhat dissatisfied with the apparent inability of antechamber to get converged geometries for small molecules, like catechol for example. I am using the Amber10 standard installation on RH Linux box. Since antechamber is running mopac under the hood, and some time ago I used that program stand alone, I gave it a try to see if I can improve upon that PRECISE keyword. To my great surprise the mopac that is compiled with Amber (using Intel compilers) was not able to optimize a reasonable guess at all.
The exact command I used was:
antechamber -i catechol.pdb -fi pdb -o oh2bnz.mol2 -fo mol2 -c bcc -s 2 -at gaff -ek "AM1 ANALYT GRAD LET GNORM=1.D-8 SCFCRT=1.D-6 XYZ"
This generates an mopac.in file that looks like this:
AM1 ANALYT GRAD LET GNORM=1.D-8 SCFCRT=1.D-6 XYZ CHARGE=0
created by wmopcrt() for mopac
C -0.0170 1 1.3780 1 0.0100 1
C 1.1690 1 2.0890 1 0.0020 1
C 2.3790 1 1.4200 1 -0.0130 1
C 2.4050 1 0.0350 1 -0.0210 1
O 3.5940 1 -0.6250 1 -0.0350 1
C 1.2110 1 -0.6800 1 -0.0130 1
O 1.2320 1 -2.0400 1 -0.0200 1
C 0.0020 1 -0.0040 1 0.0020 1
H -0.9600 1 1.9040 1 0.0260 1
H 1.1500 1 3.1690 1 0.0080 1
H 3.3040 1 1.9780 1 -0.0190 1
H 3.9460 1 -0.8200 1 0.8440 1
H 1.2460 1 -2.4380 1 0.8610 1
H -0.9260 1 -0.5570 1 0.0080 1
The output looks like a big mess to me:
SCF CRITERION = 0.1000E-05
CYCLE: 1 TIME: 0.06 TIME LEFT: 3599.9 GRAD.:999999.999 HEAT:-54.21049
CYCLE: 2 TIME: 0.05 TIME LEFT: 3599.9 GRAD.:999999.999 HEAT:-54.21157
CYCLE: 3 TIME: 0.02 TIME LEFT: 3599.8 GRAD.: 198.479 HEAT:-21.66359
GRADIENTS OF OLD GEOMETRY, GNORM= 198.479378
-15.499537 57.227984 122.266395 1.431796 -3.530504 23.028728
-35.279682 -50.180782 -38.433267 -25.461497 -3.680921 6.443132
15.413123 -45.601920 -37.698058 -13.430117 50.503021 -4.669441
34.211819 41.404541 11.990522 -34.282666 39.516530 3.591553
-30.807113 -11.363384 -6.427677 -2.833937 3.217770 -3.782121
-21.071254 2.388208 -4.099834 9.610558 -17.402015 -12.587598
GRADIENTS OF NEW GEOMETRY, GNORM=*************
************************************************************************
************************************************************************
************************************************************************
************************************************************************
************************************************************************
************************************************************************
CALCULATION ABANDONED AT THIS POINT!
SMALL CHANGES IN INTERNAL COORDINATES ARE
CAUSING A LARGE CHANGE IN THE DISTANCE BETWEEN
CHEMICALLY-BOUND ATOMS. THE GEOMETRY OPTIMIZATION
PROCEDURE WOULD LIKELY PRODUCE INCORRECT RESULTS
It is clear that the installed version of mopac that comes with Amber10 is not compiled properly.
If I follow the instructions for a stand alone mopac:
http://sourceforge.net/projects/mopac7
And get f2c (which seems to be important) to compile it with:
http://www.webmo.net/support/f2c_linux.html
Then everything seems to get back to normal and I can optimize geometry to my satisfaction.
In pdb format:
ATOM 1 C1 CAT 1 -0.691 -1.620 0.126
ATOM 2 C2 CAT 1 0.704 -1.620 0.126
ATOM 3 C3 CAT 1 1.409 -0.418 0.126
ATOM 4 C4 CAT 1 0.706 0.790 0.124
ATOM 5 O5 CAT 1 1.318 2.026 0.119
ATOM 6 C6 CAT 1 -0.706 0.793 0.128
ATOM 7 O7 CAT 1 -1.447 1.951 0.128
ATOM 8 C8 CAT 1 -1.403 -0.422 0.127
ATOM 9 H9 CAT 1 -1.239 -2.573 0.126
ATOM 10 H10 CAT 1 1.252 -2.573 0.124
ATOM 11 H11 CAT 1 2.508 -0.414 0.125
ATOM 12 H12 CAT 1 2.273 1.885 0.137
ATOM 13 H13 CAT 1 -0.832 2.700 0.136
ATOM 14 H14 CAT 1 -2.502 -0.409 0.127
This kind of things get me really worried. Then David mentioned bond type issue...
I would like to see any fixes to these problems if they are widespread.
Thanks,
-Ilja
On 8/7/09 2:19 PM, "David Mobley" <dmobley.gmail.com> wrote:
All,
I keep having issues with partial charges coming from Antechamber AM1-
BCC. In addition to the bug I mentioned previously relating to the -j
option resulting in incorrect charges when it should have no effect,
I'm also consistently finding that Antechamber generates substantially
different "AM1-BCC" partial charges than other implementations of AM1-
BCC. For a set of roughly 20 molecules, I obtained partial charges
from Christopher Bayly that he computed with his own, internal Merck
AM1-BCC implementation, and I also computed them with OpenEye's
Quacpac python toolkit. We then computed mean differences between
these charge sets. The Bayly AM1-BCC and OpenEye AM1-BCC
implementations agree extremely well, but Antechamber AM1-BCC charges
disagree with both Bayly and OpenEye partial charges reasonably
frequently.
The differences are nontrivial -- they can make up to several kcal/mol
difference in computed hydration or binding free energies.
I am attaching just one example of profound disagreement. The 2-
nitrothiophene.mol2 file contains Bayly AM1-BCC partial charges (sybyl
atom naming); 2-nitrothiophene_ante_amber.mol2 contains antechamber
AM1-BCC partial charges (amber atom naming), and 2-
nitrothiophene_oechem_amber.mol2 contains OpenEye AM1-BCC partial
charges.
In this example, the Antechamber AM1-BCC charges differ from the
others by more than 0.3 e on the nitrogen alone.
Anyway, I just wanted to put this out there to advise people. It
appears I will be ditching the Antechamber AM1-BCC implementation
until some of these problems are straightened out.
Thanks.
David Mobley, Ph.D.
Assistant Professor of Chemistry
University of New Orleans
New Orleans, LA 70148
dlmobley.uno.edu
Phone 504-383-3662
Fax 504-280-6860
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Aug 19 2009 - 23:28:32 PDT
