Re: [AMBER-Developers] [romain.wolf.gmail.com: Re: Amber atom type naming (and parms)]

From: Yong Duan <duan.ucdavis.edu>
Date: Sun, 18 Dec 2011 10:17:26 -0800

Hi Wolf,

The "fragment" thing is rather preliminary and your input is very
valuable. If possible, we'd love to have you actively involved in the
conversation. I think you know most guys I mentioned.

An important motivation is to enable polarizable GAFF in which ESP based
on high-level QM appears to be needed. To reduce the hurdle of
application, we need a method that can produce charges that are comparable
to MP2/aug-cc-pvtz level of accuracy. We think we may explore at least two
ways to move forward. One way is the AM1-BCC (or its equivalent, I.e.,
based on semi-empirical or low-level QM with BCC). Given its success in
reproducing HF/6-31G* level of charges, with re-parameterization, we think
this approach has the potential, although the long-range correlation might
be a concern.

The other is fragment-based. Here, the charges will be based on the
pre-"charged" fragments plus BCC (or its equivalent). I think these two
methods are mutually complementary. For molecules that have fragments
already built in the database, we use the fragments. For those parts that
do not exist in the database, the xxx-BCC method or full-scale
MP2/6-311++G**//MP2/aug-cc-pvtz can be used. We can start from a small set
of essential fragments and building blocks with the goal to cover 50% of
the commonly used organic compounds.

For substructure-search, optimally, we need to provide an automated tool.
If that is beyond our capability, we at least need a simple light-weight
molecule drawing tool to help the users to pick the fragments.

--
Yong Duan, Ph.D, Professor
UC Davis Genome Center and
Department of Biomedical Engineering
University of California at Davis
Davis, CA 95616
530-754-7632
On 12/18/11 5:56 AM, "Romain M. Wolf" <romain.wolf.gmail.com> wrote:
>The substructure search tool would be a must, but would not be
>sufficient. A major problem would still be the various connection points
>a single fragment might have. Unlike AAs and nucleotides, some larger
>ring fragments can have numerous attach points at which atom types would
>vary depending on the connection. Similar problems for partial charges.
>
>I guess that most users of GAFF in combination with any of the protein
>FFs start from ready, complete ligand structures (e.g. from x-ray or from
>docking). So they have the full structure (at best having to add
>hydrogens when not relying on automatic tools to do so).
>
>---regards---romain
>
>On 17 Dec 2011, at 19:02, Yong Duan wrote:
>
>> 
>> Thanks!!
>> 
>> Indeed, it's very challenging and impossible to cover all possible
>> fragments. In addition, we also face the challenge of a robust
>> sub-structure search tool.
>> 
>> What if the goal is to make a database of a subset of fragments that are
>> most commonly used or difficult to parameterize? Is a database of
>>hundred
>> fragments useful?
>> 
>> --
>> Yong Duan, Ph.D, Professor
>> UC Davis Genome Center and
>> Department of Biomedical Engineering
>> University of California at Davis
>> Davis, CA 95616
>> 530-754-7632
>> 
>> 
>> 
>> 
>> 
>> 
>> On 12/17/11 7:33 AM, "case" <case.biomaps.rutgers.edu> wrote:
>> 
>>> At Romain's request, I am forwarding parts of a recent email about
>>>force
>>> fields....dac
>>> 
>>> ----- Forwarded message from "Romain M. Wolf"  -----
>>> 
>>> Subject: Re: [AMBER-Developers] Amber atom type naming (and parms)
>>> From: "Romain M. Wolf"
>>> Date: Thu, 15 Dec 2011 21:44:24 +0100
>>> 
>>> A remark regarding atom types and the idea to use fragments for organic
>>> molecules.  My advice: DON'T!
>>> 
>>> Working in an environment where small molecules are as important as
>>> proteins or RNA/DNA, I must warn that the number of required fragments
>>> exceeds anything most people imagine.
>>> 
>>> ....
>>> 
>>> Regarding partial charges in small organics: the current 'fast' AM1-BCC
>>> method is not bad at all. A correction for symmetry would probably be
>>> a decent amendment (e.g., a methyl group should always have 3 Hs with
>>> identical partial charges). But despite this shortcoming, the charges
>>> themselves are very reasonable (and FAST to compute). I do not see how
>>> pre-computed partial charges would be incorporated into MANY thousands
>>>of
>>> required fragments.
>>> 
>>> [Junmei (or anyone): could we think about symmetrizing am1-bcc charges
>>> better?
>>> I think the code does some of this, but doesn't recognize methyls,
>>>etc.]
>>> 
>>> I have seen a few suggestions to generate comprehensive tables for
>>> all atom types in all Amber FFs. This is an excellent idea. These
>>> tables should be part of the documentation and not just be found in the
>>> parameter files. Also, definitions should be commented, by examples if
>>> necessary. Having ten definitions of carbon atoms in the style "Csp2
>>>in a
>>> ring" would not be of much help. And of course the occurrence of the
>>>type
>>> (which FFs it is used in) should be mentioned.
>>> 
>>> ---regards---romain
>>> 
>>> ----- End forwarded message -----
>>> 
>>> _______________________________________________
>>> AMBER-Developers mailing list
>>> AMBER-Developers.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber-developers
>> 
>> 
>> 
>> _______________________________________________
>> AMBER-Developers mailing list
>> AMBER-Developers.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
>--------------------------------------------
>Dr. Romain M. Wolf
>romain.wolf.gmail.com
>Tel. +33 6 63 06 94 33
>
>
>
>
>
>_______________________________________________
>AMBER-Developers mailing list
>AMBER-Developers.ambermd.org
>http://lists.ambermd.org/mailman/listinfo/amber-developers
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers

Received on Sun Dec 18 2011 - 10:30:02 PST