Re: [AMBER] AmberTools23 changes antechamber behavior for GAFF2

From: brian.radak--- via AMBER <amber.ambermd.org>
Date: Thu, 6 Jul 2023 19:51:01 -0500

Sorry to crash the thread, but this seems like an appropriate place to
re-post an issue I had a few months back with parmchk that did not
receive a response then:

On Mon, Jan 30, 2023 at 10:43 AM Brian Radak <brian.radak.gmail.com> wrote:

    Hi all,

    I accidentally stumbled across a possible bug in parmchk2 - it looks
    like specifying -s gaff2 occasionally grabs the original gaff
    parameters by mistake.

    This appears to happen stochastically for the same mol2 input with
    GAFF2 types (antechamber seems to always get this assignment right),
    but then the frcmod from parmchk2 occasionally assigns parameters to
    new GAFF2 types by analogy with old GAFF types (specifically nu and
    nv get the old nh parameters).

    Has anyone seen this before? I have a MWE.

    Cheers,
    Brian


On 7/5/23 9:20 AM, Junmei Wang via AMBER wrote:
> Hi John,
>
> Here are my answers to your questions:
>
> 1. For the majority of users, we recommend to use the latest version of
> gaff/gaff2 force fields. For those who want to do comparison, the easiest
> way is to use parmchk2 with the force field parameter file (gaff.dat and
> gaff2.dat) released in different amber versions.
>
> 2. For PS4, I will incorporate it into the antechamber package. So far, we
> support gaussian, gamess.
>
> 3. The energy function is the same for both gaff/gaff2.
>
> 4. We do not hardcode atom type definitions. Users can define his/her own
> definitions in ATOMTYPE_GFF.DEF and ATOMTYPE_GFF2.DEF.
>
> 5. Please provide me with your molecular files (such as mol2 or sdf) if you
> have difficulty/concern running antechamber/gaff, I am happy to help.
>
> All the best,
>
> Junmei
>
>
>
> On Wed, Jul 5, 2023 at 1:34 AM John Chodera<john.chodera.choderalab.org>
> wrote:
>
>> Hi Junmei!
>>
>> Thanks so much for the quick response on a holiday!
>>
>> For the gaff and gaff2 force fields, I recommend using the latest
>>> versions if possible, which should give you better results.
>>
>> We are happy to recommend that users always use the latest versions, but
>> we believe it is important for users to be able to specify exact versions
>> for reproducibility, and that access to historical releases is essential
>> for benchmarking and comparison purposes.
>>
>>
>>> For gaff2, the RESP charge calculated using HF/6-31G* ESP is preferred,
>>> but you can still use am1bcc for a large set of molecules.
>>
>> If you can recommend any workflows that involve conda-installable QM
>> packages such as psi4, we would love to be able to integrate these as well!
>>
>> gaff has the same atom type definitions for all the versions, in the later
>>> versions of gaff2, c5/c6 atom types have been introduced. However, parmchk2
>>> applies a special mechanism allowing a user to correctly assign force field
>>> parameters even if there is a mismatch on the atom type definition and
>>> force field parameter file. For example, there is c5 or c6 defined in one
>>> molecule, but c5 and c6 do not show up in the early version of gaff2, all
>>> the parameters involved by c5/c6 will be replaced with the ones of c3, as
>>> c3 is the corresponding atom type of c5/c6.
>>>
>>
>>> We do not recommend using the old versions of force field parameters, but
>>> if you want to make comparison, you still can assign atom types using
>>> antechamber (-at gaff for gaff or -at gaff2 for gaff2), and then generate
>>> correct frcmod file with an older version of gaff or gaff2 by using -p
>>> argument.
>>>
>>
>>> But, it is a common mistake if the atom type definition is gaff, but the
>>> force field parameter file is for gaff2, or vice versa. Please make sure to
>>> use "-at gaff" running antechamber and "-s 1" running parmchk2, or "-at
>>> gaff2" running antechamber and "-s 2" running parmchk2.
>>
>> Here is the link to a gist where we do this, but the molecule ends up with
>> different frcmod files only if we use AmberTools23:
>> https://gist.github.com/jchodera/fffb36ca605ea3b5654644cc1fe2ec16
>> Earlier versions of AmberTools 19.5 through 22 produce consistent frcmod
>> files using this mechanism.
>>
>> Is it the case that the energy function would be exactly the same for the
>> AmberTools23 version that involves new c5/c6 parameters as for the earlier
>> versions?
>> Or is there a way to get AmberTools23 to produce consistent frcmod files
>> with the earlier versions of AmberTools?
>>
>> My earlier suggestion was to enable us to override the exact atom
>> definition files that antechamber users, which are hardcoded by the `-at`
>> flag based on the supplied string token.
>>
>> Thanks so much!
>>
>> John
>>
>> --
>> John D. Chodera
>> Member, Computational and Systems Biology Program
>> Memorial Sloan Kettering Cancer Center
>> email:john.chodera.choderalab.org
>> book a meeting:https://calendly.com/john-chodera
>> office: 646.888.3400
>> mobile: 415.867.7384
>> url:http://www.choderalab.org
>>
> _______________________________________________
> 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 Thu Jul 06 2023 - 18:00:02 PDT
Custom Search