[gmx-developers] pdb2gmx default water model

[Berk Hess]

Hi,

Thanks for the large response on this topic.
I know this is a delicate issue.

I realized my initial proposition of spc/e as a default is not a good idea.
For my own purposes this is usually the best choice, especially because
of the dielectric constant, but for others the trade-offs will be different.

Strictly using the water model used to parametrize the force field is also
not a good idea. CHARMM uses tip3p with LJ on the hydrogen which is
twice as slow in Gromacs, but also has even less structure than the already
very understructured tip3p model.

I am currently leaning towards always presenting the user with a list of
choices,
of which the first model is the one we recommend, with "recommended" on
the line.
I would then recommend spc for Gromos, tip4p for OPLS and standard tip3p
for Charmm and Amber.
That way the user at least realizes that there is a choice to make and
might look
more in the issue or simply pick the recommended model which is (in most
cases)
not a bad choice.

Berk


Thomas Piggot wrote:
> I agree. This seems like very a sensible compromise to me.
>
> Tom
>
> Justin A. Lemkul wrote:
>> Maybe there should be a hybrid approach from everything that's been
>> said.  Since even the use of SPC/e may be complicated depending on
>> the user's purpose, I don't think there is any true way to make
>> pdb2gmx incredibly smart.  I would vote for providing a default water
>> model for each force field (which is already in place for Encad,
>> anyway), but also always print a big, bold note (much like what
>> grompp does) at the very end of the pdb2gmx output saying that a
>> choice has been made and that this may or may not be suitable for the
>> user's purpose.  I don't see a way to satisfy everyone.  If I choose
>> Gromos96 53A6 for a protein, I may want SPC/e, but if that protein is
>> in a membrane, I want SPC.  There's no way for pdb2gmx to outsmart
>> the user, nor should it try :)
>>
>> GROMACS can only do so much to help the learning curve, but I do
>> agree with Michael's assessment that one needs to do some homework
>> before plowing ahead with eyes shut and fingers crossed :)
>>
>> -Justin
>>
>> Thomas Piggot wrote:
>>> Maybe I am wrong but I thought that different water models are more
>>> (or less) appropriate for certain systems, as well as being
>>> forcefield dependant?
>>>
>>> For example SPC has been shown to perform better than SPC/e when
>>> simulating membranes (Tieleman and Berendsen 1996 J Chem Phys) but
>>> SPC/e is better than SPC when simulating proteins (Hess and van der
>>> Vegt 2006 J Phys Chem B). Doesn't this make things more complex than
>>> having a single appropriate default model for a forcefield?
>>>
>>> And also what about the issues where it is not so clear as to which
>>> model is the appropriate default. For example CHARMM TIP3P Vs.
>>> standard TIP3P model for use with the CHARMM forcefield. Would it be
>>> standard TIP3P because of the increase in speed as discussed in the
>>> recent ffcharmm paper or would the more commonly used CHARMM TIP3P
>>> be a better choice?
>>>
>>> Because of these points I think having SCP/e as default with a clear
>>> message saying that the user should be warned that this may not be
>>> the best water model for their system would be the best way to
>>> handle this.
>>>
>>> Cheers
>>>
>>> Tom
>>>
>>> Anton Feenstra wrote:
>>>> On 16/06/10 17:24, Shirts, Michael (mrs5pt) wrote:
>>>>>> I thought about that, but I think that choice is asking too much in
>>>>>> terms of knowlegde
>>>>>> from the general user.
>>>>> See, I'm all in favor of making them think a little.  If you don't
>>>>> know what
>>>>> water model you should use, you don't deserve to be running a
>>>>> simulation.
>>>>>
>>>>> Make them look up what they should be running, as long as the
>>>>> error message
>>>>> pretty clearly says that - something like:
>>>> [...]
>>>>
>>>> I don't agree here. Yes, the user should be responsible for her
>>>> choices, but also a good tool should come with a *full* set of
>>>> appropriate default options. Then, it is clearly the responsibility
>>>> of the user to assert their validity for a particular application.
>>>>
>>>> Is/are there (a) paper(s) that say that either SPC or SPC/e is the
>>>> better water model? If most non-expert users would prefer SPC /or/
>>>> most reviewers would require that (for historical reasons of no
>>>> other /or/ some well-cited papers highlight SPC as the best option,
>>>> then I'd argue to leave that the default.
>>>>
>>>> Expert user will (want to) know better ;-)
>>>>
>>>> B.t.w., how transferable are water models between ff's? I've always
>>>> been thought that they are actually non-transferable (or at least
>>>> that is what I remember), making e.g. Amber/SPCe a bad option, as
>>>> would gromos/tip4p.
>>>>
>>>>
>>
>