[gmx-developers] implicit solvent free energy

[Mark Abraham]

On Fri, Aug 30, 2013 at 8:44 AM, David van der Spoel
<spoel at xray.bmc.uu.se> wrote:
> On 2013-08-30 01:47, David Mobley wrote:
>>
>> Dear Devs,
>>
>> I wanted to get a sense how many people are interested in getting the
>> implicit solvent and free energy codes playing nice together, and how
>> likely it is to happen.
>>
>> We're very interested in this, as we do a lot of work related to free
>> energy methodology. This means in general we are more interested in
>> whether something works (and how well it works) given the force field
>> parameters than we are in having the solvent model right. In general,
>> implicit solvent will run a lot faster, so we could do a lot of our work
>> much more quickly if the free energy code would work well with implicit
>> solvent (such as generalized Born, etc.). We've also seen enough work on
>> implicit solvent free energy calculations elsewhere (and done some
>> ourself in AMBER) to be optimistic about the accuracies we could achieve
>> for a number of the problems we're interested in.
>>
>> In any case:
>> - Are other people similarly interested in this?
>> - Are there any efforts underway to get the GB code to work properly
>> with free energy?
>> - What are the problems/obstacles?
>>
>
> First problem is that GB does not work at all anymore in 4.6 and higher
> (correct me if I'm wrong). There are some loops missing IIRC. Maybe you
> could look into this if you're interested.

SIMD-accelerated all-vs-all kernels are certainly broken (patch in
http://redmine.gromacs.org/issues/1237 forces the use of the C
all-vs-all kernels; these seemed to work fine in my hands today,
though I was not looking closely). I believe finite cut-offs work with
Verlet scheme, not sure about group scheme. So I do not think there
will be mountains to move. If the kind of single-node case where these
kernels might work best is the general target, then I can probably
help fix the SIMD kernels. However, I won't prioritise that unless
someone is taking the lead for "I want to run these kinds of
calculations to publish with later, here's the theory we need to
implement, here's some test input, here's a sketch for what we don't
have that we still need." I have more than enough open-ended code to
write already ;-)

>From my memory of http://redmine.gromacs.org/issues/1237, if GB plus
any FE ever worked (AFAIK the implementation Per and Erik never had
that as a target?), it surely does not now.

I'm quite literally a noob at free energy calculations, so people can
expect very little code help from me on that side, unless there's
already a manual section written documenting the planned
implementation :-)

Mark

> Since I am rather skeptical about GB due to its approximate nature I am not
> sure that it is worth the trouble. You would have to compute the derivative
> of the GB term with respect to the born radius, right? The manual in section
> 3.18 is not entirely complete, but that might be tedious and expensive to
> compute as well.
>
>> Thanks!
>> David
>>
>>
>> --
>> David Mobley
>> dmobley at gmail.com <mailto:dmobley at gmail.com>
>> 949-385-2436
>>
>>
>
>
> --
> David van der Spoel, Ph.D., Professor of Biology
> Dept. of Cell & Molec. Biol., Uppsala University.
> Box 596, 75124 Uppsala, Sweden. Phone:  +46184714205.
> spoel at xray.bmc.uu.se    http://folding.bmc.uu.se
> --
> gmx-developers mailing list
> gmx-developers at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-developers
> Please don't post (un)subscribe requests to the list. Use the www interface
> or send it to gmx-developers-request at gromacs.org.