On 2/26/13 9:23 AM, Sebastien Cote wrote:
Dear Justin (and Yun),
Since our last discussion on implicit solvent simulations, I have read a bit
more about how cutoffs in GBSA (or variants) implicit solvent simulations are
used in other softwares such as AMBER, CHARMM and NAMD.
It appears to me that many use switch or shift cutoffs. Have you tried this in
your tests? Such boundary conditions should remove any non-conservation of
energy problem. As you said, protein stability should also be tested though.
I haven't tested switching or shifting. In a theoretical sense, that would make
a lot of sense. My issues were practical, though. My implicit solvent
experience is all in the 4.5.x series of Gromacs, where one could use OpenMM
acceleration for GPU. I found a lot of features were unsupported (like
switching and shifting) and/or unreliable (systems crashing). So I gave up on
OpenMM for GPU and moved to CPU, where I encountered a bug that prevents use of
more than 2 CPU for running MD (which still exists). Since I was relegated to
only 1 or 2 processors for these runs, I tried to get the maximum performance
while still maintaining accuracy. Switching and shifting were obviously slower
than cutoffs, and as I said before, finite cutoffs were terrible. I decided on
exclusive use of infinite cutoffs due to the fact that they were (1) more stable
and (2) allowed the use of the accelerated all-vs-all kernels, which helped
performance.
Please note that none of the above applies to the native GPU acceleration in
Gromacs 4.6, but the bug limiting implicit solvent simulations to 1 or 2 CPU
still exists. The present GPU acceleration also does not support implicit
solvent simulations, IIRC. I have not needed to test it yet, though.
Moreover, many use a non-zero salt concentration giving rise to a Debye
screening (exp(-r*kappa), kappa^-1 ~ 1 nm under physiological concentration).
Such screening makes the electrostatic energy tends to zero faster. In GROMACS,
it is possible to specify a salt concentration for the implicit solvent scheme
in the .mdp file, but such variable has not been coded for in the source code
(as explained in the manual). From what I saw, this is still on the to-do list
of the developers. Is it still under development?
Theoretically, but I've not seen any movement on that issue. I will file a
request on redmine.gromacs.org. Anything not filed there typically gets lost in
the development process, so if there are things people want to see implemented,
they need to be filed.
Finally, there as been a post on the Gromacs mailing list late 2012 about some
energy differences between the AMBER and GROMACS implementation of OBC GBSA. It
was found that the parameters of the gbsa.itp file should be changed to obtain
a better energy agreement for the GB part. The SA part also showed some energy
difference that could not be corrected for and were attributed to different
surface accessible calculation algorithm. See :
http://lists.gromacs.org/pipermail/gmx-users/2012-November/076230.html.
Probably a worthwhile improvement. I'll file that on Redmine as well.
-Justin
Date: Tue, 26 Feb 2013 06:56:40 -0500
From: [email protected]
To: [email protected]
Subject: Re: [gmx-users] Why not PBC for implicit solvent?
On 2/25/13 10:30 PM, Yun Shi wrote:
Hi everyone,
Previous posts mentioned setting pbc = none for MD simulations with
implicit solvent. But I am trying to see the behavior of certain
concentration of ligands (small molecules, no big biomolecules) in
solvent, so I wonder if setting pbc = xyz would cause any problem for
my system?
I see no theoretical problem with running NVT with "pbc = xyz" with implicit
solvent, but definitely not NPT since the box will shrink inwards and lead to
periodicity artifacts (if it even remains stable at all). I usually set a
nonperiodic box because it allows me to use the infinite cutoff approach, which
is the only one I have found to give sensible results.
Should I also stay with the normal cutoff values for vdw and coul interactions?
Maybe, but do some serious testing before relying on the results. Maybe for
small molecules a finite cutoff will work, but for proteins I have tried cutoffs
of 1.0, 2.0, 4.0, and even 8.0 nm and all have unfolded or distorted.
-Justin
--
========================================
Justin A. Lemkul, Ph.D.
Research Scientist
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin
========================================
--
gmx-users mailing list [email protected]
http://lists.gromacs.org/mailman/listinfo/gmx-users
* Please search the archive at
http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
* Please don't post (un)subscribe requests to the list. Use the
www interface or send it to [email protected].
* Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
--
gmx-users mailing list [email protected]
http://lists.gromacs.org/mailman/listinfo/gmx-users
* Please search the archive at
http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
* Please don't post (un)subscribe requests to the list. Use the
www interface or send it to [email protected].
* Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
--
========================================
Justin A. Lemkul, Ph.D.
Research Scientist
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin
========================================
--
gmx-users mailing list [email protected]
http://lists.gromacs.org/mailman/listinfo/gmx-users
* Please search the archive at
http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
* Please don't post (un)subscribe requests to the list. Use the
www interface or send it to [email protected].
* Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
