[gmx-users] Implicit solvent

[ithu]

Dear gromacs Users,

I found this in the web, but I wanted to know if there exists the
possibility now of using implicit solvent efficiently.

Thanks,
Esteban

A repeating question on the mailing list whether GROMACS can perform
implicit solvent simulations. The answer is, not really. Over the last few
years there have been quite a few papers in (good) journals about why one in
general should or should not use it. Please search literature by Ruhong
Zhou, Vijay Pande and/or Bruce Berne on the subject (and fill in the
references here plus DOI links etc.).

   - R. Zhou and B. J. Berne. *Can a continuum solvent model reproduce the
   free energy landscape of a β-hairpin in water?*, Proc. Natl. Acad. Sci.
   U.S.A. 99 (2002), 12777-12782 DOI<http://dx.doi.org/10.1073/pnas.142430099>
   - Young Min Rhee, Eric J. Sorin, Guha Jayachandran, Erik Lindahl, and
   Vijay S. Pande. *Simulations of the role of water in the protein- folding
   mechanism*, Proc. Natl. Acad. Sci. U.S.A. 101 (2004), 6456-6461
DOI<http://dx.doi.org/10.1073/pnas.0307898101>
   - Hao Fan, Alan E. Mark, Jiang Zhu, and Barry Honig. Comparative study of
   generalized Born models: protein dynamics, Proc. Natl. Acad. Sci. U.S.A. 102
   (2005), 6760-6764 DOI <http://dx.doi.org/10.1073/pnas.0408857102>

 The current state in Gromacs is that we already have very optimized
assembly kernels for the actual generalized born interaction, so that part
is done. We also have C language functions to calculate Still radii (not yet
in CVS), although these have to be ported to assembly for decent
performance.

The one big remaining issue is a fast surface calculation algorithm. The
problem with the commonly used ones (e.g. Still) is that everything else in
Gromacs (including the GB loops) is an order of magnitude faster, so that
surface calculation would take over 90% of the time. They also do not
parallelize easily.

There are some tricks we can use (e.g. only calculating surface every N
steps), but we still need a *very* fast surface calculation algorithm. The
best starting point in the literature is probably the algorithm of Brooks,
where you simply have empiric parameters for sp2/sp3/sp neighbors of
different atom types combined with a short neighborlist.

We definitely need approximate derivatives of the surface free energy with
respect to all atom coordinates, and the last couple of years there has also
been some discussion that the volume term could be even more important than
the surface, so preferably volume derivatives too. If you're interested in
helping I (lindahl at cbr.su.se) have reference code that calculates both
surface/volume and the associated derivatives analytically.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20100705/704e0e2f/attachment.html>