[gmx-users] re-equilbrating membrane protein system after g_membed and ligand addition
Peter C. Lai
pcl at uab.edu
Fri Jul 29 14:53:41 CEST 2011
On 2011-07-29 07:25:08AM -0500, Justin A. Lemkul wrote:
> Peter C. Lai wrote:
> > I was looking at the g_membed paper recently, and noticed the authors "only"
> > spent 10ns to re-equilibrate the membedded-system after g_membeding.
> > This they also did after bumping the temperature from 300 to 323K "to prevent
> > ordering of the bilayer" even though the "typical" simulation temperatures of
> > POPC is 300 to 310K...Anyone know why 323K was used? Was it to compare it
> > to the DOPC environment or something?
> I can't comment on this, but it seems odd to me, too.
> > I noticed for my particular system, I had to run a round of EM on the
> > post-membedded system in order to resolve some clashes before I could mdrun.
> > Then I ran 10ns in NPT with gen_vel=yes at 310K and position restraints on
> > the helix CA atoms. I am inferring that because the membrane patch I used had
> > been previously equilibrated after I constructed it and ran it under NPT at
> > 300K for 100ns, 10ns should be sufficient time to re-equilibrate with the
> > protein in the middle and 310K temperature?
> I generally find that membrane protein systems need at least 20 ns or so to
> really be equilibrated, but perhaps your system is a bit different. 10 ns is
> about the shortest time you can use to start seeing translational relaxation
> (rotational relaxation of lipids is shorter, roughly 5 ns).
Well I picked 10ns since the authors of g_membed claim that their g_membed
packs so well (at least compared to say, inflateGRO) *shrug*. In fact, they
claim that you can re-equilbrate in only 1ns after g_membed. In addition to
APL, they measured deuterium order parameter and the density profile of water
in z-axis columns, but they didn't elaborate on these other 2 metrics...
> > After the 10ns with protein restrained, I ran Grid_MAT on it and got a
> > reasonable APL (61-62A^2/lipid when taking protein atoms into account,
> > comparable to pre-g_membed patch) and the box dimensions look stable. Is
> > this sufficient to answer the abvove question?
> Could be. Seems reasonable.
> > Now I am going to be introducing a ligand to the protein. Is there a way to
> > preserve any state, like velocities of all the previous atoms? I was thinking
> > of the doing the EM while freezing everything but the ligand atoms and
> > allowing the ligand to change conformation (like a docking refinement). After
> > this, can I have the thermostat and barostat rescale the original velocities
> > (and gradually heat the ligand) instead of reinitialising everything?
> I doubt it. By introducing new atoms into the system, you can't use a
> checkpoint file any more so you lose the state you had before. You can preserve
> velocities in a .gro file, but I can't think of a reasonable way to have those
> read (while not having velocities for the ligand) and then heat up the ligand.
> You'd have to use "gen_vel = no" in conjunction with simulated annealing, which
> sounds like a recipe for instability. Is there any particular reason you think
> a specific set of velocities is required? Random sampling is all part of
> running proper simulations, so you need multiple, independent states to converge
> to get a reliable result, anyway.
Yeah didn't think so, but worth a thought. Just trying to look ways to save
computational time. Wolf et al (the g_membed authors) seem to suggest that
if the structure is already close-to or at equilibrium, then it should take
little to to re-equilibrate when you have minimal perturbation...
Peter C. Lai | University of Alabama-Birmingham
Programmer/Analyst | BEC 257
Genetics, Div. of Research | 1150 10th Avenue South
pcl at uab.edu | Birmingham AL 35294-4461
(205) 690-0808 |
More information about the gromacs.org_gmx-users