[gmx-users] A questin to Justin on methods in the PPARγ-RXRα-DNA complex paper
jalemkul at vt.edu
Sat Feb 20 20:18:22 CET 2016
On 2/20/16 1:59 AM, Timofey Tyugashev wrote:
> I was looking for examples of calculation setups. So I have a couple of
> questions about one used in this paper
> Parameters are amber99sb-ildn and gaff:
> "Simulations were carried out with GROMACS [40,41], version 4.6. All bonds were
> strained using the P-LINCS algorithm , allowing an integration time step of
> 2 fs. The Verlet
> cutoff scheme  was used with a minimum cutoff of 1.0 nm for short-range
> interactions and the real-space contribution to the smooth Particle Mesh Ewald
If you have questions about any of my papers, it's generally preferred to send
them directly to me, rather than airing them out over the mailing list.
> Why all-bonds constraints are applied? Isn't AMBER FF are supposed to be used
> with h-bond constraints? Is there any notable difference between 1 and 2fs
> timestep in this setup?
Any difference between constraining all bonds and h-bonds is probably
negligible. As has been noted several times on the list in response to your
questions, there are implementation differences and other assumptions that are
often very different between programs. Different nonbonded schemes, constraint
algorithms (LINCS vs. SHAKE), etc. make for a lot of potential differences that
have rather unknowable consequences.
I don't really know how to answer the second question. There's no point wasting
time at 1 fs when the simulation can stably be simulated with dt = 2 fs,
especially for something this large.
> Is box-solute distance of 1.0nm sufficient for 1.0nm LJ cuttof?
This was already answered the other day. A box-solute distance of 1.0 nm leaves
2.0 nm between solute images. It's enough.
> And why run three independent 500ns simulations instead of one or two of longer
> duration? Only to check for reproducibility or are there any additional reasons?
Mostly for better sampling. All simulations behave somewhat differently, but
generally you can observe the same average behavior. A single, long simulation
may get stuck in a minimum and give the false impression of convergence.
Multiple simulations that converge to compatible behavior is more credible.
Justin A. Lemkul, Ph.D.
Ruth L. Kirschstein NRSA Postdoctoral Fellow
Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 629
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201
jalemkul at outerbanks.umaryland.edu | (410) 706-7441
More information about the gromacs.org_gmx-users