Voytek,
I think you indicated that you didn't see the structural problems
when running explicit solvent MD .... thus I don't think the problem
lies in your initial structure. (To be sure, start a GB simulation from
from a minimized structure taken out of your explicit simulation)
My experience with RNA simulations using GB would support the
"GB will be terrible for highly polar systems such as DNA / RNA"
statement. There may be small improvement when using the
updated chi torsion parameters (see ff10), but in general I think
GB struggles with nucleic acids. We still have problems with RNA
in explicit solvent, so using GB will likely lead to even worse results.
I have kept RNA together in GB by enforcing NMR distance restraints.
But all those restraints limit sampling and thus limit what you can learn.
Ditto, regarding Ross's comments about the tutorial.
--Niel
________________________________________
From: Ross Walker [ross.rosswalker.co.uk]
Sent: Tuesday, September 27, 2011 10:13 AM
To: 'AMBER Mailing List'
Subject: Re: [AMBER] GB protocol problems for large RNAs (follow-up)
Hi Voytek,
> Thank you for the thorough review of my GB protocol for a large
> RNA structure. I implemented all the indicated changes (most
> importantly
> moving to the Langevin thermostat), ran the equilibration for a total
> of 1ns and
> followed with MD. Unfortunately, the final results are a slow-motion
> version
> of the earlier structural failures, with the base pairs involving the
> 5' and 3'
> nucleotides, in the middle of otherwise robust, all-paired helices (
> can be
> viewed as two stacked helices) opening first, followed by the failure
> of the
> stacking (bending), rotations of the two halves around the connecting
> backbone and slow flattening of the helices. If these are indicators
> of known
> problems (with prep or GB itself), I would appreciate more
I am still hoping that someone familiar with running nucleic acid
simulations with GB will chip in here. From what you describe though it
looks like something is more fundamentally wrong with the simulation than
just issues with how GB describes nucleic acids. I am assuming you see these
changes on a fairly short <10ns timescale?
It would be useful to see your output for the first step to see if any of
the energies look off the scale. Beyond that I would check the GB radii in
the prmtop file - perhaps there is something in there with a strange radii,
zero perhaps? - Although I would expect this to show up in the GB energy in
the output. The other options might be an atom with a very high charge or
some other strange parameter.
I also would not rule out any steric clashes going on - so you might want to
check for close contacts.
I would also try just turning off the restraints entirely at the beginning
and see if the behavior is still there. It could be something weird going on
with the restraints.
Also, did you run the test cases for AMBER? - Just to make sure it isn't
something wrong (a compiler bug for example) in the compilation of the GB
code.
> information on it.
> No such issues occur in the explicit solvent approach.
>
> Dr. Walker commented that "GB will be terrible for highly polar systems
> such
> as DNA / RNA," and my experience seems to second that. On the other
> hand,
> GB protocol tutorial I read uses a small DNA fragment. Again, any
> enlightening
> comments would be appreciated (say, good for small nucleic acid
> structures,
> based on crystallography data, bad in other cases.)
You are referring to the first tutorial here which is Tutorial B1. The
purpose of this tutorial is to get brand new users up to speed on running
AMBER. It is not designed to make any comment on the perfect input settings
to use or the correct for field description for a simulation. DNA was used
here since it is easy to build with the tools in AMBER so avoids the
complexity of dealing with pdb files from the beginning. It is also good for
demonstrating the issues with gas phase simulations. Beyond that GB is used
just to highlight that one can run implicit solvent simulations as well as
explicit solvent simulations. It is not designed to suggest that GB is
optimum here for the arbitrary choice of solute molecule.
> Also, with a cutoff of 999, the GB execution speed (even with PMEMD)
> for
> my 264nt RNA structure appears to be comparable to the explicit solvent
> MD runs (on the order of 1ns of simulation time per day on a 24 core
> cluster),
> as suggested by Dr. Walker.
Oh you are good then. The performance can be very hardware specific but one
thing you may find is that the GB run will scale further than the implicit
solvent case.
All the best
Ross
/\
\/
|\oss Walker
---------------------------------------------------------
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Adjunct Assistant Professor |
| Dept. of Chemistry and Biochemistry |
| University of California San Diego |
| NVIDIA Fellow |
|
http://www.rosswalker.co.uk |
http://www.wmd-lab.org/ |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
---------------------------------------------------------
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Sep 27 2011 - 10:00:05 PDT
