> I performed MD simulation using pmem.cuda (AMBER 2011 package) on GPU
> workstation and found that trajectory does not stabilized even after 10ns.
> the details are as follows :
>
> 1) MD simulation using ff99bsc0 forcefield: even after ~12ns of MD run
> the trajectory is not stabilized (TRAJECTORY SHOWN BELOW ) ???
>
> 2) MD simulation using FF99SB force field: in this case also trajectory not
> stabilized (RMSD plot below) ??
It is difficult to judge the results without more details regarding the
initial structure model, size of the system, etc., however as pointed out,
such RMSd values are not unreasonable for a 15-20mer duplex and the
fluctuations you are seeing may in fact be reasonable.
Some things to think about...
(1) what RMSd value do you expect? You need to consider movement of
chemically equivalent groups (like the hydrogens on the thymine methyl)
that have distinctive names, thermal fluctations (B-factors), expected
dynamics (opening, breathing, ion and water movement, ...), movement away
from the initial model structure (especially if it is canonical).
Clearly the RMSd to the initial structure will not be zero. A better
metric is convergence to the average structure and also consideration of
block averages.
(2) 10 ns is short, especially on a GPU, i.e. this is likely only 1 day of
simulation unless the system is rather large (in which case a 3A RMSd is
small). On what time scale do you expect the DNA structural properties to
converge? Ion/water distributions? Twist/bend fluctuations? These
definately converge on a timescale longer than 10 ns. Likely longer than
100 ns. Figure out how you are going to judge convergence noting that
very few simulations to date have demonstrated full "equilibration" i.e.
convergence of all of the structural properties monitored. Also note that
some things converge faster than others (for example solvent distributions
converge faster than ion distributions).
(3) Flat RMSd does not imply stability; stability is a thermodynamic
measure and low RMSd values only weakly correlate with stability. As
mentioned by others, you need to look at other indicators, for example
2D-RMSd, clustering, energetics (MM-PBSA), ...
(4) I would not suggest running DNA with ff99SB since you will see
anomolous transitions of gamma->trans that are effectively irreversible.
Note, however, that to see these accumulate often takes longer than 50 ns
so your structure at 10 ns may not be corrupted, yet...and yes, it is
clearly not converged.
--tec3
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Jan 17 2012 - 15:00:02 PST
