Re: Random seed and equilibrated structure

From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Fri Jul 27 2012 - 03:52:15 CDT

On Thu, Jul 26, 2012 at 11:40 PM, Dr. Eddie <eackad_at_gmail.com> wrote:
> Thanks Aron. I guess there is no way to make a small perturbation the an
> already equilibrated protein. I want to have good statistics for distance
> distributions in the equilibrated protein so having multiple runs helps. But
> I guess changing the seed can just thrust me back into a parameter space
> where it will likely take just as long to re-reach equilibrium. Right?

please keep in mind that there is not *the* equilibrated structure,
but rather an ensemble of equivalent structures. given the extremely
large number of degrees of freedom in a typical MD simulation,
this ensemble contains quite a large number conformations. please
keep also in mind, that some of those are probable and others are
less probable and that there are many conformations that cannot
be reached on the time scale of classical MD simulations due to
either having high (free) energy barriers around them or only very
narrow paths in phase space to get to them.

thus to create an ensemble of equivalent equilibrated structures
on your own, you will likely do a bit more to your system than
just using different random seeds.

as it was mentioned, parallel tempering replica exchange MD
is a method that can help get you over those barriers and thus
sample your equilibrium phase space better (of course you
sacrifice the dynamics of your system in the process).

..and if you know which barriers and transitions are of importance
to you, you can use one of the many other enhanced sampling
methods (MTD, SMD, TMD, US, TAD, PRD, ...).

cheers,
     axel.

> I have compared my rmsd to the initial frame as well since it seems to be a
> standard approach and does make sense, but I am also curious if others have
> a different method. My understanding was this is done because once the rmsd
> has leveled off your results are converged. I have noticed that sensitive
> things like the distance distribution between two residue alpha-carbons
> becomes very narrow once the rmsd of the protein has equilibrated. Prior to
> equilibration I get broader and multipeaked distributions.
>
> Thanks,
> Eddie
>
>
> On Thu, Jul 26, 2012 at 4:07 PM, Aron Broom <broomsday_at_gmail.com> wrote:
>>
>> I'm not sure if that would show you what you want. If your run has
>> stabilized, it suggests you are now in some kind of reasonably deep minimum.
>> Presumably you are thinking of doing multiple runs to make sure that it is
>> actually the deepest minimum near to your starting point, and if so,
>> randomizing the seed from your already equilibrated structure may end up
>> just telling you that you have high walls around that minimum, even if there
>> are other deeper minima nearby.
>>
>> Also, in terms of using the RMSD a measure of the simulation stabilizing,
>> do you just plot the RMSD as compared to the first frame? That is generally
>> what I've done, but I wonder if others have different approaches? The big
>> problem with that is the ambiguity of the RMSD, that is, if the RMSD levels
>> off at 3A, it means that two structures in the last part of your simulation
>> are different from the starting structure by 3A, but they could be more
>> different than that from one another. I've tried using VMD to do an RMSD
>> versus lag time or something, which would be a kind of fluctuation measure,
>> and found that could be more sensitive. I'd like to know what others do
>> with this, maybe there is something obvious I've missed?
>>
>> ~Aron
>>
>>
>> On Thu, Jul 26, 2012 at 4:55 PM, Dr. Eddie <eackad_at_gmail.com> wrote:
>>>
>>> Hi all,
>>> I've been simulating a protein with NAMD using constant pressure and
>>> temperature and PME. It takes anywhere from 10-24 ns for the rmsd of the
>>> protein's backbone to level-off. I'd like to have multiple runs to ensure
>>> good statistics but that overhead is debilitating. Would I be able to simply
>>> continue a run that has stabilized (after, say 30 ns) using different random
>>> seeds to get the same result as having multiple runs start with different
>>> random seeds and each running for 30+ ns?
>>> Has anyone tried this? I'd be interested in success or failure since my
>>> time scales are too long to try it myself.
>>> Thanks,
>>> Eddie
>>>
>>
>>
>>
>> --
>> Aron Broom M.Sc
>> PhD Student
>> Department of Chemistry
>> University of Waterloo
>>
>
>
>
> --
> Eddie
>

-- 
Dr. Axel Kohlmeyer  akohlmey_at_gmail.com  http://goo.gl/1wk0
International Centre for Theoretical Physics, Trieste. Italy.

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:21:51 CST