[gmx-users] Representative structure of a long simulation

[David L. Bostick]

Very well..

Perhaps averaging would give some strange results, however, in principle, a
system which samples it's free energy minimum should generate a set of
structures where {x}avg = {x}best .. taking {x} to be its configuration.
The spirit of essential dynamics will capture this because in a condensed
system such as a protein, many degrees of freedom may be "essentially"
constrained, and the rest will not be.  It is precisely this
"unconstrained" subspace that is most likely to cause strange things to
happen when averaging over an entire trajectory.  

I've never used the program, but g_cluster, can generate clusters of
structures.  I'm just brainstorming, but could you cluster your trajectory
of structures.  Find the most populated cluster, and average over this?


David 

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
David Bostick					Office: 262 Venable Hall
Dept. of Physics and Astronomy			Phone:  (919)962-0165 
Program in Molecular and Cellular Biophysics 
UNC-Chapel Hill					
CB #3255 Phillips Hall				dbostick at physics.unc.edu	
Chapel Hill, NC 27599	           		http://www.unc.edu/~dbostick	
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

On Mon, 29 Apr 2002, Marc Baaden wrote:

> 
> Hi there,
> 
> 
> thanks a lot for all those useful comments about my query.
> I will start trying these approaches, but some comments beforehand:
> 
> 
> - Nobody considered an energy criterium as discrimination possibility
>  (although Y. U. Sasidhar mentioned that the most populated conformation
>   should be representative).
>   Why ? Isn't it that in principle the frame with minimum energy should
>   be the one which will (statistically in the long term) be the most
>   populated ? I agree, that being interested in just a protein structure
>   leads to the question as to which energy has to be used (as the whole
>   system contains other components, too), but it seemed a canonical
>   choice to me.
> 
> 
> Berk Hess said:
> >> My suggestion would be to perform a covariance analysis [..]
> >> For large (membrane) proteins convergence can be a problem.
> 
> I agree, but the spirit of my question was more: let's forget about
> sampling issues and assume I have the structures in this trajectory,
> and just based on this trajectory which would be the one most representative
> for the sampled region of phase space.
> 
> 
> dbostick at physics.unc.edu said:
> >> This may seem too simple for you, but if I were looking for a
> >> "representative" structure of a protein, I would just take the time
> >> averaged structure.
> 
> No, it's not too simple, but as has been pointed out by Anton Feenstra
> and Y. U. Sasidhar, this average structure is not one of the snapshots,
> but might be an unphysical structure in terms of bond lengths, etc.
> This I wanted to avoid.
> 
> Cheers,
>   Marc
> 
> 
> -- 
>  Dr. Marc Baaden - Laboratory of Molecular Biophysics, Oxford University
>  mailto:baaden at smplinux.de  - ICQ# 11466242 -  http://www.marc-baaden.de
>  FAX/Voice +49 697912 39550  -  Tel: +44 1865 275380  or  +33 609 843217
> 
> 
> _______________________________________________
> gmx-users mailing list
> gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the 
> www interface or send it to gmx-users-request at gromacs.org.
>