Dear Thomas, thank you for such a representative :) and detailed answer. The main difference for me is indeed that average is unreal anyway and that is the main cause of strange parameters, while representative is real and anyway should have some normal free energy landscape even with the maximums and mimums but maximums anyway shouldn t be somehow unreal. So to test this indeed I will test much more structures from the trajectory.
Best regards,
Andrew
16.09.10, 19:39, "Thomas Cheatham III" <tec3.utah.edu>:
>
> > So maybe the main question is about difference between representative
> > structures and average. As I understand the representatie structure is
> > one of the real structures in the trajectory.
>
> The representative structure should be equivalent to one of the structures
> read in from the specified "trajin" list of files. Although I know this
> to be true, I went ahead and tested this to be sure and this valid...
>
> So, why might the representative display "unusual" backbone angles? It
> comes down to issues similar the arguments about the utility of the
> average structure. If you have something in motion, perhaps oscillating
> back and forth between two states, the average structure becomes unreal or
> smeared. For example consider the two conformers below in exchange,
> followed by the expected average structure...
>
> \ / |
> | <-> | = |
> \ / |
>
> conf1 conf2 avg
>
> Now if we want the representative, is it conf1 or conf2? If we have lot's
> of conformations, the representative chosen will be the structure that
> is the closest to the "average" of the cluster (and therefore may
> display similarities to the motionally averaged structure).
>
> Can minimization "fix" either the average or the representative? Sure,
> but in the example above it will move towards either conf1 or conf2.
> Which is more representative, conf1 or conf2? [Both are.]
>
> What this suggests is that if you want truly representative structures
> that are not averages of interconverting substates that you need to have a
> sufficient number of clusters to match the number of definitive substates.
>
> For a simple system like the above, this is easy; for a mobile molecular
> system it becomes complicated. To see how complicated, read the
> paper that discusses clustering in ptraj and the influence on various
> metrics, algorithms, ... [Shao et al, JCTC (2007) 3, 2312-2334].
>
> When there is a clear distinction among sub-states, then clustering is
> easy. Real life (aka protein simulation) however is complicated.
>
> [p.s.
>
> For your docking investigations, I would suggest comparing protocols:
> (a) representative structures
> (b) minimized representative structures
> (c) randomly chosen structures
> (d) structures chosen at set intervals
> ...and also check whether the results depend on the
> # of structures chosen...
> ]
>
> --tec3
>
>
> _______________________________________________
> 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 Sat Sep 18 2010 - 04:30:03 PDT
