[gmx-users] Cosine content

[Tsjerk Wassenaar]

Hey :)

This is pretty incorrect... A high cosine content is an indicator of
unidirectional motion in phase space. I've elaborated on that on this
list some while ago.

> Regarding your case, this means that, even if the RMSD is stable starting
> from 5ns, the protein still experiences "random" motions.

So the opposite is true: the protein configuration is adrift, i.e.,
it's undergoing conformational rearrangement, not random (thermal)
fluctuations. This rearrangement may happen in phase space far away
from the reference structure, thus not noticeable in the RMSD, which
will just say 'far away' (the larger the RMSD, the larger the volume
of phase space to be moving in).

> You should therefore consider as statistically relevant the trajectory
> starting from 30ns.

Arguable. The conformational rearrangement may be relevant and it may
even be a functional mode, which is just undersampled.

> Basing on my experience, I advice you to extend the simulation lenght
> (i.e.100ns) to ensure a larger dataset.

Of course, sampling longer makes sense :)


> Il 14/12/2011 00:37, R.S.K.Vijayan ha scritto:

> I have a 50 ns trajectory and looking at the RMSD plot, i set aside the
> first  5 ns as the time required for stabilization and subsequently carried
> out a essential dynamics for the backbone atoms post 5ns.

It is very likely that equilibration of the system takes longer than
5ns. Actually it is highly unlikely that it has converged with 5ns.

>  But when i perform a cosine content analysis for the eigen vector 1, i get
> these strange results
>
> a) Performing a g_analyze run using -b 5000 to -e 50000 ( ie 5- 50 ns )
> produces  a cosine value of 0.92

This suggests a transition. Look a the RMSD matrix for the trajectory
to see if you can spot a transition. As mentioned above, the
transition may be part of the equilibration, or reflect a functional
mode. That should be assessed using extended simulation(s).

>   but when i perform a cosine analysis using -b 5000 to -e 25000 ( 5-25ns) i
> get a cosine value of 0.24
>   also when i perform a cosine analysis using -b 30000 to -- 50000(30-50 ns)
> i get a cosine value of 0.05
>  so how should i interpret this result,

It's like doing PCA on the position of a canoe around a waterfall. If
you look at the whole stretch, including the drop down, you'll see the
thing getting closer to the fall more and more rapidly, taking the
plunge and then being pushed away from the fall. If you only look at
what happens before, there's just a slow drift towards the waterfall.
If you only look at the what happens after, the canoe is just wobbling
in the pond below randomly.

> does it imply that all the conformational changes are taking place between
> 30 to 50 ns ????

No, there appears to be a conformational rearrangement before 30ns.
Likely most obvious between 20 and 30 ns. Again, look at the RMSD
matrix.

> or is there some thing wrong with the convergence of the system??? Papers
> tell that the cosine value does decreases with increased time scale due to
> enhanced sampling, but strange a 25 ns simulation has a low cosine value on
> the contrary a 50 ns simulation has an increased cosine value.

Nothing strange. It depends on the system and what is happening in it.
But it doesn't exclude the possibility of bad convergence.

Hope it helps,

Tsjerk

-- 
Tsjerk A. Wassenaar, Ph.D.

post-doctoral researcher
Molecular Dynamics Group
* Groningen Institute for Biomolecular Research and Biotechnology
* Zernike Institute for Advanced Materials
University of Groningen
The Netherlands