Re: Generation of average trajectory

From: Joshua Adelman (
Date: Mon Oct 12 2009 - 09:43:56 CDT

Hi Monika,

Depending on your system, this "average trajectory" is likely to be
unphysical and difficult to interpret. This depends heavily on how you
align the individual frames of your 4 hypothetical trajectories and if
the structures in each trajectory are similar/they proceed through the
same folding pathway. Since these trajectories are stochastic and
progress at different rates, it is unlikely that the average of any
number of conformations at a particular time, t, is going to be

I would suggest something more similar to what Nicholas, suggested.
Cluster all of your structures according to RMSD, and then look for
meta-stable states. You could then see if there are typical channels
that lead from cluster to cluster from your starting to unfolded
state. The Pande Lab at Stanford has done some work in this area as it
relates to building Markov models of folding and they have code
available. There are probably other ways that people have discussed in
the literature, but I'm not familiar with the field to a degree that I
can direct you to particular articles.

Best of luck.


On Oct 12, 2009, at 10:24 AM, Monika Sharma wrote:

> Hello.
> Thank you for suggestion. This is actually different from what I
> asked, but I appreciate your suggestion. It is a good way for
> representation.
> My question was that can simple coordinates averaging per time frame
> for a group of trajectories yield a meaningful average
> representative trajectory for that group?
> Reframing the question, if we have 4 multiple trajectories 1, 2, 3
> and 4. So, Is a trajectory generated as average of coordinates
> Rij = (Rij0 + Rij1 + Rij2 + Rij3) / 4
> be called as representative trajectory of that group , where i is
> atom number and j is the frame number.
> For instance, the coordinate of atom 2 (R23av) of this average
> trajectory at say time 3 ps be calculated as
> R23av = ((R23)0 + (R23)1 +(R23)2 +(R23)3) / 4
> ie sum average of atom 2 coordinates at 3ps of all the four
> trajectories.
> And also if I can use this average trajectory for further PCA
> analysis, will it be meaningful?
> All suggestions are welcome. If I am wrong, please correct me.
> Thanking you,
> Regards
> Monika
> On Mon, Oct 12, 2009 at 3:36 PM, Nicholas M Glykos
> <> wrote:
> > I have simulated 50 simulation trajectories of a protein unfolding
> > starting from same structure, but with different seed numbers.
> Then I
> > have clustered them according to their RMSD and Rg behavior. So
> for each
> > cluster, I wish to look at the average trajectory, which can be
> looked
> > as representative trajectory of that respective cluster. If I just
> > average out the coordinates per timeframe, then will it be a
> realistic
> > average trajectory representative of that set of 10 simulations? I
> am
> > going through the literature, but I am stuck and am not able to
> find a
> > better solution. If there is any other better and correct way of
> > generating an average trajectory, please suggest.
> I'm not sure what this 'average trajectory' may be. If you want an
> image
> showing the amount of mobility encompassed by a cluster, ie. something
> like this figure
> then, all you need is a set of representative (superimposed) PDB files
> corresponding to trajectory frames that belong to a given cluster
> (and,
> possibly, colored according to atomic fluctuations as calculated from
> the cluster).
> Apologies if this is not what you had in mind.
> --
> Dr Nicholas M. Glykos, Department of Molecular Biology
> and Genetics, Democritus University of Thrace, University Campus,
> Dragana, 68100 Alexandroupolis, Greece, Tel/Fax (office)
> +302551030620,
> Ext.77620, Tel (lab) +302551030615,

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:53:21 CST