AMBER Archive (2008)

Subject: Re: AMBER: silhouette width/coefficient after clustering by PTRAJ9.9

From: Jianyin Shao (jyshao2004_at_gmail.com)
Date: Wed May 28 2008 - 10:51:12 CDT


On 5/27/08, Dong Xu <dxu_at_mccammon.ucsd.edu> wrote:
>
> >> If I want to fix the seed structure that I provide for decoy, I will
> >> have to specify "iteration 1", correct?
> >
> > If you specify "iteration 1", the clustering will stop after assigning
> each
> > frame to the closest seed. If no iteration value is specified, the
> K-means
> > refinement will kick in with the default iteration value of 50.
>
>
> If I use the default iteration (50 or 100?), could the seed structures
> that I provide for decoy (decoy_structure) be swapped to other
> clusters during the refinement?

It is possible that after refinement two decoy conformations may end up in a
same cluster especially when they are very close at the first place. The
decoy is designed to guide or initiate the K-means clustering algorithm. So
you can use some well-defined folding structures, not necessarily the
structures from MD trajectory, as the decoy structures. The refinement is
based on K-means algorithm, which will update the centroids of clusters and
reassign each frames to a closest centroids.

>
> >
> > The centroid is the average structure of the a cluster, it is not a real
> > structure and may not look like a real one.
>
>
> So, does this mean the *.avg.ci files created by ptraj are the cluster
> centroids?

Yes.

> The representative structure is
> > a real structure from the trajectory that is closest to the centroid of
> its
> > cluster. If the averagelinkage clustering result is different from the
> decoy
> > result, it is possible that their representatives are not same.
>
>
> If I set iteration=1 and provide the averagelinkage cluster
> representatives as decoy seeds structure, why are the decoy cluster
> representatives different from the seeds since "the clustering will
> stop after assigning each frame to the closest seed" as you mentioned
> above?

Your decoy seed structures are the representatives of clusters from
averagelinkage. When you do decoy clustering, the clustering (partition) of
the trajectory may be, most likely in your case, different from the
averagelinkage clustering. Please compare the aveagelinkage output and the
decoy output (the .txt files). After all, these are two different
algorithms.

Best,

Jianyin

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
      to majordomo_at_scripps.edu