Hi Jianyin,
Sorry for the delay in replying. I could not check my mails for some
days. My replies are interleaved below.
On Fri, Sep 3, 2010 at 11:09 PM, Jianyin Shao <jyshao2004.gmail.com> wrote:
> Hi Arvind,
>
> It looks to me that this approach probably may not work. When snapshot A and
> B both have same rmsd values to the reference structure, it does not mean A
> and B are similar, especially if the rmsd value is large. You would want to
> compare snapshot A and B to see if they should be in same cluster.
I agree that similarity in rmsd by any measure need not imply
similarity of the two snapshots. I would be certainly going back and
checking if snapshots that are clustered together indeed turn out
similar on whatever basis we are evaluating them.
> We are working on ptraj clustering so that other distance metrics may be
> used. Dihedral angles will be one of the possibilities. However, due to the
> circularity, the sum and average of dihedral angles could be tricky. I would
> suggest you still use the rmsd as the distance metric. If you have suspect
> some specific angles or dihedrals, focus on those specific residues.
I think there is some misunderstanding. I do not want ptraj to
calculate rmsd on the basis of angles. I have already written a code
for this purpose, and i did use directional statistics pointed out by
Bill. The code gives me a two-column output, with the first column
being the time instant, and second being the rmsd value (similar to
the ptraj output for coordinate-based rmsd). I need to cluster those
rmsd values on the basis of different clustering algorithms such as
k-means, bayesian etc. Ptraj already has a sub-package which does
clustering by all these algorithms. I was asking if i could input the
two-column rmsd output generated by me as input file to the clustering
sub-package of ptraj. In other words, rather than using a structural
coordinate file as input to ptraj, i wish to use a file with rmsd
values as input to ptraj. Is that possible? Hope the question is clear
now.
thanks,
arvind
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Sep 12 2010 - 23:30:03 PDT
