AMBER Archive (2008)
Subject: Re: AMBER: silhouette width/coefficient after clustering by PTRAJ9.9
From: Dong Xu (d1xu_at_ucsd.edu)
Date: Thu May 29 2008 - 14:19:53 CDT
On Thu, May 29, 2008 at 11:15 AM, Jianyin Shao <jyshao2004_at_gmail.com> wrote:
> On 5/28/08, Dong Xu <dxu_at_mccammon.ucsd.edu> wrote:
>> On Wed, May 28, 2008 at 9:07 PM, Jianyin Shao <jyshao2004_at_gmail.com>
>> > On 5/28/08, Dong Xu <dxu_at_mccammon.ucsd.edu> wrote:
>> >> Hi Jianyin,
>> >> Does the cluster command already include rms fit? I tried on a
>> >> trajectory without doing rms fit at first and the output
>> >> representative structure look they are rms aligned on the atom select
>> >> mask used in the cluster command.
>> > The clustering does not do an explicit rms fit. However, it makes sense
>> > to
>> > do rms fit before calculating the centroid (average structure). So all
>> > structures will align to the representative structure of the most
>> > populous
>> > cluster before outputing. For those agglomerative algorithms and k-mean
>> > algorithm, rms fit is not required. Other algorithms will need an
>> > explicit
>> > rms fit command before clustering.
>> >> If this is the case, how to disable rms fit in cluster command? Do we
>> >> have something similar to "nofit" in rms command?
>> > As I just said, rms fitting is required for calculating meaningful
>> > centroids. Is there a specific reason you don't want rms fitting?
>> Yes, I have an oligosacchride molecule which has 5 sugar units. I
>> first aligned all frames on the 1st sugar unit and then wanted to
>> performed clustering on the rmsd matrix of the remaining 4 sugar units
>> which are not aligned. So I did averagelinkage on the mask selection
>> of the remaining 4 sugar units. When I look at the output
>> representative structures, they are no long aligned on the 1st sugar
>> units, instead they seemed aligned on the rest of 4 sugar units
>> probably due to the mask selection. That's why I suspect the "cluster"
>> command actually does rms fit during the clustering, does it not?
>> The distances reported in the output *.txt file are much lower than
>> the rmsd I got from "rms" command with "nofit" on mask selection of
>> the remaining 4 sugar units, which means the frames were probably
>> aligned during the clustering.
>> I hope my questions make sense to you and your help is much appreciated!
> Well, I think there is no switch to turn off rms fitting in the clustering.
> You may want to try clustering on all 5 sugar units. Another alternative is
> to rms fit by sugar 1 first, then run SOM/cobweb/bayesian clustering on
> sugar 2-5. But still I think the output structure will be aligned after the
> clustering and the pSF or DBI values are calculated based on the aligned
> average structures (centroids of the clusters).
So, it looks that clustering indeed perform rms fitting on the mask
selection. Does this imply that prior rms fitting of the trajectory is
not needed since the "cluster" command does it anyway?
Is there a way to find out the frame numbers in the input trajectory
that corresponds to the output representative structures?
I wish there is a way to turn off rms fitting during the clustering.
Is it absolutely required to calculate the statistics such as pSF and
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)