AMBER Archive (2008)

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

From: Jianyin Shao (
Date: Thu May 29 2008 - 16:47:43 CDT

> 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?

The agglomerative algorithms and k-means will perform rms fitting, but
SOM/bayesian/cobweb will not do furing clustering. For the latter three, rms
fitting is recommended to remove rotations accumulated in the MD.

> Is there a way to find out the frame numbers in the input trajectory
> that corresponds to the output representative structures?

In the output (.txt) file, the lines that look like ".....XXXXX....XX.." are
the composition of the clusters. Each "X" represents the position of frame
in the original trajectory. If you just don't like the fact that clustering
altered position of the frames, you can rms fit the resulting clusters on
the sugar 1.

There might be problems though. Say, if there are two subgroups of
conformation in which sugar 2-5 are relatively similar, but very different
in their position/configuration relative to sugar 1, the clustering will
assign them in a same cluster since the sugar 2-5 are similar. But the
question is why you don't run clustering on sugar 1-5, or just sugar 1-2 to
capture the conformation change of the linkage between sugar 1 and 2.

> 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
> DBI?

That probably needs a bit of code modification.



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