Re: [AMBER] how to assess the convergence of the clustering

From: Thomas Cheatham III <>
Date: Tue, 28 Jun 2011 15:36:56 -0600 (Mountain Daylight Time)

> I have performed the clustering analysis employing the
> average-linkage method by using the ptraj module. Is there way to
> measure the convergence of the clustering. Can you give me some
> suggestions ? A kind professor suggest me that check if the same
> clusters obtained when the analysis is run over only half the trajectory
> as opposed to the full one, but i am confused this and i don't know what
> to do. Thanks very much and i am looking forward to your reply.

Block averaging; i.e. cluster the first half, then separately cluster the
second half of the trajectory: Are the representative structures similar?
(Noting that they will not necessarily be numbered the same, i.e.
cluster0 may not match cluster0 from the second clustering).

[Note that the question is probably not whether the clustering is
"converged" but whether the MD simulation is converged and does the
clustering give a representative picture of the structures sampled during
the MD.]

You could also try clustering with a different algorithm. Or different
sieve, or different number of clusters. Or compare DME to RMSd for
clustering... See the Shao et al. clustering paper, JCTC 3, 2312-2334

To see how many clusters are expected, visualize a 2D RMSd plot.

AMBER mailing list
Received on Tue Jun 28 2011 - 15:00:03 PDT
Custom Search