AMBER Archive (2009)

Subject: Re: [AMBER] ptraj usage for clustering of protein ensembles manual or tutorial

From: Thomas Cheatham (tec3_at_utah.edu)
Date: Sun Dec 06 2009 - 11:23:48 CST

> Yes, larger motions have some influence as far as the binding site is in
> the interface of homodimer and there are some little oscillations in
> monomers distance. But I actually also want to optimize somehow the
> number of clusters.

You can read our paper on clustering MD trajectories in JCTC (Shao et al.)
and from this you will see that the metrics are not always so clear. In
practice, my general path is to look at a 2D RMSd plot and count the
number of big squares on the diagonal. Alternatively, I look at the
cluster counts (i.e. how many frames in each cluster) which is reported at
the bottom on the *.txt file... i.e. below it looks like there are 4
major clusters (1, 2, 5, 6) based on the count; the graphical time course
suggests that the clusters are not revisited much...

#Clustering: divide 258425 points into 7 clusters
#Cluster 0: has 6647 points, occurence 0.026
#Cluster 1: has 133732 points, occurence 0.517
#Cluster 2: has 58864 points, occurence 0.228
#Cluster 3: has 243 points, occurence 0.001
#Cluster 4: has 1110 points, occurence 0.004
#Cluster 5: has 32013 points, occurence 0.124
#Cluster 6: has 25816 points, occurence 0.100
#Cluster 0 . ... .. . . . 54.1 . ....
#Cluster 1 9XX999X99XXX999999X9X9X98. . .....54.. .. ..
#Cluster 2 . ...... . . .1999999799994...
#Cluster 3 . ....... .
#Cluster 4 .1.
#Cluster 5 . .. .5781...29997
#Cluster 6 ..89997 ..2

-- tec3

_______________________________________________
AMBER mailing list
AMBER_at_ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber