[gmx-developers] Bug in g_cluster/gmx_cluster

[Berk Hess]

>The cluster centers, defined as the structure with the lowest RMSD to all
>cluster members, are not correctly calculated in gmx_cluster (of
>gmx-3.2.1). The problem is that the function 'analyze_clusters', which
>determines the cluster centers, is called (line 1376) after the function
>'plot_clusters' (line 1364), where one half of the RMSD matrix is replaced
>by the cluster indices. Thus, not only the cluster centers are incorrectly
>determined, but also the average RMSD values reported in the log-file
>(average RMSD of all pairs of structures in cluster as well as average RMSD
>of all cluster members to cluster center).

Looking at the code I guess you are right. But I could not reproduce the 
error.
I had lots of trouble running g_cluster, as it often wrote an empty xpm 
file.
I have worked on g_cluster several years ago, but after that there have
been a lot of changes.
Hopefully somebody else can fix this problem plus the writing of empty xpm 
files.

>Apart from that, I think that the clustering method denoted "linkage" does
>not represent a full linkage, but rather a single linkage clustering.

I programmed this and I think it does what is advertized in the g_cluster -h 
help.
So do you think (and if yes why) g_cluster does not do what is written in 
the help?
Or is full linkage not what has been written in the help?

>Yet another (though rather unimportant) detail, this time referring to
>gmx_rms.c: when calculating  an RMSD-matrix with -m, the reported average
>RMSD value is calculated incorrectly (except when using two trajectories
>with -f2).

I have fixed this.

Thank you for the bug reports,

Berk.

_________________________________________________________________
Play online games with your friends with MSN Messenger 
http://messenger.msn.nl/