[BioC] average linkage clustering in Eisen's Cluster vs. that in hclust

[Antoine Lucas]

Hi,

Xcluster does same algorithm as Eisen software. This software can
be use within R with ctc package.

(Xcluster: http://genetics.stanford.edu/~sherlock/cluster.html)

One advantage of this method is that we are not oblige to store the
whole distance matrix: Xcluster requires very few memory allocation.

A problem could arrive by example when agglomerating 3 points of equal 
distance (an isometric triangle): 
 * first node will agglomerate two corners with a distance 'd'
 * second node will agglomerate the last corner with the midle of
   the other corners with distance 'd'.sqrt(3)/2 

Your tree will be difficult to draw as distance are not increasing while
agglomerating more and more... but can be easily skip.

Antoine Lucas.


Le Tue, 29 Nov 2005 13:27:13 -0500
"Wittner, Ben" <Wittner.Ben at mgh.harvard.edu> a écrit:

> Eisen's cluster software implements average linkage clustering by replacing
> nodes with an average of the leaf nodes represented by the node and by then
> computing distances from that average. hclust(), on the other hand, uses an
> average of the distances between all pairs of points in two nodes.
> 
> Does anyone have any information/opinion as to why one method might be better
> than the other?
> 
> Also, does anyone know of any code in R that implements Eisen's algorithm?
> 
> Thanks.
> 
> -Ben
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> 


-- 
Antoine Lucas
Centre de génétique Moléculaire, CNRS
91198 Gif sur Yvette Cedex
Tel: (33)1 69 82 38 89
Fax: (33)1 69 82 38 77