[R] about hcluster

Jari Oksanen jari.oksanen at oulu.fi
Tue Mar 20 12:32:33 CET 2007


XinMeng <xmeng <at> capitalbio.com> writes:

> 
> Hi all:
> As to hcluster,how can I control the cluster is performed according to
rows(genes for instance) or
> columns(samples for instance)?
> I can't find the parameters for it.
> 

Function hclust (if that was intended with 'hcluster') needs a dissimilarity
matrix as an argument. If those dissimilarities are between rows, you will get
clustering of rows ("samples"). If you dissimiarities are between columns
("variables") then you will get a clustering of columns. Technically this is,
easy. If d <- dist(x) gives you Euclidean distances between rows, then d <-
dist(t(x)) gives you Euclidean distances between columns (function t()
transposes its argument). In practice, the problem is that you need to find a
dissimilarity measure that is meaningful for columns. Such dissimilarity
measures are rare, but there may be some alternatives floating around in R.

cheers, Jari Oksanen



More information about the R-help mailing list