[BioC] "3D" hierarchical clustering question

Aedin Culhane aedin at jimmy.harvard.edu
Wed May 30 02:27:13 CEST 2007

Dear Philippe
There are bi-clustering approaches available, however I don't know of a 
"3D" HC approach that is available in BioC. Maybe someone else does? 

However if you wish to use a principal components analysis approach, you 
can use coinertia analysis (CIA) available in the Bioc package made4 (or 
multiple coinertia analysis available in the R package ade4). There is 
an example of how to use CIA to link genes across different studies in 
RNews in Dec 2006. In that example we link across different platforms, 
in your case as all of the studies are on the same platform it will be 

Coinertia analysis constrained the axes of  the principal component 
analysis so that they are maximally covariant.  Therefore the axes 
capture the variance (principal gene expression trends) from each 
dataset, and will highlight those that are covariant across datasets.  
So you visualize correlated gene expression patterns across datasets.

We also have described how to link gene expression data from different 
studies using samples (Culhane et al., 2003, BMC Bioinformatics 
//*4(1):*59).  If you wish to use a supervised approach to find which 
genes are most associated with a classifier across datasets ,we have 
described a supervised extension to CIA (Jeffery et al., 2007 
/Bioinformatics/ 23(3) 298-305) and in a paper in press we describe 
linking protein and gene expression datasets (Fagan et al., Proteomics 
In Press), So this method can be applied to different types of data, not 
only gene expression profiles.

As you are specifically interested in one gene and genes "associated" 
with it, the  "iterative gene signature algorithm" approach described 
by  Bergmann et al.,  (Phys Rev E Stat Nonlin Soft Matter Phys. 2003 
Mar;67(3 Pt 1):031902.). I don't know if its available in BioC, however  
Jan Ihmels  implemented it in Expression Profiler available from the EBI 

Please contact me if you need further information about CIA, or how to 
use the package made4.


I would like to know if a package already available on BioC can do this:

I have data from multiple series of micrarrays coming from different 
experiments dealing with different tissus (different questions and 
projects but all on Affy U133A) etc... I would like to know more about 
one gene and the genes that are "linked" to this one.

What I do is 2D hierarchical clustering (samples/genes)Â for each 
experiment/project, and look at which genes are close to mine and look 
with Venn D through all projects what are the common genes close to the 
one I m interested in.

Is there a way/package that could "run" some kind of hierarchical 
clustering adding a third dimension (regarding the Projects) so that we 
obtain some kind of a 3D hierarchica lclustering...integrating the 
variability accross the different projects (different tissus etc...).

I would greatly appreciate any comment or help on this 


Philippe Guardiola, MD

More information about the Bioconductor mailing list