The R function 'hclust' will provide you a standard way of calculating the hierarchical clustering.

Perhaps you can find the following link useful, and, more generally, literature on ensemble clustering (a relatively recent area) should point you in the right direction.

http://users.cis.fiu.edu/~taoli/pub/icdm10-ensemble-zheng.pdf

Good luck!

Antonio Ciampi

Thank you for the help!

This is a recent area? I thought this was something mathematically simple and much like calculating a consensus tree from multiple classifications derived from bootstrapping or similar (that is done by many phylogeny-related software). The only difference is that I want to introduce the input distance matrices myself and generate a consensus tree from the combination of my inputs.

As far as I can see, Ensemble Clustering is a type of Consensus Clustering and there are many algorithms and methods available... http://en.wikipedia.org/wiki/Consensus_clustering

Lets see if I can find some software to help me with this... they use METIS and hMETIS algorithms... they can be downloaded, but I have no idea what they are... so I am happy there is a solution to my problem, but I am scared to look up at the learning curve that stands in front of me...

Hi,

I would recommend reading this article - they are doing exactly what you want and provided their code as R package.

Merged consensus clustering to assess and improve class discovery with microarray data.

2010 Simpson, Armstrong, Jarman

cheers

don't you want a Mantel test? that is a correlation test between distance matrices, here's an example

http://www.ats.ucla.edu/stat/r/faq/mantel_test.htm

Hi,

I am a formulation development scientist. I need to know a brief way on various designs like,box bekhen, D-optimal,Thiguchi etc. Is there any suitable website or free online books to get .

You can make cluster pattern from several different distance matrices through minitab 13 softweare programe or spss10

Thank you for the help!

Ok, what I've done is download PHYLIP, which can import a series of trees (compiled into a single file) and calculate a consensus tree using Ml methods (Margush and McMorris, 1981). These methods include partitions that happen in more than a proportion (between 0.5 and 1) of the input trees into the consensus tree. l=0.5 is called Majority rule (only groups that occur in more than half of the trees) and l=1 is called strict consensus (groups that occur in all trees). It also has an extended majority rule that attempts to resolve the whole tree by starting on l=0.5 and moving down on the required proportion, trying to fit everything. This last one is what I used, and I have a tree!

However, I will look at other options to see if they will better fit my problem. I am happy that I have a tree to work with if everything else fails.

@Corey Sparks: A mantel test will only calculate correlations among two matrices at a time. What I did was a second stage classification, by calculating Spearman rank correlations for every pair of distance matrices. I then clustered these results based on correlation and chose 11 matrices that form the biggest cluster from an initial 30 matrices. This was a way of reducing the "noise" and looking for consensus in a subset of matrices (although I am calculating consensus trees based on all initial matrices for comparison).