Hi, I m currently trying to run some clustering on some expression arrays and I was wondering about the best way of doing it, I have 81 samples on hgu133plus2 (55000), I have filtered this down to approximately 10000 (X, Y, low variabilty, control probes), and wanted to try hierarchical clustering on these both by arrays and genes. I was planning on using hopach as this seems an easy and obvious choice. How long would such a lot of comparisons take? I make it something like ( 81 * 10000 ) ^ 2 comparisons, I have a machine with 24gb of memory. Has anybody ever done something like this before? and what was the amount of time it took to actually do it? Given it might take a while are there any suggestions for how I might decrease the running time for such a program? I am already creating the distance matrix prior to clustering. Why is it better to use cosangle for gene clustering and euclidean distance for arrays? Is there a good reason for this and why would you use one distance over another. Many thanks in advance, Nathan [[alternative HTML version deleted]]

You may want to look at kmeans clustering instead of hierarchical if you are interesed in genes with correlated expression patterns across the samples. Imposing a hierarchical structure/model on 10,000 genes is probably incorrect -- gene A and B may be correlated but independent in terms of function, evolutionary history, pathway etc. In terms of how long it takes you would have to calculate a 10000*(9999)/2 = 49,995,000 element distance matrix -- my best suggestion is start the distance calculation and see if it gets finished in a reasonable amount of time. Bill Shannon, PhD Associate Professor of Biostatistics in Medicine Washington University in St Louis President-elect, Classificatin Society ________________________________________ From: bioconductor-bounces@stat.math.ethz.ch [bioconductor- bounces@stat.math.ethz.ch] On Behalf Of Nathan Harmston [iwanttobeabadger@googlemail.com] Sent: Monday, July 21, 2008 8:55 AM To: bioconductor at stat.math.ethz.ch Subject: [BioC] HOPACH clustering of genes Hi, I m currently trying to run some clustering on some expression arrays and I was wondering about the best way of doing it, I have 81 samples on hgu133plus2 (55000), I have filtered this down to approximately 10000 (X, Y, low variabilty, control probes), and wanted to try hierarchical clustering on these both by arrays and genes. I was planning on using hopach as this seems an easy and obvious choice. How long would such a lot of comparisons take? I make it something like ( 81 * 10000 ) ^ 2 comparisons, I have a machine with 24gb of memory. Has anybody ever done something like this before? and what was the amount of time it took to actually do it? Given it might take a while are there any suggestions for how I might decrease the running time for such a program? I am already creating the distance matrix prior to clustering. Why is it better to use cosangle for gene clustering and euclidean distance for arrays? Is there a good reason for this and why would you use one distance over another. Many thanks in advance, Nathan [[alternative HTML version deleted]] _______________________________________________ Bioconductor mailing list Bioconductor at stat.math.ethz.ch https://stat.ethz.ch/mailman/listinfo/bioconductor Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor