I would like advice about software and appropriate approaches for clustering genes. I have results from a series of 45 cDNA arrays comparing RNA sample extracted from a hela human cell line. My collaborators have over- expressed a dozen different transcription factor proteins and compared the results back to a reference. There are 3-4 replicate arrays for each transcription factor. I have classified each gene as likely up, down, neither for each transcription factor. Now I would like to do a cluster analysis of the genes. I know that there is a long tradition of cluster algorithms for microarray results in the cancer context, but I don't have any experience with clustering using R. What software have people found useful for this problem? There are the cluster and class packages on CRAN, both of which look promising. Is there anything on bioconductor or anything in R specifically directed at microarray results? Many thanks Gordon ---------------------------------------------------------------------- ----------------- Dr Gordon K Smyth, Senior Research Scientist, Bioinformatics, Walter and Eliza Hall Institute of Medical Research, 1G Royal Parade, Parkville, Vic 3050, Australia Tel: (03) 9345 2326, Fax (03) 9347 0852, Email: smyth@wehi.edu.au, www: http://www.statsci.org

Hi, The packages 'mva', 'cluster' and 'cclust' are sure places to shop if you want to do clustering with R. >From my experience, many will listen to you as you elaborate about clustering, but will conlude with "it is like Eisen's clustering ?" (those red and green plot probably contain subliminal messages ;) ). To do "kinda the same" when this is an absolute requisite, I use something like: dist.likeEisen <- function(x, use="all.obs") { c.x <- cor(x, use=use) a.x <- acos(c.x) return(as.dist(a.x)) } hc <- hclust(dist.likeEisen(exprs(eset))) Hopin' it helps, L. -- - -------------------------------------------------------------- currently at the National Yang-Ming University in Taipei, Taiwan -------------------------------------------------------------- Laurent Gautier CBS, Building 208, DTU PhD. Student DK-2800 Lyngby,Denmark tel: +45 45 25 24 89 http://www.cbs.dtu.dk/laurent On Fri, May 16, 2003 at 01:01:21PM +1000, Gordon Smyth wrote: > I would like advice about software and appropriate approaches for > clustering genes. > > I have results from a series of 45 cDNA arrays comparing RNA sample > extracted from a hela human cell line. My collaborators have over- expressed > a dozen different transcription factor proteins and compared the results > back to a reference. There are 3-4 replicate arrays for each transcription > factor. I have classified each gene as likely up, down, neither for each > transcription factor. Now I would like to do a cluster analysis of the > genes. I know that there is a long tradition of cluster algorithms for > microarray results in the cancer context, but I don't have any experience > with clustering using R. > > What software have people found useful for this problem? There are the > cluster and class packages on CRAN, both of which look promising. Is there > anything on bioconductor or anything in R specifically directed at > microarray results? > > Many thanks > Gordon > -------------------------------------------------------------------- ------------------- > Dr Gordon K Smyth, Senior Research Scientist, Bioinformatics, > Walter and Eliza Hall Institute of Medical Research, > 1G Royal Parade, Parkville, Vic 3050, Australia > Tel: (03) 9345 2326, Fax (03) 9347 0852, > Email: smyth@wehi.edu.au, www: http://www.statsci.org > > _______________________________________________ > Bioconductor mailing list > Bioconductor@stat.math.ethz.ch > https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor

Hi Gordon, here some methods that I have sometimes found useful for visualization: - the function heatmap() in package:mva (together with package:RColorBrewer and package:geneplotter::colorRamp) - Correspondence Analysis - package:MASS::corresp - Multidimensional Scaling - package::mva::cmdscale I am not aware of universally optimal recommendations on which metric or variable scaling to use, so you'll probably need to try. For clustering the TFs, the class discovery package "isis" by Anja von Heydebreck could be useful [1]. And rather than completely unsupervised clustering, it may also be fruitful to define a "query profiles" and then look for genes that are close to these. Useful query profiles could be ones that are only "up" for a single TF, or for a pair of TFs, etc. Best - Wolfgang [1] http://www.molgen.mpg.de/~heydebre/software.html ------------------------------------- Wolfgang Huber Division of Molecular Genome Analysis German Cancer Research Center Heidelberg, Germany Phone: +49 6221 424709 Fax: +49 6221 42524709 Http: www.dkfz.de/mga/whuber ------------------------------------- On Fri, 16 May 2003, Gordon Smyth wrote: > I would like advice about software and appropriate approaches for > clustering genes. > k