Hello everyone, I'm trying to produce a heat map that clusters 7120 genes into 6 groups based on 500 conditions. I'm using kmeans and then image, but I've two problems. The first one is that kmeans sometimes doesn't converge even with 10 restarts, and the second one is that the image produced is basically all read (I'm using the standard color scheme), not to mention it's size is massive and very hard to deal with. Does anyone have any suggestions on how I could accomplish this task efficiently, or is this data just too big to cluster? Thanks a lot, -Gaston

Hi Gaston, > I'm trying to produce a heat map that clusters 7120 genes > into 6 groups based on 500 conditions. I'm using kmeans and > then image, but I've two problems. The first one is that > kmeans sometimes doesn't converge even with 10 restarts, and > the second one is that the image produced is basically all > read (I'm using the standard color scheme), not to mention > it's size is massive and very hard to deal with. Does anyone > have any suggestions on how I could accomplish this task > efficiently, or is this data just too big to cluster? Genesis should be able to handle datasets that large (http://genome.tugraz.at/genesisclient/genesisclient_description.shtm l) Adapting the color scale is very easy. I can't comment on the convergence of k-means, this could depend on the data. Regards, Gerhard

Think you need to filter those genes as you almost certainly have too much noise . Suggest you filter the genes: if you are doing class discovery perhaps use the ratio of mean/sd - choose ones which the most variation and used about 500-1000 ish . The heatmap won't be readable in any case ....I would perhaps try principle components/ spectral decomposition like: the.pca <- prcomp(data,scale = TRUE) # for samples/genes (try using attributes(the.pca ) dim(the.pca$x) ### estimate PCA's you need the.pca.var <- round(the.pca$sdev^2 / sum(the.pca$sdev^2)*100,2) plot(c(1:length(the.pca.var)),the.pca.var,type="b",xlab="# components",ylab="% variance",main="Scree Plot for Hits",col="red",cex=1.5,cex.lab=1.5) savePlot("scree plot.jpeg",type="jpeg") centers<-15 the.cl<-kmeans(the.pca$x[,1:2],centers=centers,iter.max=1000) #Do kmeans colours <- rainbow(centers) ##2D plot(range(the.pca$x[,1]),range(the.pca $x[,2]),xlab="PCA1",ylab="PCA2",main="Spectral clustering of differential hits") text(the.pca$x[,1],the.pca$x[,2],label=rownames(the.pca $x),col=colours[the.cl$cluster],cex=0.75) library(scatterplot3d) ### 3D s3d<-scatterplot3d(range(the.pca$x[,1]),range(the.pca $x[,2]),range(the.pca $x[,3]),xlab="PCA1",ylab="PCA2",zlab="PCA3",main="Spectral clustering of differential hits",angle=120) text(s3d$xyz.convert(the.pca$x[,1],the.pca$x[,2],the.pca $x[,3]),label=rownames(the.pca$x),col=colours[the.cl$cluster],cex=0.75 ) points(s3d$xyz.convert(the.pca$x[wanted,1],the.pca$x[wanted,2],the.pca $x[wanted,3]),col=color,cex=5.0) Otherwise if you have class labels SAM or PAM. Hope that helps Cheers Paul -----Original Message----- From: Gaston Fiore <gaston.fiore@gmail.com> To: bioconductor@stat.math.ethz.ch Subject: [BioC] Heatmap with 7120x500 array Date: Fri, 27 Aug 2010 15:39:37 -0400 Hello everyone, I'm trying to produce a heat map that clusters 7120 genes into 6 groups based on 500 conditions. I'm using kmeans and then image, but I've two problems. The first one is that kmeans sometimes doesn't converge even with 10 restarts, and the second one is that the image produced is basically all read (I'm using the standard color scheme), not to mention it's size is massive and very hard to deal with. Does anyone have any suggestions on how I could accomplish this task efficiently, or is this data just too big to cluster? Thanks a lot, -Gaston _______________________________________________ Bioconductor mailing list Bioconductor@stat.math.ethz.ch https://stat.ethz.ch/mailman/listinfo/bioconductor Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor [[alternative HTML version deleted]]