Hi

I have clustered my normalised gene expression data using two methods. 1. NMF aheatmap 2. Using R's hclust function. I am trying to work out why these two methods give different clustering patterns despite me specifying the aheatmap function to use the same methods as the default R's function. Observe.

tdata <- t(data)
Raw2 <- tdata
dissimilarity <- 1 - cor(Raw2, method = 'pearson') # abs has been removed
distance <- as.dist(dissimilarity)
hc <- hclust(distance, method = 'complete')
plot(hc, main="Dissimilarity = 1 - (Correlation)", xlab="")

aheatmap(data, distfun = 'pearson', hclustfun = 'complete', scale = 'row')

This has been annoying me all morning, I have checked the aheatmap source code and it seems to use as.dist and hclust just as I am doing here, but the order of genes/probes is different - hence the clustering pattern is different.

Best,

Chris

Chris,

changes of a reply should increase with a reproducible code example (minimal, and incl. small dataset) and the plots you are seeing.

Best wishes

Wolfgang

Hi

It could be because aheatmap includes a scale or re-order function.  However looks 

 plot(hclust(dist(t(USArrests), method="euclidean"), "complete"), hang=-1)

aheatmap(USArrests, distfun = "euclidean", hclustfun = "complete") 

However when you have more than about 10 columns, aheatmap doesn't size the dendrogram and heatmap consistently.  Try  aheatmap(t(USArrests), distfun = "euclidean", hclustfun = "complete") 

I would recommend using ComplexHeatmap instead.  It has a lot of very nice features and works well.

Good Luck

Aedin