Comparing Clusters From The Dendrogram Using R Programming?

Question

[Mkl]

Hi all,

I need to compare the clusters of dendrogram by using R program? I did hierarchical clustering(protein sequences) using hclust function and I got 27 clusters. Next I would like to know the names of proteins in each cluster for comparison. Which function do I have to use for this?I used cutree function to cut dendrogram at a particular height.But I don't know how to find the elements of each cluster.I used following code to do Hierarchial clustering.

fit= hclust(as.dist(seq), method = "single") 
plot(fit)
democut=cutree(fit,h=20)
plot(fit, labels = as.character(democut))
table(democut)

How could I get the elements from each cluster and is there any package in R to solve this problem?

[Michael]

Add the protein names as rownames and colnames to your matrix seq, then you should be fine.

Answer 1

[Michael]

This is exactly in the documentation, read ?cutree and run example(cutree). In short, run:

hc <- hclust(dist(USArrests))

cutree(hc, k=5) #k = 1 is trivial
# output:
Alabama         Alaska        Arizona       Arkansas     California 
         1              1              1              2              1 
  ...

If you don't get the names, but only numbers, then your input data to hclust lacks names.

[Michael]

In this case, Alabam, Alaska, Arizona, are the "protein names". The integer is the cluster number.

[Mkl]

@Michael . I think this code cuts the dendrogram as five clusters. Actually I know the number of clusters in my dendrogram. My aim is to find the names of members(protein names) in each cluster. Any solution for this?

[Mkl]

I tried again this code for my data.But I got only the number of clusters. I didn't get a table like this.My data contains 228 sequences. I cut the dendrogram at h=20 using cutree function.From this I got the number of clusters as 27.

[Michael]

As I said before your input data must have rownames for this to work. Type rownames(yourdata), what do you see? otherwise show us the result of head(yourdata).

Answer 2

[Stevelor]

a more interactive way is:

plot(as.dendrogram(hclust(dist(USArrests)))

x <- identify(hclust(dist(USArrests)))

Use right click to stop..all selected subtrees are printed out... You can use x as a normal R object for further stuff

But it's sometimes a little bit stupid to use :D

HTH!