getting clusters from R's heatmap function?

Question

I am using R's heatmap function to do hierarchical clustering (http://www.inside-r.org/r-doc/stats/heatmap). Is there a function that takes the object returned by heatmap and gives back the clustering you get by applying a threshold alpha to the dendrogram? A function like:

h <- heatmap(...)
clusters <- get_clusters(h, alpha=0.5)

where clusters is an array of same length as number of columns of the input dataframe, and returns a number between 0 and n-1 for each entry, where n is number of columns (assignment of columns to clusters). Does this exist in R?

Answer 1

You're looking for the keep.dendro argument of the heatmap function. The dendrograms are then stored in the Rowv and Colv elements of the output.

x = matrix(rnorm(25), ncol=5)
h = heatmap(x, keep.dendro=TRUE)

Next, convert the dendrogram object to a hclust object with the as.hclust function. Then, as stated in the comments, you can use cutree to get clusters.

row.clusters = as.hclust(h$Rowv)
cutree(row.clusters, k=3)  # break into k=3 clusters

Of course, you could have gotten to row.clusters by just doing

row.clusters = hclust(dist(x))