extract dendrogram information from heatmap generated by heatmap.2

[crazyhottommy]

I asked the same question in Biostar, try my luck here:

Hi all,

I am having a hard time to extract the corresponding subgroup members in the heatmap generated by heatmap.2 in gplots. I have looked here: http://stackoverflow.com/questions/1...lustering-in-r and http://stackoverflow.com/questions/1...s-hierarchical and here https://stat.ethz.ch/pipermail/bioco...er/020521.html

I have a matrix m, the heatmap can be generated by

heatmap.2(m, colv=FALSE, dendrogram= "row", trace="none")

it gave me very good looking heatmap, now I want to extract the subgroup members.

I know I can use hclust and cutree to extract the information:

hc.rows<- hclust(dist(m))

plot(hc.rows) # it gives me just the dendrogram, and it looks the same as the dendrogram appeared in the heatmap, only differ in the order. it seems to me the order is reversed.

I can cut the tree based on either the number of group (k), or the height (h)

ct<- cutree(hc.rows, h=10) # it gives me 6 groups

rect.hclust(hc.rows, h=10) # draw red rectangles to mark the subgroups

table(ct)

ct

1 2 3 4 5 6

196 248 294 119 42 4

I can get the labels of each row, and the assigned the group number by:

tableclust<- data.frame(m,ct)

my question is that how do I map the cluster number here back to the dendrogram? cluster 6 has 4 members, but it looks like to me cluster 6 appears in the most left of the dendrogram, cluster 5 has 42 members, it appears in the second most left (based on the width of the red rectangle in the dendrogram), cluster 4 has 119 members, but it appears not next to cluster 5....

http://i43.tinypic.com/2hhktw9.png

I am confused.....

Thanks for your advice.

Tommy Tang

[Simon Anders]

Try this:

hm <- heatmap.2( m )
hc <- as.hclust( hm$rowDendrogram )
cutree( hc, h=10 )

The trick is that heatmap.2 returns an (invisible) list that contains an element "rowDendrogram" which you can feed to cutree (after converting it from a dendrogram object to an hclust object).

[crazyhottommy]

Try this:

hm <- heatmap.2( m )
hc <- as.hclust( hm$rowDendrogram )
cutree( hc, h=10 )

The trick is that heatmap.2 returns an (invisible) list that contains an element "rowDendrogram" which you can feed to cutree (after converting it from a dendrogram object to an hclust object).

Thanks Simon!
I did not know that I can convert the dendrogram object to hclust object.

however, I can do the reverse way:
hc<- hclust(dist(m))
rowDend<- as.dendrogram(hc)
heatmap.2(m, Rowv=rowDend, colv=FALSE, dendrogram="row")


Anyway, I figured out I can get the cluster members using:
cutree(hc, h=10) [hc$order]

in this way, I put the rows in the same cluster together, and the order is the same as appear in the heatmap. right?

Many thanks!

[sindrle]

I just get this error....

hc <- as.hclust( hm$rowDendrogram )
Error: all(vapply(s, is.integer, NA)) is not TRUE

[frymor]

I am getting the same error massage.

I have tried the example from here and it works great.
Also when I am trying to do it the way crazyhottommy did, it also works

But when I try my data I get the same error message as sindrle.

I have tried several subsets of my data and got partial success. I uploaded here a subset of my data (subset.txt), where I get the same error message as well as the heatmap resulted from this data set.

This is what I tried:
mydata <- as.matrix(read.delim2("subset.txt", row.names=1, quote=""))
mydata <- apply(mydata, 2, as.numeric)
test <- heatmap.2(mydata)
dend <- as.hclust( test$rowDendrogram )

But I get this error:

Error: all(vapply(s, is.integer, NA)) is not TRUE

Can it be, that the error happens due to the fact, that some of the genes (at a certain level) don't belong to any cluster at all?
Or maybe because a lot of the values are '0'?

Thanks
Assa

> sessionInfo()
R version 3.1.0 (2014-04-10)
Platform: x86_64-apple-darwin13.1.0 (64-bit)

locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

attached base packages:
[1] stats graphics grDevices utils datasets methods base

other attached packages:
[1] gplots_2.13.0

loaded via a namespace (and not attached):
[1] bitops_1.0-6 caTools_1.17 gdata_2.13.3 gtools_3.4.0
[5] KernSmooth_2.23-12 tools_3.1.0

[zzta]

Tried getting dendrogram but no success

Dear all,

Sorry for reviving the thread, but I have plotted a heat map and was proceeding to tree-cutting.

But the value rowDendrogram was no where to be found when I did:

> hm <- heatmap.2(log_tmm_selected_var_norm[,c(1,7,8,13)], scale='row', trace='none', col=redgreen(225))
> hm$rowDendrogram
NULL
> hc <- as.hclust(hm$rowDendrogram)
Error in as.hclust.default(hm$rowDendrogram) : 
  argument 'x' cannot be coerced to class "hclust"character(0)

I am not experienced with R and I keep getting this error. @@
I thought by default the dendrogram will be calculated for both rows and columns (that is why I see them on the device) but the value is absent.

<confused>

[crazyhottommy]

Just try to plot the heatmap, do not assign it to hm. Do you see a heatmap?

Dear all,

Sorry for reviving the thread, but I have plotted a heat map and was proceeding to tree-cutting.

But the value rowDendrogram was no where to be found when I did:

> hm <- heatmap.2(log_tmm_selected_var_norm[,c(1,7,8,13)], scale='row', trace='none', col=redgreen(225))
> hm$rowDendrogram
NULL
> hc <- as.hclust(hm$rowDendrogram)
Error in as.hclust.default(hm$rowDendrogram) : 
  argument 'x' cannot be coerced to class "hclust"character(0)

I am not experienced with R and I keep getting this error. @@
I thought by default the dendrogram will be calculated for both rows and columns (that is why I see them on the device) but the value is absent.

<confused>