How To Generate A Smaller, Simpler Heatmap Similar To A Tile?
2
3
Entering edit mode
11.4 years ago
k.nirmalraman ★ 1.1k

I am trying to make a heatmap and the was able to achieve the one I wanted. However, I have a feeling it looks clumpsy. Y is a 4 column matrix with expression values

y<-ave_scaledCounts[Gene_Markers,]
hr <- hclust(as.dist(1-cor(t(y), method="pearson")), method="centroid"); 
hc <- hclust(as.dist(1-cor(y, method="spearman")), method="centroid")  
# Cuts the tree and creates color vector for clusters.
mycl <- cutree(hr,k=4, h=max(hr$height)/2); 
mycolhc <- rainbow(length(unique(mycl)), start=0.1, end=0.9); mycolhc <- mycolhc[as.vector(mycl)] 
myheatcol <- bluered(75) 
pdf("Hierarchial_Clustering/Final_HeatMap Hierarchial clustering.pdf",width=9,height=7)
heatmap.2(y, Rowv=as.dendrogram(hr), Colv=as.dendrogram(hc), col=myheatcol, scale="row", density.info="none", trace="none",   
     +    RowSideColors= mycolhc,margins=c(10,10),keep.dendro=FALSE)
dev.off()

. I would like the heatmap in itself to be smaller in area... preferable like a tile!! The dendograms are not neccessary and the color key could be even thinner! ANy help would be great! Thanks!

EDIT: Would it just be possible to make a better heatmap if I just have the clusters in order already?? May be ggplot2 package enter image description here

heatmap • 5.5k views
ADD COMMENT
0
Entering edit mode

clearly this scale is not appropriate for correlations. wouldn't you normally calculate some kind of distance value?

ADD REPLY
0
Entering edit mode

I thought I was doing it here as.dist(1-cor(t(y)

Do you mean the color scale is not appropriate? If you dont mind could you expand a bit more??

ADD REPLY
0
Entering edit mode

have you tried just euclidean distances on the raw data?

dists<-dist(t(y))
heatmap( as.matrix( dists ))
ADD REPLY
0
Entering edit mode

This just made a huge square with same color, I mean the clusters cannot be seen. I added col=myheatcol"

ADD REPLY
1
Entering edit mode
11.4 years ago

Perhaps you could try myheatcol <- colorpanel(3, "#0000ff", "#ffffff", "#ff0000") or myheatcol <- bluered(3) to "force" each cell to fall into one of three colors.

Even narrower would be myheatcol <- colorpanel(2, "#0000ff", "#ff0000") or myheatcol <- bluered(2). In that case, a cell is either blue or red.

This probably won't work too well for smaller clusters, but it should work okay for larger ones, I'd think.

If you want to keep a blue-to-white gradient for values from -1.5 to 0 and have all z-scores greater than 0 be painted one shade of red, you could construct a color palette vector like so:

> myheatcol.bluewhite <- colorpanel(128, "#0000ff", "#ffffff") 
> myheatcol.red <- rep("#0000ff", 127)
> myheatcol <- c(myheatcol.bluewhite, myheatcol.red)
ADD COMMENT
0
Entering edit mode

Thank you!! Actually, the color gradient is just fine.. I would however like to know how can I

  1. reduce the size of the heatmap (width of the heatmap matrix)
  2. remove the dendogram without changing the order (the data can also be already pre-ordered if required... however, HC does a )good job too
ADD REPLY
0
Entering edit mode
11.4 years ago
Fwip ▴ 500

Have you tried heatmap.2( ..., dendrogram = "none" ... ) to turn off the dendrogram? Remember to check ? heatmap.2 for usage info.

ADD COMMENT
0
Entering edit mode

I tried dendogram="none" but did not help :( The problem is I am loosing the clusters if I do not include rowv and colv...

ADD REPLY

Login before adding your answer.

Traffic: 1736 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6