RPM values of RNASeq for clustering and heatmap

Question

[cheischra]

Hi,

We have RPM (not RPKM) values for 7 different plant lines (3 reps each). However, I want to perform an unsupervised clustering and plot a heatmap to look for interesting patterns. My question is how to mean center and normalise these RPM values for visualisation of heatmaps and clustering? Will log2 transformation of these RPM values help me in representing that? For example, in the method section (expression data analysis) of this paper http://www.plantphysiol.org/content/168/4/1684/tab-figures-data they say that "RPM values were centered around the mean and normalized using the sum-of-squares method". How do I actually do like this?

Thanks.

Answer 1

[ATpoint]

Not discussing now if RP(K)M is a good choice for count normalization or not (it is not), what you probably want is Z-normalization aka standardization.

In R, given a count matrix with column = samples and rows = genes that would be:

zscored <- t(scale(t(your.rpm)))

I suggest you log2-transform your data before you do that.