This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Heatmap with normalised count from DESEQ2

I was wondering which is better (or if one of this method is wrong ) for plotting in a heatmap normalised count from deseq2 : counts(dds,normalized=TRUE) : z-score( log2(norm count +1) ), z-score(norm count ) or just log2(norm count +1)

deseq2 heatmap
• 8,289 views • link
updated by Kristin Muench Kristin Muench • written by ZheFrench ZheFrench

Take a look at how they are used in the DESeq2 vignette: Heatmap of the count matrix.

Normalised [unlogged; untransformed] counts are used for QC purposes and are neither logged nor Z-scaled. In addition, the default passed to pheatmap is scale="none".

For downstream analyses, you should really be using the regularised log or variance stabilised counts, and you can Z-scale these, if you wish. You do not have to log these.

Elsewhere, from EdgeR, people use logged CPM-normalised counts for heatmaps. For other tools that derive values in FPKM, people usually log those too (they can also be Z-scaled via the zFPKM function). You'll never see me use FPKM data, though.

• 1 views • link
written by Kevin Blighe Kevin Blighe

1 answer

I've used scale() with scale=TRUE, center=TRUE on rlog-normalized, which is a bit like z-score. I've seen people also plot log-scaled expression values too.

One tip - if you use asinh() instead of log2(), you can create the same kind of scaling, but you don't have to add 1 in order to account for possible cells that equal 0. For comparison:

enter image description here

enter image description here

• 1 views • link
written by Kristin Muench Kristin Muench

Log in to answer this question.

More posts like this