rna-seq heatmap z-score calculation - TPM?

Question

[lyc1ne]

I'm wondering if visualizing heatmaps with z-scores for TPM is an appropriate approach?

From what I've read on here, people typically work with log normalized raw counts. However, I've already done a number of heatmap visualizations z-score scaling TPM from star-salmon. Is this approach invalid? I'd like to avoid re-doing this work if not necessary. From my own checks the values seem similar to raw count z-scores.

Thanks in advance for any help!

Answer 1

[Kevin Blighe]

For visualisation purposes, it's okay to Z-scale TPM values. The idea is that, by scaling, it improves the visual interpretation, i.e., to the reader / human eye. Note that which ever heatmap function that you are using may already scale by default; so, please check.

The statistical inferences that you make on the data for the purposes of differential expression obviously would not have been derived from the Z-scale TPM values. I assume that you calculated these via a standard tool, like EdgeR, limma-voom, or DESeq2.

Kevin

[lyc1ne]

Amazing, thank you for the quick response! Yes, have done statistical inferences using standard RNA-seq tools like the ones you've listed.

Answer 2

[ATpoint]

Scaling is appropriate and makes sense in most scenarios. In RNA-seq we are typically interested in emphasizing differences between groups. A heatmap without scaling though will emphasize absolute expression levels. That is not a meaningful choice for both clustering and visualization. You can find details here in this older post of mine: Scaling RNA-Seq data before clustering?