Z score

Question

[Akash D]

Hi!

I am a bit confused here

1. Is Z score applied only on Expression Values? (say Normalized counts, VST transformed normalized counts, TPM, FPKM)
2. Can I perform a Z scoring on log2FC values of the genes across my treatment samples and visualize?

Thanks!

Sky

[Akash D]

I made this. Z scored on log2FC? will anyone support me on this?

3 groups A,B,C

Answer 1

[ATpoint]

You can apply Z-scores on whatever you want, but it must make sense and depends on what you want to do. What is the context? See for example Scaling RNA-Seq data before clustering?

Typically though it is applied to data on log2 scale because without log2-transformation standard deviations, hence Z-score range is very large and probably meaningless.

[Akash D]

Ok. However, if i perform it on log2FC, there wont be any control samples visible in my heatmap.

Your answer is perfect.

[ATpoint]

logFCs are a bit difficult to cluster I think, because fold changes without stats attached to it are meaningless. They can be very big, due to noise. I would cluster logcpms or vst. Not logFCs.

[Akash D]

Alright! Got you...

[Akash D]

I wonder what is the difference between a heatmap plotted with Z score (on normalized expression values) and the one plotted with simple log2FC?

What is the objective of making a Z score graph?

[ATpoint]

Difference is that plotting Z-scores plots each group and logFC plots differences between groups. Simply a different way of showing things. Depends on what you want to show.