Normalize Bimodal Distribution

Question

[sp29]

I am analyzing a single cell dataset from SMART-SEQ having 300 cells. I am following Seurat's guided tutorial for clustering. I am confused about how to normalize the data, and I am using NormalizeData() from Seurat.

However, I do not see much difference between raw and normalized counts. The distribution still appears to be bi-modal; how should I proceed with the normalization? I there a workaround through which I can evaluate the normalized counts? It is a follow-up question from the previous post

Here's how the data looks,

[LChart]

It looks like you plotted the same thing 4 times with different labels. In particular, your raw counts here looks very different than in the previous question (i would expect to see some counts near 0 and no counts near 10^6)

Answer 1

[jared.andrews07]

Likely because count normalization has nothing to do with the plotted metric - it's just the number of RNA molecules detected per cell. Count normalization isn't going to change that value.

See this answer for more info.

[sp29]

Hi Jared, thanks for your response. I wonder if there is a way to evaluate whether the normalization has been done correctly or not. I can always plot counts vs. density, but is there any other measure or plot?

[LChart]

You could compute some of the metrics used in this pre-print and compare them between multiple normalization methods:

https://www.biorxiv.org/content/10.1101/2022.05.06.490859v1.full

[sp29]

Thanks, LChart, appreciate your quick response.