Hello transcriptomic folk,

DESEQ2 is very useful to assess genes differential expression but I was wondering at which point the DESEQ2 variance normalization step could distort transcriptomic profiles.

For example, we can imagine two different treatments with the same 100 expressed genes. In one treatment all the genes are expressed at the same level and in the other treatment very few genes represent the main part of the total reads. This has a great biological meaning but DESEQ2 would mask such information during the variance normalization step.

In this way, I would like to know if the subsampling could be a suitable approach for RNA seq data normalization. Subsampling would conduct to a loss of information, but I think that the loss of information is not a good argument if we can generate a normalized dataset which is closer of the “biological reality”.

Kind regards, Dr. François Maillard.

I'm not sure I understand the question fully. Are you referring to the size factor estimation? The median ratio method is fairly robust to the kind of differences seen in typical RNA-seq experiments. If you are concerned about a particular experiment, and global changes to the distribution, you could use the quantro Bioconductor package to try to detect this situation.