Hi,

I have samples where one gene accounts for more than 40% of the total number of reads in normal conditions. In one phenotype that I consider, that gene is up-regulated.

How will that impact the differential expression of the other genes?

How does DESeq2 do the normalisation to avoid considering these other genes artificially down-regulated because of that?

Let's look at 'fake' numbers of RNA molecules:

Phenotype 1

Total number of molecules: ~ 2000

Phenotype 2

Total number of molecules: ~2500

The sequencing depth might be the same between the two, so normalising by sequencing depth is not going to help correct for that. Also, DESeq2 assumes a log normal distribution for the gene expression levels, but I was wondering if such a high read count for one single gene might make that assumption wrong?

I am unsure if this is simply equivalent to half of the genes being up-regulated in the sample, with no genes down-regulated, which DESeq2 is clearly equipped to tackle, or if it is different?

Could you explain how DESeq2 accounts for cases such as this one?

Thanks very much,

Delphine

EDIT: I attach an MAplot, and changed up to down and down to up as my plot was the other ay around compared to what I had written (the gene I am talking about is up-regulated in this plot, because of the condition considered as baseline)

hi,

So the DESeq2 normalization (and similarly with edgeR's normalization method) is not thrown off by a minority of genes with differential expression, because it uses the median of ratios across all genes. Even though a single gene accounts for 40% of the reads, it has little leverage on the size factor calculation because it is just one gene out of thousands, and the median across genes is used.