Can a lack of exome coverage disproportionately call snRNAs as significantly changing (and hide standard genes)?

Question

[Daniel]

Hi

I ran a question two days ago, trying to work out why some RNAseq data that I was analysing for colleagues wasn't particularly fruitful in differentiate genes (RNAseq comparisons (tophat->HTSeq->edgeR) return only a few sig diffs, and all snRNA. Pos. bias? Further options?) and we came to the conclusion that "it is what it is".

However, I hadn't thought to consider read depth originally, as I assumed (perhaps naively) that this would have been sufficiently decided before sequencing (and before I was consulted). I now see what I would consider a low-level of coverage, and I wonder if this could be affecting the results, particularly why we're observing these snRNA targets commonly (~12 out of 14 sig. diff genes).

This is a 2.5GB bovine genome with around 40million 2x75bp nextseq reads per sample.

Thanks for looking.

[Nicolas Rosewick]

Did you check on IGV the alignment on several genes ? How many replicates do you have ? Did you plot a MA-plot ?

Answer 1

[DG]

I'm not sure that a straightforward coverage analysis like this is necessarily a good way to look at RNA-Seq data, unless you controlled for unexpressed genes in your analysis. In humans (and it should hold for all mammals) any given tissue or cell type only expresses about 20% of the genome, so having close to 80% of your targets having essentially zero coverage , and a plateau with close to 100x coverage for 20% of targets seems pretty reasonable to me.

[Daniel]

That makes sense, thanks. I wouldn't normally try this, I'm just trying to work out if there was anything wrong with this dataset, but your reasoning is sound.