Hello Everyone,

I'm working on an RNAseq dataset from human cells in which transcription was globally affected (shut down) and would like to learn more about the limitations of using DESeq2 on such data when it comes to calling differential expression. I believe such global effects can distort the assumptions made by the statistical model behind DESeq2 and potentially make the downstream analysis heavily biased.

Is there such an effect really at play? And what are the best ways to combat this (both from the sample prep angle and the statistical analysis)? I would appreciate some pointers and possibly some literature covering this?.

Best regards, Piotr

Basically all methods that perform in silico normalization need to make the assumption that there are not global changes in expression, which are essentially effects that are 100% confounded with sequencing depth.

Alternatively, you can provide specific controls to estimateSizeFactors() as controlGenes. These could be genes that you know are not affected by the otherwise global up-regulation or spike-ins.