How would you explain to an 8-yr old why the log2 value output from DESeq2 is not simply the log2 of the averaged HTSeq count ratio ? Not that I expect 8-yr olds to be using DESeq2, but maybe your boss who is unfamiliar with differential expression methods and shows some scepticism towards the analysis.
HTSeq count cont HTSeq count treat DESeq log2
An example, Gene A 9 10 11 48 50 52 1.5
A five fold change in read count, but only 3-fold change in the log2.
Bill Gates and Warren Buffet are attending a large conference with many many "break out rooms". You randomly abduct 10 people from one of the break out rooms (as one does) and interrogate them until they tell you their net worth. The measured mean net worth is ~20 billion dollars, but of course it's highly variable because you just happened to have abducted two billionaires. Since the police are on to you, you can't afford to abduct anyone else to answer your burning question as to what the expected net worth at the conference is. So, do you go with your 20 billion dollar measured value or seriously decrease that because you know you don't live in a land of make believe? How much, then, should you decrease the measured value by? Well that would depend on (A) your sample size (if you'd abducted 1000 people you'd probably believe the estimate) and (B) the variability in values (maybe you're at a billionaires conference).
If you add some Gaussian and other distributions in then that's essentially how DESeq2 and other empirical Bayes-based methods work (i.e., why and vaguely how they shrink values, since obviously they don't abduct anyone...that we know of).
Sometimes there aren't simple explanations to complicated problems :(
Or, if you can't explain it simply, then you don't really understand it. Sounds cliche, but I believe it.