concern about statistical model after more than 78% of genes are differentially expressed according to DESeq2?

Question

[boaty]

Hi guys,

I got a RNA-seq data from the extraction of cellular droplet. Thanks to extraction method, a fraction of cytosol material : droplets, mixed with RNA and protein are separated from the whole cytosol. And now I want to do a comparison between all RNA in cell and RNA present in droplet like organelles in cytosols.

After deseq2 analysis, 78% of genes are differentially "expressed" . I put "expressed" here because those genes are not really expressed differently, they just concentrate at a special cellular organelle or enriched at that special organelle. the RNA's composition in this organelle is not the same as whole transcriptome in cell

And my concern is about that Deseq2 make an assumption about the majority of genes shoud not be differentially expressed. So with this huge percentage of differentially "expressed" gene, is deseq2 statistical model still good?

if a data is naturally very different between control and sample, can we still use deseq2 for the statistical analysis? any bias for the p value?

Thanks

[ATpoint]

it's a comparison between all cell

What does that mean, please elaborate on which cells were used and what the protocol is.

[h.mon]

because of extraction method, 78% of genes(all genes for that sp, up and down) are differentially expressed (extraction enriched some RNA number)

Please explain this also. If what you say is true, and extraction method causes expression differences between treatments, then you can't untangle true biological differential expression from the technical artifact.

[boaty]

thanks. the extraction will extract Droplet-Like Organelles in cytosol which only some special RNA is enriched. This is not a conventional DE, I am not comparing the expression between different times or different conditions. the comparison is between RNA in Droplet_like Organelles and all RNA in cell

[ATpoint]

Then you a priori know that most genes will change in a traditional DEG analysis. What exactly do you want to show with this experiment?

[boaty]

thanks again.

the experiment will give us insight of what gene's transcript will concentrate in that organelle. So here comes my concern, Deseq2's pvalue from nbinomialWaldtest() is key for filtering genes but if Desqe2 statistical model is not fit for my data, those p value will misguide the selection of potential candidates for biological validation.

[boaty]

thanks

I modified my text, the comparison is between cytosol droplet organelles RNA and all RNA in cytosol. So in droplet, a small part of RNA will be enriched.

my concern is that the stats model build by Deseq2 will not fit this data and give more false negative.

[boaty]

thanks

I modified my text, the comparison is between cytosol droplet organelles RNA and all RNA in cytosol. So in droplet, a small part of RNA will be enriched.

my concern is that the stats model build by Deseq2 will not fit this data and give more false negative.

[Friederike]

my goal here is to prove that extraction works

What would that entail? I.e., are you comparing two extraction methods and hoping that there are no genes differentially expressed when comparing the extraction methods?

[boaty]

thank you

I just updated my question, it will answer your question