still too many differentially expressed genes.

[thejustpark]

Hello,

I am doing a DE analysis on TCGA BRCA data.
In particular, I took 85 tumor-normal pairs from the data and am trying to figure out which genes are differentially expressed.
And surprisingly, I found that at least 9,000 out of 20,000 genes are.
So my question is if my analysis with this radical number can be right, and if not, what's wrong in the process.

To this end, I am employing the step 14 suggested in S. Anders et al. in Nat. Biotech. 2013 (Count-based differential expression analysis of RNA
sequencing data using R and Bioconductor), especially for pre-processing steps. Since DESeq2 is not mentioned, I did the analysis only with edgeR. Attached are the figures from plotMDS,plotMeanVar,plotBCS, and plotSmear functions according to the suggestion. I think they imply my analysis can be right, in the sense that many tagwise variations are less than common variation and the cases are separated quite well in MDS.

Interestingly, my another DE analysis using a widely-known non-parametric tool, SAMSeq, suggests around similar number of DE genes.

Again, can my analysis be OK? but I have never seen this kind of case with this number of DE genes.
So if my analysis is not correct, can you please tell me what's wrong?
Thanks in advance.

[dietmar13]

yes, why not.

what do you think, how many genes are differentially expressed between normal tissues (what is exactly the normal progenitor - the lobular or ductal epithelial cell - and where is the stroma, where are the blood and lymph vessels, ...) and the cancerous tissue (a mixture of cancer cells and cancerous remodelled microenvironment, like tumor stroma, cancer specific immune cells, ...).

i would suspect, that ALL genes are differentially expressed between tumor tissue (mixtures) and normal tissue (mixtures).

the number you get are reliable, IMHO.

the problem is that (all) normalization methods and many statistical methods rely on the assumption that most of the genes are not differentially expressed - which is violated in comparison of such different tissue-mixtures.

and another problem is that absolute expression heights matter, and not relative (to stably expressed genes, if such genes exist at all)...

dietmar

[thejustpark]

Thank you dietmar13,

Your comments are very insightful.
So what you are saying is that this can be actually a case that this number of genes are DEed in TCGA BRCA between tumor and normal even with its heterogeneity. Then, it is strange why this obvious and global pattern were not reported in a literature. Hm.....

But as for the problem you raised about the assumption of statistical methods that most genes are DEed, that is why I did this analysis again using SAMSeq, a sampling based DE analysis tool that (I believe to ) inspect each gene individually.
When SAMSeq returns a similar number of DE genes, I take it is not a problem.
What do you think?

HJ.