Hi guys,

Probably my problem is not so complex but reasoning about the appropriate settings to test differential expression is often a source of confusion for me. Briefly, I wanted to test differentially expressed miRNA from miRNA-seq data. My dataset is composed by 8 samples in total, subdivided in 4 classes of 2 samples (replicates) each. I have 4 sample classes since the experimental design has two different factors with two conditions for each factor.

Now I just would like to use DESeq2 package to test DE miRNAs in the following comparisons: 

Exosome_wt vs Intracellular_wt

Exosome_mut vs Exosome_wt

Intracellular_mut vs Intracellular_wt

Exosome_mut vs Intracellular_mut

Obviously, the 'intracellular' condition refers to intracellular miRNAs and 'exosome' refers to miRNA expression from exosomes.

My main doubt is how to test these contrasts with DESeq2. Initially I supposed to create the DESeqDataSet object using both the experimental factors:

dds <-DESeqDataSetFromMatrix(countData = ReadCountTable, colData = sampleinfo, design = Cell_component ~ Type).

But then, if I understood correctly, the results function of DESeq2 will extract logFC/pvalue/adj.pval only for comparisons between levels of one factor, for example: 

res <- results(dds, contrast = c("cell_component", "exosome", "intracellular") OR

res <- results(dds, contrast=c("type", "mut", "wt")

But in my case, I wanted to test DE between combinations of factors. One solution I have tried: creating a new column in colData containing for each sample the corresponding combination of factors: intracellular_wt, intracellular_wt, intracellular_mut, intracellular_mut, exosome_wt, etc... and then using results to extract each time the output of each comparison on the new column:

for example:

res <- results(dds, contrast = c("new_column", "exosome_wt", "intracellular_wt") ).

However, I'm not sure at all this is the correct procedure. Can somebody help me?

Thanks a lot,

Francesco

Yes, I think your approach of adding a combination factor (design = ~ new_column) is a correct way to carry out the analysis.  It would be possible to achieve something similar if you had a design = ~Cell_component * Type with an interaction, but it wouldn't be as transparent as the approach you've suggested.  (I'm not sure your 'design = Cell_component ~ Type' is a typo, as generally DESeq2 designs are specified without a left-hand-side to the formula).  

One warning is that if you go on and look at set intersections of these genelists (e.g. mutation-differential in exosome but not in intracellular), then you're doubling up on potential statistical errors, and there are two-way designs which answer similar questions in one pass, so may be more appropriate - a local statistician would be able to advise.