Dear all,

I have a design with temperature (2-levels: 21º C and 24º C) and strain (3-levels: strain1, strain2, strain3) and I would like to find differentially expressed genes specific to each strain for the 24º C vs 21º C comparison. That means:

1) finding those genes that are differentially expressed in strain1 for 24ºC vs 21ºC comparison that are NOT differentially expressed for strain2 for 24ºC vs 21ºC neither for strain3 for 24ºC vs 21ºC. 

2) finding those genes that are differentially expressed in strain2 for 24ºC vs 21ºC comparison that are NOT differentially expressed for strain1 for 24ºC vs 21ºC neither for strain3 for 24ºC vs 21ºC. 

3) finding those genes that are differentially expressed in strain3 for 24ºC vs 21ºC comparison that are NOT differentially expressed for strain1 for 24ºC vs 21ºC neither for strain2 for 24ºC vs 21ºC. 

Th model I selected is using the combination of temperature and strain as a single factor: design~temperature_strain, but then I have to perform overlappings to find specific DE genes.

Is the model with interaction: design ~strain+temperature+strain:temperature, a better model to extract the desired (specific) DE genes directly? what would be the contrasts?

thanks,

There is not a contrast that will give you the genes which are DE in you strain but not DE in the others. Instead, I'd suggest to build these tables manually, by compiling across individual results tables. Note that we have a test in DESeq2 for "not DE", if you can specify an LFC threshold below which you can consider it practically "not DE", see here:

https://bioconductor.org/packages/release/bioc/vignettes/DESeq2/inst/doc/DESeq2.html#tests-of-log2-fold-change-above-or-below-a-threshold