Hi,

I would appreciate some advice for a complex experimental design (complex at least for me!). I have a dataset with 3 factor variables, as follows:

4 genotypes A, B, C, D 

2 tissues A, H 

2 infection status UI, I 

Each combination of factor levels was replicated 2 times, giving 4 x 2 x 2 x 2 = 32 samples total.

I want to find the genes that respond differently to infection in the two tissues; and I am particularly interested in genes that would show the same pattern over all four genotypes. I guess that I have to look to the interaction tissue x infection effect over the four genotypes.

Assuming the above reasoning makes sense, does the following code accomplish the goals above?

dds = DESeqDataSetFromMatrix(countData = tab,
                             colData = colData,
                             design = ~ genotype + tissue + infection + tissue:genotype + genotype:infection +
                               infection:tissue)

I compared the likelihood of this model with that of a reduced model:

dds_full2 <- DESeq( dds, test="LRT", reduced = ~ genotype + tissue + infection + tissue:genotype + genotype:infection)

I got only 13 genes from this analysis where I expected much more based on previous analysis. So my code may be incorrect.  

I would greatly appreciate any help.

Thank you very much

This LRT will find genes where the infection effect is different across tissues, after accounting for baseline differences and accounting for each tissue having a genotype-specific baseline and each infection having a genotype-specific baseline.

"I got only 13 genes from this analysis where I expected much more based on previous analysis."

I can't help you here. If these genes look like there is in fact a differing infection effect across tissue, then you can either ask for a set with higher FDR or you can add samples. Remember, you have only 2 biological replicates.