Hi all,

I am attempting to meaningfully compare two resulting list of significantly (padj < 0.05) differentially expressed genes in an experiment with a shared control, although I am getting unexpected results. This experiment used 5 replicates for each treatment and 5 shared controls.

For example, when comparing two different treatments against the control, this yields 1025 and 579 DE genes respectively. When comparing these two different treatments against each other, this only yields 18 DE genes. In addition, when looking at the shared genes of the resulting list when comparing the aforementioned two different treatments against the control, this yields 329 DE genes.

This is confusing, as it seems the two treatments are not that different from each other when directly comparing them against each other. But when comparing to the shared control independently, this tells a much different story. My question is - what may be the reason for this unexpected difference?

In addition, is there any available software or a better method to deal with these types of comparisons?

Thanks,

-R

hi R,

A and B can both be found to have significant differences when comparing to C, but might not be found to have significantly differences from each other. This isn't an inconsistency.

Imagine below I draw the range of sample gene expression within a group (just one gene):

A: [====]
B:     [====]
C:                   [====]

So here A and B might both be found significantly different than C, but their range of expression values overlap, so perhaps you don't find them significantly different from each other (you'd need more samples perhaps).

The numbers you have above don't need to add up to the same number, because there are other genes which may be different but not significantly so according to a test.