Model design for interactors / likelihood ratio test

Question

[andi.moor]

I need advice with RNAseq analysis & interactor model design for the likelihood ratio test. I would like to determine if the fold changes of genes in a versus b change according to fasting versus feeding. Hence I want to compute a fold change of fold changes. I intend to compare first a versus b by taking into account the pairing variable for both conditions separately and then to calculate fold changes of the results (FC a versus b) between condition fast and fed.

Here my experimental setup:

Location, Pairing, Condition

a, 1, fast

b, 1, fast

a, 2, fast

b, 2, fast

a, 3, fed

b, 3, fed

a, 4, fed

b, 4, fed

I want to specify this experimental setup in one interactor model and run it with Sleuth / likelihood ratio test (Pachter lab) to make use of all bootstrapping information that would be lost if I perform the analyses separately. Is this feasible / how do I have to specify the interactors location, pairing and condition for this purpose?

Thanks Andreas

[andrew.j.skelton73]

As far as I know, this isn't possible with the current implementation of Sleuth, but have you considered a multi-level experiment approach using limma? - That involves treating the pairings as a random effect using the duplicateCorrelation function from statmod. See section 9.7 from here