I am analysing in DESeq2 an experiment with the following design :
There are samples from 3 developmental time points. For each time point, there are three genotypes (knockout, heterozygote and control).
Besides testing for the effect of the timepoint & the genotype, the researcher is interested to see whether there is an interaction between the development and the genotype. That is, running DESeq2 with the design ( ~timepoint+genotype), and testing also for (~timepoint+genotype+genotype:timepoint).
As far as I understand, in regression, when the interaction term turns out to be significant, then the main effects in the design without the interaction terms cannot be trusted.
It seems to make sense that something similar would also take place in DESeq2. That is, if when including the interaction term many more genes become significant, then the analysis without the interaction term cannot be trusted.
My first question is whether am I correct in this.
My second question is how to test the significance of the interaction?
I was thinking of performing a likelihood ratio test, once with design (~ timepoint + genotype + genotype:timepoint), once with the reduced design (~timepoint + genotype). If the number of genes in this LRT test will be large, then there is no sense in using the design without the interactions. Is that a good way?