Confused about which comparison to give a more accurate answer in RNA seq experimen t

Question

[mohsamir2016]

Dear All, I have two cell types (wild type and knockout), which is once is mock-treated and once is infected with a virus. I would like to know what are the differential genes that are a results of knockout after infection. The design looks like this

To answer this question, I have seen two scenario :

1. To get the list of DE genes comparing Wt mock vs Wt infected with virus (list 1) [supposed to be the genes different due to infection of wild type)], and get also list 2 (comparing mock knockout vs virus infected knockout) [list of genes differ due to infection in knockout]., and then get unique genes that appeared only in list 2
2. is to directly compare virus infected wt cells vs virus infected knockout cells. Here I am skeptic if this is even valid, because the DE genes here is not solely due to knockout effect but also different cell background.

Which one of these 2- scenario would reveal the most accurate gene list ?

Thanks

Answer 1

[swbarnes2]

The way to properly do 1 is to use interactions. What that essentially does is give you the difference in fold changes (which of course you could just do yourself), but it also gives you a p-value for the difference.

2 is a different question. I would think that #1 is more interesting.

[mohsamir2016]

Thanks a lot for the answer: Can you please elaborate more on 'interaction': What I usually do is to run normal DE analyses c(i.e. using edgeR or any other packages) to get the DE between mock and Virus 1... in the design matrix for that, there is an interaction as you might know. Then I do the same for the other list, and use Venn diagram to reveal common or unique genes appearing in both or either comparison. Am I understood correctly ?

Could you also mention what is this different question you told about for the 2nd scenario ?

Thanks

[swbarnes2]

Read what the vignettes say about interaction. Simply comparing gene names list is kind of primitive; EdgeR has tools to do a more sophisticated analysis.