Hi, suppose I have an experiment where I compare Drug A to control however I do this experiment with 3 cell lines. Lets say I run an RNA.seq on this and PCA analysis  shows a clear separation of cell lines but within each cell line, Drug and control separates out nicely.  The question is should I run the the analysis separately for each cell line and take the intersection OR should I run it all in the same model but correcting for cell line?  For example using limma it would look something like this. 

design <-  model.matrix(~0+ key$group + key$cell  )

the header would look something like this: 

 control  druga  cell1 cell2 

and the makeContrasts(pn=drug1-control  , levels=design)

if I do take this method how would I define my contrast to see the DGE for all three cell lines seperately? 

Or should I just do the analysis separately for each cell line and then take the intersection of the DGE from each cellline? 

so in summary its a two part question. 

1. should I run the model together or separately?

2. and if I run it together how do I define my contrast so that I find the DGE for each cell separately.  Say I want to find DGE only in cell3? 

thanks in advance! 

Ahdee

You kinda answered that yourself - if you are interested in the general effects of the drug (independent of cell type) you should use the combined model. 

If you are interested in the cell-type specific effects you would need to either include an interaction term in the combined model (model.matrix(~0+ key$group * key$cell  ) or do 3 separate models.

I would probably go with one combined model as you will utilize mode of the available data to estimate the variance.