DEseq2 interactions

Question

[drajangirija]

I am conducting DESeq2 analysis with three conditions and four celllinea, focusing on identifying differentially expressed genes (DEGs) for the comparisons of Control vs. Treated and Expt. Control vs. Treated within each cell line separately. The DESeqDataSet object has been filtered to include only Day 3 samples, and the design formula used is,

~ -1 + Cellline + Condition + Cellline:Condition.

The design matrix includes the following terms:

 ~-1 + Cellline + Condition + Cellline:Condition
 [1] "CelllineO"                       "CelllineP"                        "CelllineS"                      
 [4] "CelllineT"                         "ConditionExpt.control"               "ConditionTreated"                       
 [7] "CelllineP.ConditionExpt.control"  "CelllineS.ConditionExpt.control" "CelllineT.ConditionExpt.control"  
[10] "CelllineP.ConditionTreated"           "CelllineS.ConditionTreated"          "CelllineT.ConditionTreated"

I then generated the DESeq2 results for each condition comparison within the cell lines as follows:

• Day3_O_Treated_vs_control: Using results(dds, name = "ConditionTreated") for the comparison of Treated vs. Control in Cellline O.
• Day3_O_Treated_vs_Expt.control: Using results(dds, name = "ConditionExpt.control") for the comparison of Expt. Control vs. Control in Cellline O.
• Day3_P_Treated_vs_control: Using results(dds, name = "CelllineP.ConditionTreated") for the comparison of Treated vs. Control in Cellline P.

• Day3_P_Treated_vs_Expt.control: Using results(dds, name = "CelllineP.ConditionExpt.control") for the comparison of Expt. Control vs. Control in Cellline P. and so on for other celllines

  Just wanted to know what I am doing is correct?

Thanks

Answer 1

[swbarnes2]

You probably want to have data for all days in the same object, for better normalization and dispersion calculations.

Depending on what your PCA looks like, you may not want all the cell lines together in the same object. The assumptions underlying normalization may be violated. You might want to find DE genes for each cell line, then just compare fold changes to compare between lines. You don't seem to be really using interactions anyway.

If you want to compare just Day 3 Treated to day 3 Untreated, make a Day_Treatment column in your ColData, and make that your design. That's clearer than what you have above.

I don't know that I've ever seen an example of DESeq2 with a -1 in the design.