Perform DESeq2 with One factor with three levels vs Two factor with two levels

Question

[وفاء]

Hi everyone

I have patients RNA-seq data before and after treatment for one type of cancer. My goal is to use machine learning to predict drug response (response or resistance) and the treatment is one drug or two drugs. So, my question is what the best experimental design for Desq2 is:

1. One factor with 3 level: untreated (control) vs treated single drug vs treated double drugs

2. Two factors with 2 level: Factor 1: untreated (control) vs treated Factor 2: treated single drug vs treated double drugs

So in DESeq2, what is the main difference between one factor with 3 levels and two factors with two levels?

Answer 1

[allingt]

In scenario 2, what would the design formula be in your DESeq2::DESeq() call?

DESeq2 models gene counts according to your design. Then when you extract the results using DESeq2::results(), you specify the contrast to look at, by specifying a factor and the (I think only two) levels to contrast. For example, this could be control vs single drug, or control vs double drug. If single drug vs double drug is interesting to you, you should also make sure to get those results. I feel that log2 fold change only makes sense to calculate for a pair of levels at a time, but other experts could fill in on this.

Take a look at the DESeq2 tutorial: https://bioconductor.org/packages/devel/bioc/vignettes/DESeq2/inst/doc/DESeq2.html and specifically the section "Note on factor levels".

Answer 2

[swbarnes2]

With only three kinds of samples, I don't think two factors makes sense. Two factors would be like, having multiple genotype backgrounds which get tested against a treatment and a control. You essentially have a control and two treatments. So one factor, and specify in the contrast statement what two groups you are comparing to each other.