Hi Everyone,
I want to use DESeq2 to perform statistical testing for:
A. The effect of “treatment” (IR_plus vs IR_minus), for each specific value of “tissue” (GFP_plus, GFP_minus, or All)
B. The interaction effect between "tissue" and "treatment".
That is, if changing “tissue” from GFP_minus to GFP_plus (or from All to GFP_plus) will change the effect of “treatment” (IR_plus vs IR_minus).
On the following design matrix:
tissue treatment pi_group
All IR_minus group2
All IR_minus group3
All IR_minus group5
All IR_plus group1
All IR_plus group4
All IR_plus group6
GFP_plus IR_minus group2
GFP_plus IR_minus group3
GFP_plus IR_minus group5
GFP_plus IR_plus group1
GFP_plus IR_plus group4
GFP_plus IR_plus group6
GFP_minus IR_minus group2
GFP_minus IR_minus group5
GFP_minus IR_plus group1
GFP_minus IR_plus group6
where “tissue” is the tissue type, “treatment” is the presence and absence of Ionizing Radiation treatment, and “pi_group” refers to the same tissue source that was separated into three tissues (All, GFP_plus, and GFP_minus).
One particularity of this data is that we do not have any data for "GFP_minus" tissue, from the tissue sources “group3” and “group4”.
If possible, please let me know which design matrix and contrasts would work best!
Thanks,
-Victor Missirian
I was hoping to perform statistical testing of the interactions between "tissue" and "treatment".
Specifically how changing the tissue from GFP_minus to GFP_plus (or from All to GFP_plus) would change the treatment effect (IR_plus vs IR_minus).
Thanks!
-Victor
I'm not sure you can use the group information in a fixed effects model (the kind DESeq2 uses) because it is confounded with treatment. How can you fit terms for the 2,3,5 group and the treatment minus group, because these are identical samples? Because you are interested in making comparisons across treatment, I don't see how you can control for group here.
If you just use tissue and treatment, this is a fairly straightforward interaction model.
Use a design of
~tissue + treatment + tissue:treatment.When you run DESeq(), set betaPrior=FALSE (this will be the default for designs with interaction in the next release of DESeq2).
"A. The effect of “treatment” (IR_plus vs IR_minus), for each specific value of “tissue” (GFP_plus, GFP_minus, or All)"
The treatment effect in the results will be the treatment effect for the reference level of tissue. That is, the first level in this vector:
See the vignette on how to set the reference level.
For the other levels, the treatment effect is the reference level effect plus the interaction term. You can add terms like so:
results(dds, contrast=list(c("treatment_plus_vs_minus","tissueX.treatmentplus"))(check resultsNames(dds) for the names of these).
"B. The interaction effect between "tissue" and "treatment"."
You can test if the treatment is different for tissue X vs the reference tissue like so:
You can test if the treatment is different for tissue X vs tissue Y (neither reference levels) like so:
results(dds, contrast=list("tissueY.treatmentplus", "tissueX.treatmentplus"))Thanks! This really helps me out!
-Victor
I have updated my original question to clarify what I have wrote above!