I was asked to run a 2-way ANOVA using LIMMA with strain (AJ, BALBC, C57BL6, DBA, FVB) and treatment (drug, naive, vehicle) as my factors. They want to know which genes change in expression due to strain, treatment, and the interaction. I ran the two-way ANOVA using treatment-contrast parametrization with C57BL6 naive as my baseline but they have asked me to run it without a reference. However, if I do not specify the reference then it will pick AJ drug as my reference.  The contrasts are different but the F statistics are the same. From reading the limma guide, my understanding is that the factorial 2-way ANOVA needs to have a baseline.  Is it possible to run a 2-way ANOVA without a baseline?  

Thanks in advance for your help, 

Theresa

snippet of script from my original 2-way ANOVA

#Model design
strain <-factor(info$strain)
strain <-relevel(strain, ref="C57BL6")
treatment <-factor(info$treatment)
treatment <-relevel(treatment, ref="N")
design = model.matrix(~strain*treatment)
rownames(design)<-colnames(oExpr)
design

#Model fitting 
fit<-lmFit(oExpr, design)
fit2<-eBayes(fit)

#Differential gene expression

out<-topTable(fit2,number=nrow(fit2), adjust.method="BH", p.value=1, sort.by="none")

What do they mean by running the analysis without a reference? You could re-parametrize the design matrix with no intercept, e.g. using a design of ~0+strain:treatment and then forming the appropriate contrasts, but you would still be doing the same analysis and would get exactly the same p-values and logFC values.

Hi Ryan,

Thanks. Treatment naive (not treatment) is fine but there is no particular genetic background that should serve as the baseline. 

Thanks,Theresa

 Thank you for the additional explanation. I have a much better understanding now. I will give your suggested design a try.