Hi

I have just switched to analysing my data with a continuous instead of 2 group variable with limma. Is this correct with the addition of covariates in this manner?

design <- model.matrix(~des4$DAS28.0M+des4$AGE+des4$GENDER+des4$RACE)
matrix <- data.matrix(data3)
fit <- lmFit(matrix,design)
fit <- eBayes(fit)
top2 <- topTable(fit,coef=2,number=Inf,sort.by="P")

significant <- subset(top2, top2$adj.P.Val < 0.01)

Thanks,

Chris

It's hard to say if it's correct or not, as you don't really say what you are after.

This model does control for race and gender specific effects, as well as for continuous changes due to age. But you are implicitly assuming that these changes are simply a mean shift rather than directional. In other words, for a given gene you are assuming that males and females react the same to whatever DAS28.0M is, but perhaps that one sex has higher or lower intrinsic expression of that gene. So if you partition that difference into the gender coefficient, you in effect set the males and females to the same baseline expression, and you can then test for changes due to DAS28.0M.

But if the gene goes up in males but down in females when you treat with DAS28.0M, this model may not alert you to that fact. The same is true of race. And if the gene expression tends to go up with age in one race, but down with age in another race, well you get the point.

So yes, it's correct to specify a model in this manner, but whether or not it's correct for the experiment at hand is a different question altogether.