For a microarray experiment, there are 4 biological replicates for several tissue types, ages (old and young, not numerical), and treatments (a factored experiment). I was originally doing differential expression analysis without SVA in limma, using a group means parameterization design matrix for each combination of tissue, age, and treatment created with model.matrix(~0+group) where each group is a unique combination of tissue age and treatment, and using a contrast matrix for group comparisons of interest with each contrast being something like groupA - groupB.

However, after reading the vignette for SVA, the matricies that the vignette recommends to create is something like mod = model.matrix(~tissue+age+treatment) which I understand will allow lmFit to fit coeficients and an intercept to each group. But this of course is completely different than the current matrix that estimates a mean coefficient for each unique group. How would I go about trying to do something similar for the SVA design matrix for lmFit so that the output TopTables would be for the exact same groupA - groupB comparisons? If this isn't possible, how would I go about replicating this effect? Would any other type of design matrix work for SVA?