Hi, I am trying to set up the correct formula for an analysis using the package limma. I have read the user guide, but I am still not sure if I am approaching this the right way. Thanks in advance for your help. I have microarray data for six patients. Four responded to treatment and two did not. For each patient I have a sample from before treatment and a sample from after treatment. Thus, twelve arrays total. I have two objectives: 1) Identify genes differentially expressed between responders and nonresponders before treatment 2) Identify genes that change (from before treatment to after treatment) differentially between responders and nonresponders For objective one, my initial strategy was to: 1. ignore the post-treatment data 2. use the formula "~ isresponder" 3. and run toptable for isresponder But then I wondered whether I was throwing information away about variance by ignoring the post-treatment data. Therefore I tried a new strategy: 1. keep all data 2. create a factor variable identifying my four types of patients 3. use the formula "~ 0 + responderpre + responderpost + nonresponderpre + nonresponderpost" 4. and run toptable for the contrast responderpost-responderpre But then I wondered whether it was a problem that I was not accounting for the fact that there were pairs of samples in the analysis. I tried a more complicated modeling strategy, but on running eBayes I got the error message "No residual degrees of freedom in linear model fits." Does anyone have suggestions? For objective two, I am using the approach: 1. use all data 2. use the formula "~ isresponder*ispost" 3. run toptable on isresponder:ispost Again, this ignores the fact that observations are paired. I tried augmenting the formula to "~isresponder*ispost + patient_id", but this returns "Coefficients not estimable." What is the appropriate modeling strategy here? Thanks, Mike [[alternative HTML version deleted]]