Thank you for the input, Fangxin, but I am not sure if a fixed effect for the lab factor can be modelled in my case, because each lab conducted different experiments, not replicates of the same experiment. Lab 1 did: "tissue1 treated" and "tissue1 untreated" and lab 2 did: "tissue2 treated" and "tissue2 untreated". So I don't see a way to distinguish between lab factor and tissue factor, right? Thanx a lot for any comments, Julia Fangxin Hong wrote: >>I wonder if I can compare Affymetrix arrays of the same type (ATH1) >>which were made in different laboratories and with different tissue >>types and different references. I have: "tissue1 treated", "tissue1 >>untreated" from one lab and "tissue2 treated", "tissue2 untreated" from >>the other lab. >>The references (untreated) are different because of the different >>tissue types. I am interested in the difference between tissue1 treated >>and tissue2 treated, so I thought I could use limma to make a contrast: >>(tissue1_treated-tissue1_untreated)-(tissue2_treated- tissue2_untreated). >>I am not sure if this is valid, though? For example, I do not account >>for the different labs that way. >>Maybe it is just possible to analyse each experiment by itself and >>compare the results at a latter stage, say compare lists of >>differentially expressed genes? >> >> >Based on what I observed when study data generated at different lab, lab >effect can't not be completely removed by normalization step. If you do >have some replicates or several data sets from each lab, and you want to >combine data together, I would suggest you to inlcude a fixed effect for >lab factor. >Hopefully this will help. > >Fangxin > > > >