[BioC] limma - Considering Batch Effect

[Nathan Haigh]

I know this topic has arisen a few times from searching the mail
archive, but I'm still not clear on what/how to do this.

Here is my experimental setup:
Array    Param1   Param2   Rep   Batch
1        H        4        1     1
2        H        4        2     2
3        H        4        3     2
4        H        20       1     1
5        H        20       2     2
6        H        20       3     2
7        H        37       1     1
8        H        37       2     2
9        H        37       3     2
10       L        4        1     1
11       L        4        2     2
12       L        4        3     2
13       L        20       1     1
14       L        20       2     2
15       L        20       3     2
16       L        37       1     1
17       L        37       2     2
18       L        37       3     2

All rep 1's were processed in batch 1 and reps 2 and 3 were processes in
batch 2. From using GeneSpring, and conducting a cluster analysis on
experimental parameters, there is a clear batch effect. In GeneSpring, I
normalised the arrays, such that gene1 from array1 was divided by the
median of gene1 over arrays from batch1. Likewise, gene1 from array2 was
divided by the median of gene1 across arrays in batch2. These
successfully removed the batch effect. I'd like to be able to remove the
batch effect doing something similar, or accounting for it in limma.

In essence, I want to remove the variability in the reps caused by the
batch effect.

Could anyone explain how this can/should be done?
Thanks
Nathan