I've got a small question regarding limma and couldn't find a direct answer anywhere: Does it make sense to perform limma on only a subset of genes?
An example: I am only interested in if there are differences in gene expression between the groups of interest in 700 previously defined genes that had been identified (by limma) in another independent experiment (using a different experimental technology). So one could say it would make sense to test only the 700 genes in order to reduce the multiple testing burden. But I was wondering if this would make
sense, as limma (eBayes) borrows information across all genes. There are about 50 biological replicates per group available.
Would you recommend to 'feed' limma only with the 700 genes of interest, or to perform the analysis on all genes and do the filtering afterwards?