B P - 2010-09-21

Dear friends,

Another question about LIMMA. I performed experiments with Illumina platform and there are around 45000 probes for the whole genome. I want to compare a drug treatment.  When I used Limma module, and iinput normalized results from all the probes, I could not get any significant genes out. But  if I only used the genes with significant detection values (Then I ended up around 18000 genes), I got several hundreds significant genes. I wonder which result I should trust.  I do not know much about the math behind LIMMA. But is that because the genes with non significant detections ( so with lower and more variable signals) will affect the multiple testing dramaticly?

Thanks a lot!