Three Reasons to Distrust Microarray Results
December 10th, 2008
…the paper actually demonstrated that is it possible to distinguish microarray experiments conducted on one day from experiments conducted another day. That is, batch effects from the lab were much larger than differences between patients who did and did not respond to therapy… As is so often the case, data were mislabeled. In fact, 3/4 of the samples were mislabeled.
It reminds me of a book of scientific humour that I read as undergrad (but originally published in the USA in the 60s), titled “Physics joke”, or something like this. Especially last two sentences. Not much has changed since then, eh?
This is hardly a new observation. RNA-seq may eventually provide a means of performing expression measurements with better quantified sources of bias, however it remains in its infancy. Until then, I think it’s safe to say that we’re stuck with microarrays, and that their effective use is going to remain something of a black art.
Microarrays can, and do, produce useful results, particularly in the case where experiments are designed with competent statisticians beforehand to minimise non-biological influences on the measurements, and where they’re analyzed by competent statisticians afterwards.
It’s disconcerting how inconsistent microarray experiments are – not just between people or between experiments but between platforms as well. A colleague just presented some work to our lab on a new data set of human tissue samples run on 3 different platforms and showed there was essentially no correlation (of ranked differentially expressed genes) between any of the platforms. Crazy.