I'm using R in order to get expression values from raw microarray experimental data (downloaded from GEO). After I obtained the expression values I'd like to know which probes I should remove. Looking around I saw people using the method nsFilter. I need to analyze the expression values using a data mining approach (I'm not a biologist). Is it enough remove all the control probes? Should I keep all the other probes even the ones with low variance? What is a typical approach?
•
written
by
ncl.lazzarini
At the end of the day it depends completely on your goals (N.B., saying "...a data mining approach" is rather like saying that you'll go from point A to point B with a "movement approach"). In any case, you'll likely want to get rid of probes lacking significantly higher signal than that seen in control probes, as these likely represent background noise.
How can I get rid of those probes? Is it enough by applying the GCRMA normalization?
I don't recall that GC-RMA actually removes those probes, though it makes nice use of them in subtracting background signal. The control probes usually have names beginning with AFFX, so it's pretty easy to determine which ones they are. BTW, you might consider fRMA, in case you start adding additional samples from GEO.
Log in to answer this question.
