How to deal with undetected probes in microarray data?

Question

[Kristin Muench]

I'm working with GSE28551_RAW. In this dataset, each of the 37 samples has a flag which seems to indicate whether the program calls the probe detected (flag=0) or non-detected (flag>0).

Currently, my first filtering step is to throw out any probes when any of the 37 samples has a "not detected" flag. However, this is quite stringent - I go from having 37,842 probes to analyze, to 9,102.

Is this the appropriate way to filter out undetected probes? If not, what do I do if the probe was not detected in some samples and detected in others? Should I include a probe in my final analysis if more than 50% of the samples detected the probe? I'm not sure how I would do the analysis if I include the data for that probe in some samples and not for other samples...

Thank you for your help!

Answer 1

[Jean-Karim Heriche]

The fact that a probe gives different signals (including below detection level) in different samples is the basis for differential expression analysis. If a probe that could have been detected is not detected in a sample, this could be an indication that the corresponding transcript is not expressed in that sample. What you do with those depends on the question you want to address. As a first filtering step, I would remap the probes to a current genome annotation and filter out obsolete probes which in principle never get a chance to give a signal other than noise. This post may get you started with microarray data analysis.