Is Variance Stabilization Neccessary When Using Sam?

Question

[Sebastian]

I'm working on some miRNA microarray data and I'm wondering if I have to do variance stabilization before I use SAM (significance analysis of microarrays)? If I understand the SAM method correctly it already computes a term to minimize the variance of intensities.

Answer 1

[David Quigley]

SAM does not minimize the variance per se. The SAM statistic d for two-class unpaired data is (difference in means) / (standard deviation + S0 ), a modified T statistic. The constant term S0 in the denominator reduces the score d of genes where the difference in means between the two classes is very small. Such genes are usually expressed at very low levels. The signal-to-noise ratio of genes expressed at low levels is very low, so the practical consequence is that you avoid calling them significantly differentially expressed. If you compare SAM results to a raw T statistic you can get a feel for what the difference is.

See their paper (Tusher PNAS 2001) and the SAM manual for gory details on how S0 is calculated.

Answer 2

[toni]

I am not really used to analyze miRNA microarrays but variance stabilization is about normalizing your dataset whereas SAM methodology is about doing a supervised analysis between two classes of interest. You can do a variance stabilization before a SAM analysis unless you already normalized your data with another method (like RMA for RNA microarrays).

This is a question you should better ask to the BioConductor community through the BioC mailing list.

best,

tony