I have a question about one or multiclass SAM for timecourse arrays. I have Affy arrays with >10k genes represented. Experimental design is for 27 samples (9 timepoints, 3 replicates each). When data is loaded to MeV for One Class SAM analysis, a huge number of significant genes (~4k) are identified. This is more genes than is practical to process by clustering methods. I wonder if we are using the correct type of SAM analysis. Is there a better way to use SAM for timecourse analysis? For example, would grouping samples into 9 groups of 3
replicates each be beneficial, and help reduce the significant hits?
Thanks for any advice.
Most likely, you do not want to be running One Class SAM on your affy data. One class is typically used for two-color arrays or data with a meaningful mean value. You might try running multi-class SAM with 9 groups and see how that turns out.
Additionally, MeV offers a new statistical method tailored specifically to time course data, BETR (Bayesian Estimation of Temporal Regulation). It sounds like you might be interested in this module for "One-condition". MeV will also release version 4.5 on November 13th, which will include the widely used LIMMA method- another commonly used time course analysis.