.05 for p and -11 may be the best option. No specificity as such.

I like Jochen's approach but I would first sort all genes based on log2FC and then pick the top 50 or 100 which have significant Bonferroni-corrected P-values (FDR is less stringent). The idea behind this is that you are looking for substantial changes in expression between cases and controls but you also want to make sure the variation within each group is minimal and does not overlap, hence a statistical test (t-test or Mann-Whitney) comes into play.

Hope it helps!

Thank you everyone for sharing and discussing different perspectives.

There is no specificity in selecting the cut-off values and they are subjective.

Computing delta values on the basis of thresholds will be meaningful? and can we associate the same delta value as a cut-off for fold change too?

like, delta.table

The list of genes identified as DE by whatever statistical approach is just the beginning to understanding the data from the experiment. The next step generally used is to take the DE genes and identify the over/under represented gene ontology (GO) terms among these genes. Unless the perturbation is strong and/or you have many samples, in practice you need to use fairly relaxed thresholds of DE, e.g. FDR = 0.05  to get enough genes with associated GO terms to find anything useful in terms of developing new hypotheses. It helps too if your study is designed with additional controls to understand which gene(s) are directly relevant to the question. For example, if you are studying the effect of a drug, microarray data could show hundreds of DE genes. If you have microarray data from other drug treatments, that would help you eliminate the uninteresting DE genes and GO terms  common to all treatments. 

Hi all,

Did the PM(perfect match),MM(mismatch) matrix will be helpful for the analysis for mapping with the expression matrix?

If yes, how we can create a PM-MM matrix of an affymetrix microarray data?(as my data is from GEO database)

Is it from the GDS or GSE files?