That is the wrong question. The question should be: what is the minimum number of samples I need to have adequate power to detect some effect that will replicate, i.e., contribute to generalizability and be useful to science. 

Adequate power will depend on whether the "populations" (usually classes rather than populations) actually differ for the relevant traits being measured. The degree by which they differ will determine how many samples are necessary.

That is the wrong question. The question should be: what is the minimum number of samples I need to have adequate power to detect some effect that will replicate, i.e., contribute to generalizability and be useful to science. 

Adequate power will depend on whether the "populations" (usually classes rather than populations) actually differ for the relevant traits being measured. The degree by which they differ will determine how many samples are necessary.

For performing microarray expression analysis you just require only two samples (1 control + 1 traeated). But for advanced work and publication you may concentrates on 2 or more datasets which contains more samples (dataset 1: 5 control+ 5 treated; dataset 2: 5 control+ 5 treated). I hope you understood)

http://link.springer.com/chapter/10.1007/978-981-287-260-9_9

I agree with Gerard - there is no magic number.  You need to know something about this particular cancer, and have some knowledge of the populations A and B, and then sit down with a statistician and determine an optimal number of biological replicates to adequately sample those populations for that cancer.

A N of 1 for a population based genomics experiment is never acceptable, in my opinion.  You cannot claim to have sampled adequately to make discrimination's at the population level without actually sampling both populations.  That means you need an absolutely bare minimum of 3 biological replicates to be able to compute a mean and a variance about that mean in order to discern differences between the populations.  Ideally, your optimal N will be more than that bare minimum.

(for the animal based toxicology studies I work on, we like to have typically between 5 to 10 biological replicates, although cost does play into the final number for any given experiment)

It dramatically depends on variance in your populations.

there is an easy to use calculator  in free software BRB Array Tools.

http://linus.nci.nih.gov/BRB-ArrayTools.html

Thanks everyone for your clear answers.

In my study I compared 7 male breast cancer VS 17 female breast cancer. I did ANOVA and I selected differentially expressed genes based on fold-change and FDR. Anything else I should do?