In fact, it is quite a statistical question rather than biological question. After we compare the expression data and get the p value for different genes(iso forms, tss, or any other groups of interest), people sometimes need to control FDR(false discovery rate). Here is the method used in cufflinks to achieve this goal (a wiki page, not from the cufflinks paper):

I understand the steps used here, but I don't know how can the software come up with a q value based on this procedure.