FDR calculated from the Spira et al. data. The dotted line indicates where the optimal cut-off is estimated to be in the sense of minimising the sum of false positives and false negatives. At this point the FDR of BH, pava and qvalue is close to 0.4, and corresponds to the 298th smallest p-value. This high proportion of false positives may be acceptable given some high throughput validation procedures, but may be unacceptable in other contexts. Note that BUM produces the overly pessimistic estimate FDR ≡ 1, and hence appear at the upper limit of the graph. For comparison the local FDR  has been added to the graph.