Significance analysis of microarrays using rank scores

The Significance Analysis of Microarrays (SAM) software is a very practical tool for detecting significantly expressed genes and controlling the proportion of falsely detected genes, the False Discovery Rate (FDR). However, SAM tends to find biased estimates of the FDR. We show that the same method with the data replaced by rank scores does not have this tendency. We discuss the choice of the rank score function in view of the power of this nonparametric multiple testing procedure. Moreover, we introduce a testing formalization of the popular 2-fold rule. This testing procedure is more selective than the basic procedure and it enables the scientist to make a stronger statement about the selected genes than with the 2-fold rule. All procedures are illustrated with the example one-class data available in the SAM software.