Figure 3

Receiver Operating Characteristic curves depicting the performance of the methods in simulation. For each of the six methods compared in the simulation experiment, the curves plot the percentage of the (in total) 400,000 simulated differentially expressed (d.e.) genes that would be returned in gene-lists of varying length (the sensitivity) against the percentage of the 1,857,500 simulated non-differentially expressed genes that are correctly not-returned in the same gene list (the specificity). Two trivial cases are the genelist of length zero (none of the d.e. genes are correctly returned, none of the non-d.e. genes are incorrectly returned = 0% sensitivity and 100% specificity) and that of length 22,575 (100% sensitivity and 0% specificity). The diagonal line indicates the performance that could be acheived by random classification. As a general rule, the nearer to the top left corner the curve passes, the better the method.