Figure 23

Comparison of performance of three single-slide methods on simulated data. (a) ROC curves for simulated data for each method. Each point represents one of the ten cut-offs used in Table 4 for each method. Area under curve is one in the case of an ideal method. ROC curves do not take prevalence into account and upper curves have better accuracy than lower ones (see text for details). (b) PPV curves which represent the probabilities that a gene identified as differentially expressed represents a true positive. (c) Likelihood ratio (or Bayes' factor) curves are calculated as Sensitivity/(1-Specificity) and can also be defined as the ratio of posterior odds to prior odds. PPV values takes into account prevalence of being differentially expressed in the simulated population.