Figure 1

ROC curve for the simulation study. The plot shows the ROC curve for the Bayesian models averaged over the 10 simulated dataset: in each case we simulated 200 differentially expressed and 800 not differentially expressed genes. We have implemented the models either combining the five pre-processing methods together (solid line) or analysing each one separately and ranked the tail posterior probability of differential expression. The ROC curve for the combined model is above that of each pre-processing method, highlighting the benefit of using a combined model in terms of specificity and sensitivity.