A multi-index ROC-based methodology for high throughput experiments in gene discovery

We address the problem of ranking differentially expressed genes in high throughput experiments using Receiver Operating Characteristic (ROC) curves. As it is generally unknown whether large expression values constitute 'positive' or 'negative' results or which group is 'healthy' or 'diseased', we generate four ROC curves per gene. We then consider classification indices based on all or part of the four ROC curves and identify genes ranked low by the area under the curve (AUC) but high by at least one alternative index, invariably resulting to the discovery of genes that would otherwise be missed by the AUC index.