A unique requirement in biomarker is the balance of sensitivity (true positive rate) and specificity (true negative rate). Ideally, we would like a test to be of high sensitivity (i.e. able to detect all positive cases) and high specificity (i.e. able to correctly tell negative cases). However, this is often hard to attain and we need to decide a cutoff point with some trade-off.
Receiver Operating Characteristic (ROC) curve analysis is the method of choice for performance evaluation in biomarker analysis. It can be easily used for visualizing results obtained from different cutoff values for either univariate or multivariate approaches. Note in multivariate context, the cutoff points do not have very intuitive interpretations as in the classical univariate ROC analysis (see the Fig. below produced from MetaboAnalyst)
MetaboAnalyst offers, through a user-friendly web interface, classical univariate ROC curve analysis, and integration with several well-established algorithms (currently support: PLS-DA, Random Forests and SVM) to assist researchers in performing common ROC curve analysis for biomarker discovery and performance evaluation in metabolomics studies.