Computation and Selection of Optimal Biomarker Combinations by Integrative ROC Analysis Using CombiROC

Abstract

The diagnostic accuracy of biomarker-based approaches can be considerably improved by combining multiple markers. A biomarker’s capacity to identify specific subjects is usually assessed using receiver operating characteristic (ROC) curves. Multimarker signatures are complicated to select as data signatures must be integrated using sophisticated statistical methods. CombiROC, developed as a user-friendly web tool, helps researchers to accurately determine optimal combinations of markers identified by a range of omics methods. With CombiROC, data of different types, such as proteomics and transcriptomics, can be analyzed using Sensitivity/Specificity filters: the number of candidate marker panels arising from combinatorial analysis is easily optimized bypassing limitations imposed by the nature of different experimental approaches. Users have full control over initial selection stringency, then CombiROC computes sensitivity and specificity for all marker combinations, determines performance for the best combinations, and produces ROC curves for automatic comparisons. All steps can be visualized in a graphic interface. CombiROC is designed without hard-coded thresholds, to allow customized fitting of each specific dataset: this approach dramatically reduces computational burden and false-negative rates compared to fixed thresholds. CombiROC can be accessed at www.combiroc.eu.