Robust Concordance Index: a Metric for Association Testing for Preclinical Biomarker Discovery
Ian Smith,Petr Smirnov,Benjamin Haibe-Kains University Health Network, University of Toronto
Abstract
Datasets in biological research often suffer from the dual problems of considerable systematic and statistical noise and limited sample size. This presents challenges in identifying associations between biological features and response variables of interest, as in the context of identifying biomarkers for sensitivity to therapy. We introduce the Robust Concordance Index (rCI), a modification to the standard Concordance Index (or Kendall’s Tau) to address these limitations. The package wCI incorporates efficient implementations of rCI and adaptive permutation testing for assessing statistical significance. In this work, we compare rCI with other statistical metrics for association testing and evaluate the statistical power of these metrics under a variety of assumptions. We offer analysis on under what conditions each metric is likely to be most powerful. Finally, we evaluate these metrics on pharmacogenomic datasets for biomarker discovery.
Keywords: Pharmacogenomics,statistics,biomarker discovery