Statistical Evaluation of Quality Performance on Genomic Composite Biomarker Classifiers

Abstract

After completion of the Human Genome Project, genomic composite biomarker classifiers (GCBCs) became available. However, quality performance of GCBCs varies. We propose statistical methods for evaluation of the quality performance of GCBCs on selection of differentially expressed genes, agreement and reproducibility. Methods: For detection of differentially expressed genes, an interval hypothesis was employed to take into account both biological and statistical significance. The concordance correlation coefficient (CCC) was used to evaluate the agreement of expression levels of technical replicates. The intraclass correlation coefficient (ICC) was suggested to assess the reproducibility between laboratories. Results: A two one-sided test procedure was proposed to test the interval hypothesis. Statistical methods based on the generalized pivotal quantities for CCC and ICC were suggested to test the hypotheses for agreement and reproducibility. Simulation results demonstrated that all three methods could adequately control the type I error rate at the nominal level for assessment of differentially expressed genes, agreement and reproducibility. Conclusion: Three appropriate statistical methods were developed for evaluation of quality performance on differentially expressed genes, agreement and reproducibility of GCBCs.

Keywords

• agreement
• differentially expressed genes
• genomic composite biomarker classifier
• reproducibility of results