Biomarker Categorization in Transcriptome Meta-analysis by Statistical significance, Biological Significance and Concordance

Abstract

With the advancement of high-throughput technology, transcriptomic studies have been accumulated in the public domain. Meta-analysis combines multiple studies on a related hypothesis and improves the statistical power and reproducibility of single studies. However, a majority of existing meta-analysis methods only consider the statistical significance. We propose a novel method to categorize biomarkers by simultaneously considering statistical significance, biological significance (large effect size), and concordance patterns across studies, accounting for the complex study heterogeneity that exists in most meta-analysis problems. We conducted simulation studies and applied our method to Gynecologic and breast cancer RNA-seq data from The Cancer Genome Atlas to show its strength as compared to adaptively-weighted Fisher’s method. We found several major biomarker categories according to their cross-study patterns, and these categories are enriched in very different sets of pathways, offering different biological functions for future precision medicine.