An Integrated Hierarchical Bayesian Approach to Normalizing Left-Censored miRNA Data

• Activity Number: 33
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Jia Kang*+
• Companies: Merck
• Keywords: miRNA ; normalization ; hierarchical Bayesian modeling ; level of detection ; variable selection

Abstract

miRNAs are small non-coding RNAs that suppress gene expression and are responsible for regulating >60% of the human coding genome. A critical pre-processing procedure for detecting differentially expressed miRNAs is normalization, aiming at removing the between-array systematic bias.

Most normalization methods adopted for miRNA data are the same methods used to normalize mRNA data; but miRNA are very different from mRNA data because of possibly larger proportion of differentially expressed miRNA probes, and much larger percentage of censored miRNA probes below detection limit.

To address the unique characteristics of miRNA data, we present a hierarchical Bayesian approach that integrates normalization, missing data (due to LOD) imputation, and hit selection in the same model. We generated simulation data to compare the performance of Bayesian normalization vs. quantile normalization vs. no normalization in detecting truly differentially expressed miRNAs. Detection performance is evaluated by sensitivity, specificity, and AUC. The simulation results clearly demonstrate the necessity of normalization ,and the superiority of Bayesian normalization over quantile normalization.