Goodness-of-Fit Tests and Diagnostics for Negative Binomial Regression of RNA-Seq Data

• Activity Number: 303
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Gu Mi*+ and Yanming Di and Daniel Schafer and Jeff Chang
• Companies: Oregon State University and Oregon State University and Oregon State University and Oregon State University
• Keywords: negative binomial regression ; dispersion models ; RNA-Seq ; model diagnostics

Abstract

This work is about assessing model adequacy for RNA-Seq datasets, particularly (1) assessing the adequacy of the negative binomial (NB) distribution, and (2) assessing the appropriateness of models for NB dispersion parameters. The tools for the first of these two aspects are appropriate for NB regression more generally; those for the second are specific to RNA-Seq data analysis. The statistical complexity resulted from small number of biological samples and large number of genes motivates us to address the trade-offs between robustness and statistical power using NB regression models. One power-saving strategy is to assume some commonalities of NB dispersions across genes via simple models relating the dispersion parameter to the (relative) mean. Many NB-based models with different assumptions on the mean-dispersion relationship have been proposed for testing differentially expressed genes. In this paper, we address how to evaluate model adequacy via simulation-based statistical tests and diagnostic graphics. We provide simulated and real data examples to illustrate that our proposed methods are effective for detecting deviations from the assumed NB model or the dispersion model.