Simultaneous Inference of Gene Isoform Expression for RNA Sequencing Data

Abstract

In this article, we describe simultaneous inferential methods in detecting differentially expressed gene isoforms based on the Poisson generalized linear models. We derive the joint asymptotic distribution of pivotal quantities. The sample size of RNA sequencing data is often small in practice. Using multiple comparison procedures based on large-sample approximation becomes problematic. The parametric bootstrap method based on pivotal quantities is outlined as a robust alternative. Moreover, we observe the validity of robustness of the bootstrap method when mild overdispersion presents in RNA-sequencing data. We demonstrate the validity of the proposed method in detecting differentially expressed isoforms through Monte Carlo simulation. It shows the proposed method controls the family-wise error rate for large-scale inference. Even though the proposed method can be extended to many experimental designs, we focus on factorial designs in this article.