A Statistical Method for Detecting Differentially Expressed SNVs Based on Next-Generation RNA-Seq Data

Abstract

Summary In this paper, we propose a new statistical method –MutRSeq– for detecting differentially expressed single nucleotide variants (SNVs) based on RNA-seq data. Specifically, we focus on non-synonymous mutations and employ a hierarchical likelihood approach to jointly model observed mutation events as well as read count measurements from RNA-seq experiments. We then introduce a likelihood ratio based test statistic, which detects changes not only in overall expression levels, but also in allele specific expression patterns. In addition, this method can jointly test multiple mutations in one gene/pathway. The simulation studies suggest that the proposed method achieves better power than a few competitors under a range of different settings. In the end, we apply this method to a breast cancer data set and identify genes with non-synonymous mutations differentially expressed between the triple negative breast cancer tumors and other subtypes of breast cancer tumors.

Zero-Inflated Poisson Models with Measurement Error in the Response

“…Count data with excess zeros arise frequently from genomic studies (eg, Fu et al. , 2017), and those excess zeros may be classified as “ structural zeros ” and “ sampling zeros ”. The term “ structural zeros ” refers to the zero counts for those individuals who are not “at‐risk” for the event, while “ sampling zeros ” refers to the outcomes for those “at‐risk” individuals who happen to be measured with zero counts.…”