Two-Step Mixed Model Approach to Analyzing Differential Alternative RNA Splicing

Abstract

18Changes in gene expression can correlate with poor disease outcomes in two ways: through changes in 19 relative transcript levels or through alternative RNA splicing leading to changes in relative abundance of 20 individual transcript isoforms. The objective of this research is to develop new statistical methods in 21 detecting and analyzing both differentially expressed and spliced isoforms, which appropriately account for 22 the dependence between isoforms and multiple testing corrections for the multi-dimensional structure of at 23 both the gene-and isoform-level. We developed a linear mixed effects model-based approach for analyzing 24 the complex alternative RNA splicing regulation patterns detected by whole-transcriptome RNA-25 sequencing technologies. This approach thoroughly characterizes and differentiates three types of genes 26 related to alternative RNA splicing events with distinct differential expression/splicing patterns. We applied 27 the concept of appropriately controlling for the gene-level overall false discovery rate (OFDR) in this multi-28 dimensional alternative RNA splicing analysis utilizing a two-step hierarchical hypothesis testing 29 framework. In the initial screening test we identify genes that have differentially expressed or spliced 30 isoforms; in the subsequent confirmatory testing stage we examine only the isoforms for genes that have 31 passed the screening tests. Comparisons with other methods through application to a whole transcriptome 32 RNA-Seq study of adenoid cystic carcinoma and extensive simulation studies have demonstrated the 33 advantages and improved performances of our method. Our proposed method appropriately controls the 34 gene-level OFDR, maintains statistical power, and is flexible to incorporate advanced experimental designs. 35