Fast and interpretable alternative splicing and differential gene-level expression analysis using transcriptome segmentation with Yanagi

Abstract

AbstractIntroduction:Analysis of differential alternative splicing from RNA-seq data is complicated by the fact that many RNA-seq reads map to multiple transcripts, besides, the annotated transcripts are often a small subset of the possible transcripts of a gene. Here we describe Yanagi, a tool for segmenting transcriptome to create a library of maximal L-disjoint segments from a complete transcriptome annotation. That segment library preserves all transcriptome substrings of length L and transcripts structural relationships while eliminating unnecessary sequence duplications.Contributions:In this paper, we formalize the concept of transcriptome segmentation and propose an efficient algorithm for generating segment libraries based on a length parameter dependent on specific RNA-Seq library construction. The resulting segment sequences can be used with pseudo-alignment tools to quantify expression at the segment level. We characterize the segment libraries for the reference transcriptomes of Drosophila melanogaster and Homo sapiens and provide gene-level visualization of the segments for better interpretability. Then we demonstrate the use of segments-level quantification into gene expression and alternative splicing analysis. The notion of transcript segmentation as introduced here and implemented in Yanagi opens the door for the application of lightweight, ultra-fast pseudo-alignment algorithms in a wide variety of RNA-seq analyses.Conclusion:Using segment library rather than the standard transcriptome succeeds in significantly reducing ambigious alignments where reads are multimapped to several sequences in the reference. That allowed avoiding the quantification step required by standard kmer-based pipelines for gene expression analysis. Moreover, using segment counts as statistics for alternative splicing analysis enables achieving comparable performance to counting-based approaches (e.g. rMATS) while rather using fast and lighthweight pseudo alignment.