Exhaustive capture of biological variation in RNA-seq data through k-mer decomposition

Each individual cell produces its own set of transcripts, which is the combined result of genetic variation, transcription regulation and post-transcriptional processing. Due to this combinatorial nature, obtaining the exhaustive set of full-length transcripts for a given species is a never-ending endeavor. Yet, each RNA deep sequencing experiment produces a variety of transcripts that depart from the reference transcriptome and should be properly identified. To address this challenge, we introduce a k-mer-based software protocol for capturing local RNA variation from a set of standard RNA-seq libraries, independently of a reference genome or transcriptome. Our software, called DE-kupl, analyzes k-mer contents and detects k-mers with differential abundance directly from the raw data files, prior to assembly or mapping. This enables to retrieve the virtually complete set of unannotated variation lying in an RNA-seq dataset. This variation is subsequently assigned to biological events such as differential lincRNAs, antisense RNAs, splice and polyadenylation variants, introns, expressed repeats, and SNV-harboring or exogenous RNA. We applied DE-kupl to public RNA-seq datasets, including an Epythelial-Mensenchymal Transition model and different human tissues. DE-kupl identified abundant novel events and showed excellent reproducibility when applied to independent deep sequencing experiments. DE-kupl is a new paradigm for analyzing differential RNA-seq data with no preconception on target events, which can also provide fresh insights into existing RNA-seq repositories.