Abstract
RNA sequencing enables a complete investigation covering the full dynamic spectrum of a tran- scriptome. It thus paves the way to a better understanding of the function of gene expression in different tissues, during development or pathological states. However, the splicing process, which generates both co-linear and non co-linear RNAs, the inclusion of sequencing errors, somatic muta- tions, polymorphisms, and rearrangements make the reads differ from the reference genome in a variety of ways. This complicates the task of comparing reads with a genome. Currently, the analysis paradigm consists in: 1. mapping the reads to a reference genome contiguously allowing as many differences as one expects to be necessary to accommodate sequence errors and small polymorphisms; 2. using uniquely mapped reads to determine covered genomic regions, either for computing a local coverage to predict mutations and filter out sequence errors (cf. program ERANGE), or for delimiting expressed exons approximately (cf. program TopHat); 3. re-aligning unmapped reads, which were not mapped contiguously at step one, to reveal splicing junctions. Limitations of this approach include lack of precision, redundant computations due to multi- mapping steps, error propagation due to heuristics and the absence of back-tracking. We propose a novel, integrated approach to analyze nowadays longer reads (> 50 bp). The idea is to adopt a k-mer approach that combines the genomic positions and local coverage to perform a complex analysis of each read and detect in a single step, mutations, indels, errors, as well as both normal and chimeric splice junctions. Comparisons with other tools demonstrate the feasibility of this ap- proach, which yields both sensitive and highly specific inferences.