Résumé
With its ability to reveal both altered gene expression levels and the production of aberrant transcripts, RNAseq is popular in the field of precision medicine. An increasing number of clinical trials uses this technology in order to discover functionally relevant alterations. Driven by myriads of projects, public RNAseq databases are exploding, to date, there is over 164,000 RNA-seq on SRA for human. This huge body of publicly available RNAseq libraries is a precious resource to identify specific transcriptional events. However, the challenges lie in the complexity of RNA biological content and the exponential increase in data volume. We want to make RNAseq data easily accessible, providing a capture of the whole transcriptome complexity, in the context of human health applications. Therefore, we developed a new framework based on a k-mer approach, constructed with several modules: 1/ a new RNAseq indexing structure that will serve as an efficient platform to request any transcribed information, 2/ a complete module to generate unique k-mers as signature of transcripts, 3/ a supporting web site to facilitate the queries for the biologists.The indexing step uses Reindeer, a new k-mer based indexation structure. To our knowledge, it's the first method capable of performing fast mapping-free quantification of variant transcripts in thousands of RNAseq libraries [1]. The methodology is already efficiently implemented for several biological applications based on public datasets (from ten to thousand of RNAseq corresponding to 100Go to 10To of data). The k-mer designing module uses Kmerator, a tool developed to extract specific k-mers (https://github.com/Transipedia/kmerator) [2]. Finally, the web application is already available to facilitate large RNAseq datasets queries by the biologists with their sequences of interest as input (fasta format).Concerning medical applications, we already requested and identified in selected public datasets, genes co-expressions, tissue specific biomarkers, as well as tumor specific signatures comparing normal and tumoral samples. As an example, we recovered known translocations and mutations in RNAseq Acute Myeloid Leukemia (AML) samples and identified new specific biomarkers (long non-coding RNAs…).With the addition of advanced Machine Learning approaches, our framework could be used to select the best signatures and to improve diagnosis and prognosis models in cancers.References:[1] Marchet C et al. REINDEER: efficient indexing of k-mer presence and abundance in sequencing datasets.Bioinformatics (2020)[2] Riquier S*, Bessiere C*, Guibert B, Bouge AL, Boureux A, Ruffle F, Audoux J, Gilbert N, Xue H, Gautheret D,Commes T. Kmerator Suite: design of specific k-mer signatures and automatic metadata discovery in large RNA-Seq datasets. Nucleic Acids Research (2021)