Résumé
Driven by myriads of projects, public RNA-seq databases are exploding. To date, over 850,000 RNA-seq are deposited on SRA for human. This huge body of publicly available RNA-seq libraries is a precious resource to identify specific transcriptional events. The challenges lie in the complexity of RNA biological content and the exponential increase in data volume. We want to make RNA-seq data easily accessible providing a better capture of the whole transcriptome complexity, in the context of human health applications. New computational methods that perform indexing of k-mers across huge datasets constitute interesting solutions to interrogate "Omics data" at a large scale from dataset collections. Here, we developed TranSipedia, a new framework based on k-mer approach, constructed with several modules: 1/ The RNA-seq indexing step constructed with Reindeer (REad Index for abuNDancE quERy; Marchet et al., 2020), a novel computational method that serves as an efficient platform to request all transcribed information, 2/ a module to generate k-mers as signature of transcripts (Kmerator; Riquier et al, 2021), 3/ a supporting web site to facilitate the queries easily shared by biologists (TranSipedia, https://transipedia.montp.inserm.fr).Reindeer performs indexing of k-mers and records their counts across a large collection of datasets. Interestingly it associates k-mers to their counts instead of only recording the presence/absence of k-mers as frequently done in previous works. Moreover, Reindeer provides an ultra-fast performance in the query process while indexing several thousands of RNA-seq. One of the great advantages of indexing raw data is also that it integrates reference-free and annotation-free approaches. For applications where gene expression level is required, the k-mer count must be sufficiently sensitive and representative to be applicable. The quantitative accuracy with k-mers counts from Reindeer indexed datasets was compared to classical quantitative methods like Kallisto. Secondly, the design k-mer module uses Kmerator, a tool developed to construct specific k-mers, already avalailable on github. Thirdly, the website is also available to facilitate index queries by the biologists with sequences on fasta file format. The Transipedia platform now includes several thousands of datasets from public and private collections mainly from acute myeloid leukemia (AML) for cancer applications. We indexed the whole CCLE cohort representing 1019 RNA-seq samples for a total of 10 To and indexes from Leucegene, BEAT-AML, SRA and TCGA collections represented with more than 1000 RNA-seq samples for AML application. Concerning biological applications, we already requested in selected public datasets biomarker tissue specificity as well as tumor specific signatures comparing normal/tumor, for simple and useful medical usage. In perspectives, based on data structures such as k-mer features, diagnosis applications are in developpement. Moreover, Machine learning models could be used to search for signatures and explore better diagnostic and prognosis models.