PUB - Publikationen an der Universität Bielefeld

The advent of next-generation sequencing technologies and their adoption to RNA sequencing (RNA-seq) has revolutionized the field of transcriptomics. RNA-seq approaches revealed an unexpected complexity of prokaryotic transcriptomes and in particular led to the discovery of a wealth of previously unknown non-coding RNAs (ncRNAs) in diverse bacterial species. Primary applications of RNA-seq in the context of ncRNAs are a) discovery of novel and (re-)annotation of known ncRNAs, b) determination of transcription start sites (TSS), and c) identification of transcripts that are associated with RNA binding proteins, such as the RNA chaperone Hfq. Given the immense amounts of data obtained from RNA-seq approaches bioinformatics methods are crucial for their analysis and interpretation. Moreover, the constant change of design and scope as well as of protocols and applications of RNA-seq experiments requires the development and adaption of computational methods.

The present thesis focuses on the development of biocomputational methods that enable the discovery and characterization of ncRNAs in prokaryotes. These methods include: First, methods for processing differential RNA sequencing data that aid in the reconstruction and classification of ncRNA transcripts. Second, methods that allow for the precise determination of transcription start sites in sequencing data obtained from RNA-seq approaches enriched with primary transcripts. Besides, methods were developed, which exploit TSS information for the prediction of promoter sequences. Third, methods for the identification of Hfq-bound transcripts in sequencing data generated by Hfq co-immunoprecipitation experiments. Fourth, methods for building RNA families. RNA-seq approaches predict hundreds of unannotated ncRNAs but do not provide much information about their biological function. Studying conservation patterns and phylogenetic distribution of ncRNAs by means of RNA family models aids in the functional characterization of ncRNAs. Building RNA family models is neither standardized nor automated. In this work a systematic construction strategy starting from single ncRNAs was designed and implemented for covariance models, the de facto standard for modeling structural RNA families. Fifth, an integrative model of the structurally varying "cuckoo" RNA family by means of thermodynamic matchers was devised and the model was used for systematic homology search in a wide spectrum of bacterial species.

The computational methods that were developed as part of this thesis were applied in several studies on the transcriptome of the nitrogen fixing alphaproteobacterium *Sinorhizobium meliloti*. This resulted in the unprecedented characterization of the transcriptomic landscape of *S. meliloti* and provided deep insights into the presence and organization of ncRNAs.