Unsupervised and Supervised Methods for Analysis of Human Multimodal Transcriptome Data

Studying the RNAs expressed in cells has enabled the discovery of different types of RNA with different functions. Many RNAs, however, have no associated function. Therefore, we need to develop novel methods to facilitate and accelerate the elucidation of these RNAs. In this thesis, I developed two machine learning methods allowing summarization and visualization of large collections of transcriptomic datasets. I also show that these methods allow generating biological hypotheses for known RNAs with unknown functions. I describe these methods in three chapters. Chapter 2 describes our SegRNA method for unsupervised annotation of multi-experiment transcriptome datasets. SegRNA annotations enabled the visualization of the most common patterns of transcriptomic signal from multiple assays: GRO-seq, CAGE, and RNA-seq. The most common patterns not only indicated the differences between these assays, but they also identified RNAs with different sizes and from different subcellular localization. SegRNA annotations are cell-type–specific. To date, the annotation that we generated for the K562 cell line is the first cell-type annotation that integrates RNA sizes, localization, and different assays. We show in this chapter that this annotation allows identifying novel short RNAs. Chapter 3 describes my collaboration with the FANTOM6 consortium. I show the experiments that led to the decision of using SegRNA with manual parameters to annotate the transcriptional changes that follow the knockdown of long non-coding RNAs, at 2 bp resolution. We used our method to generate 99 annotations that summarize 510 CAGE datasets. These SegRNA annotations successfully identified 92% of the knocked down long non-coding RNAs as downregulated, while SegRNA annotated on average 7% of the genes as downregulated. Chapter 4 describes our supervised cell lineage analysis (CLA) method for the identification of cell-type–specific, and lineage-specific RNAs. Identifying these RNAs requires running tools that generate a large number of results which makes the results hard to interpret. CLA automates these steps by identifying RNAs that discriminate cell types and summarizing the RNAs with a lineage score. We show that this approach enabled the identification of novel cell-type–specific alternative promoters. This thesis introduces tools that allow researchers to better understand the transcriptome by facilitating its interpretation.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.