Year: 2024
Type: dissertation
Abstract: Single-cell RNA sequencing promises to revolutionize our understanding of tissues at the molecular level. Single-cell transcriptomics should allow us to identify the various cell types of a tissue by their transcriptomes, as well as indicating how they coordinate to carry out tissue function by the ligands and receptors they present. To fulfill the promise of molecular models of tissue function from scRNAseq data, we must be able to: collect single-cell transcriptomes at scale; identify all contributing cells of a tissue from their transcriptome; and predict how these cell types are communicating based on their expressed ligands and receptors. In this work I address some of the challenges that have arisen in our efforts to realize this potential of scRNAseq technology. The software tool scClustViz is my contribution to the workflow required to cluster single-cell sequencing libraries into cell type transcriptomes. It proposes a novel method to determine appropriate cluster resolution by differential gene expression and fixes a common error in differential expression magnitude due to pseudocounts. Its web-based interactive interface simplifies both the annotation of clusters and the sharing of results. A systematic review of methods using scRNAseq to predict intercellular communication between cell types revealed that most, including my CCInx tool, infer communication solely based on ligand and cognate receptor expression. Specificity of these predictions may be improved by using evidence of ligand perturbation in the receptor cell transcriptome, as implemented by a handful of methods. These methods commonly expect that receptor activation yields a characteristic downstream gene expression signature. I test that assumption using both high-throughput and individual ligand perturbation assays and show that transcriptional response to ligand perturbation is cell type specific. If transcriptional response to ligand interaction can be inferred per cell type, this barrier to improving ligand-receptor inference methods can be overcome.
Source: TSpace (University of Toronto)
Author: Innes, Brendan T
Institution: University of Toronto
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7132960089