Author: Anjun Ma (Ph. D. in biomedical informatics)

Publisher:

Published: 2020

Total Pages: 0

ISBN-13:

Unraveling regulatory relations between transcription factors (TFs) and target genes can help us understand the underline regulatory mechanisms of cell functions and pinpoint crucial factors that determine phenotype. The rapid development of single-cell RNA-Sequencing (scRNA-Seq) technology provides a tremendous opportunity to study cellular heterogeneity, characterize dynamic cell differentiations, and identify specific regulation patterns programmed in a cell population. However, limited by the high-level dropout issue and heterogeneity of scRNA-Seq data, it is computationally challenging to directly predict the overall gene regulatory network at the single-cell level. Instead, we can disassemble a global regulatory network into regulons, each of which is a group of target genes regulated by the same TF, to elucidate the signature regulatory patterns in cells. Regulons that are uniquely active in a cell type are named cell-type-specific regulons (CTSRs). Identify CTSRs can substantially help to determine differential key TFs and downstream-regulated genes that determine the cell types. However, a limited number of tools in the public domain can do this job. To fill this gap, my project aims to develop new algorithms and tools to identify CTSRs from scRNA-Seq data. The project is divided into three sections. First of all, a new biclustering model was designed to identify local functional co-expressed gene modules based on a left-truncated mixture Gaussian model. Secondly, in virtue of the advantage of graph neural network in de-noising and feature selection, my labmates and I developed scGNN (single-cell graph neural network) for gene imputation and cell clustering. Finally, a novel computational framework named IRIS3 (integrated cell-type-specific regulon inference server from single-cell RNA-Seq) was designed to identify CTSRs from scRNA-Seq data based on the integration of functional gene modules, cell clusters, and cis-regulatory motif enrichment. It is the first-of-its-kind web server for CTSR inference for human and mouse scRNA-Seq data. I reasoned that CTSRs can be used to reliably characterize and distinguish the corresponding cell type from others and can be combined with other computational or experimental analyses for biomedical studies. CTSRs can, therefore, aid in the discovery of major regulatory mechanisms and allow reliable constructions of global transcriptional regulation networks encoded in a specific cell type. The broader impact of IRIS3 includes, but is not limited to, investigation of complex diseases hierarchies and heterogeneity, causal gene regulatory network construction, and drug development.