Date of Award

Fall 10-1-2021

Document Type


Degree Name

Doctor of Philosophy (PhD)


Computational Biology and Bioinformatics

First Advisor

Zhao, Hongyu


High throughput single cell sequencing has seen exciting developments in recent years. With its high resolution characterization of genetics, genomics, proteomics, and epigenomics features, single cell data offer more insights on the underlying biological processes than those from bulk sequencing data. The most well developed single cell technologies are single cell RNA-seq (scRNA-seq) on transcriptomics and flow cytometry on proteomics. Many multi-omics single cell sequencing platforms have also emerged recently, such as CITE-seq, which profiles both epitope and transcriptome simultaneously. But some well known limitations of single cell data, such as batch variations, shallow sequencing depth, and sparsity also present many challenges. Many computational approaches built on machine learning and deep learning methods have been proposed to address these challenges. In this dissertation, I present three computational methods for joint analysis of single cell sequencing data either by multi-omics integration or joint analysis of multiple datasets. In the first chapter, we focus on single cell proteomics data, specifically, the antibody profiling of CITE-seq and cytometry by time of flight (CyTOF) applied to single cells to measure surface marker abundance. Although CyTOF has high accuracy and was introduced earlier than scRNA-seq, there is a lack of computational methods on cell type classification and annotations for these data. We propose a novel automated cell type annotation tool by incorporating CITE-seq data from the same tissue, publicly available annotated scRNA-seq data, and prior knowledge of surface markers in the literature. Our new method, called automated single cell proteomics data annotation approach (ProtAnno), is based on non-negative matrix factorization. We demonstrate the annotation accuracy and robustness of ProtAnno through extensive applications, especially for peripheral blood mononuclear cells (PBMC). The second chapter introduces an integrative method improving bulk sequencing data decomposition into cell type proportions by harmonizing scRNA-seq data across multiple tissues or multiple studies. As a Bayesian model, our method, called tranSig, is able to construct a more reliable signature matrix for decomposition by borrowing information from other tissues and/or studies. Our method can be considered an add-on step in cell type decomposition. Our method can better derive signature gene matrix and better characterize the biological heterogeneity from bulk sequencing datasets. Finally, in the last chapter, we propose a method to jointly analyze scRNA-seq data with summary statistics from genome wide association studies (GWAS). Our method generates a set of SNP (single nucelotide polymorphism)-level weight scores for each cell type or tissue type using scRNA-seq atlas. These scores are combined with risk allele effect sizes to decompose polygenic risk score (PRS) into cell types or tissue types. We show through enrichment analysis and phenome-wide association study (PheWAS) that the decomposed PRSs can better explain the biological mechanisms of genetic effects on complex traits mediated through transcription regulation and the differences across cell types and tissues.