Large scale single cell methods and pan-cancer analysis

Abstract

The widespread adoption of single cell omic technologies has enabled generation of large cell atlases that often consists of more than hundreds of thousands of cells. The large data size and heterogeneity pose significant computational and scientific challenges. With limited computational resources, new scalable algorithms and software are required to properly analyze these atlas-scale single cell data sets. In this thesis, we present a collection of new algorithms, bioinformatics tools, and applications related to large-scale analysis of single cell omic data. • A scalable clustering method called FlowGrid that can significantly reduce run time when analyzing very large single cell omic data. • A web-based immersive visualization tool called StarmapVis that enables effective exploration of single cell and spatial omic data. • A high-throughput and sensitive single-nucleus total RNA sequencing technique called HHseq is developed for the detection of total RNA from frozen clinical samples such as tumors. Using HHseq, we collected more than 730,000 single nuclei from 32 patients with various tumor types. An innovative analysis platform has been established to specifically encompass the entire process from upstream to downstream. • The pan-cancer analysis focuses on dissecting the transcriptional changes happened in malignant cells. The functions of key cell types and regulatory genes are analyzed. Specifically, pan-cancer malignant cell sub-cluster are detected, ciliated-like malignant groups show important functions in cancer proliferating and signaling. Furthermore, the investigation reveals mutual mutations in small nuclear RNAs (RNU4-1, RNU6-31P) among malignant cells, highlighting their potential splicing effects on cancer markers implicated in tumorigenesis mechanisms. Overall, the thesis highlights the importance of large-scale single-cell analysis in understanding complex biological systems such as cancer. The insights gained from this research can have important implications for developing suitable bioinformatics tools and advancing our understanding of biological systems from multiple aspects.