Computational Methods for the Analysis of Single-Cell Transcriptomic Data and Their Applications to Cancer

Abstract

Single-cell sequencing methods have allowed for a closer view into the heterogeneity of cell populations, down to the level of the individual cell. In particular, single-cell transcriptomic data provides a detailed map of the diverse gene expression profiles present throughout a sample. These new methods have shown promise in many fields, especially the field of cancer genomics. However, despite the technological advances, many computational challenges remain. In Chapter 1, we provide an overview of single-cell technological and computational methods through the perspective of cancer genomics. Cancer is a notoriously heterogeneous disease whose characteristics can differ greatly from person to person and cell to cell, making treatment a very challenging proposition. Single-cell methods allow us to dissect within-tumor heterogeneity, opening up the possibility of developing individualized therapies that target specific cancer cell subpopulations within a patient. We next address two current challenges associated with the analysis of single-cell transcriptomic data. The first challenge is the difficulty of detecting rare cell populations. Current clustering methods either only detect prevalent cell types, or specifically target only the detection of rare cell types. In Chapter 2, we develop a clustering method, GiniClust2, that can accurately identify both rare and common cell types using a novel, cluster-aware ensemble method that combines clustering results from rare and common cell-type-specific clustering methods. The second challenge we address is the inference of cell-type composition from bulk gene expression data when single-cell data is unavailable. In Chapter 3, we propose a gene expression deconvolution method that estimates cell-type composition using a gene signature derived from single-cell data. We also introduce a novel dampened weighted least squares algorithm (DWLS) for estimation that adjusts for biases present in existing estimation methods, to create a method that can more accurately detect diverse cell types. Finally, in Chapter 4, we conclude with two applications of single-cell RNA-sequencing data analysis to the discovery of immune response mechanisms in cancer. This highlights the impact such methods can have on helping cancer immunologists identify drug targets and assess their effects.