Integrating Single-Cell Transcriptomics Data with Spatial Imaging Data

The advent of sophisticated single-cell RNA sequencing (scRNA-seq) techniques now allows investigation of the transcriptomic landscape of tens of thousands of genes across tissues at the resolution of individual cells. However, scRNA-seq necessitates dissociation of the sample, thereby destroying any spatial context which can be crucial to the understanding of cellular development and dynamics. The loss of spatial information in scRNA-seq data can be partially mitigated by referring to known spatial expression patterns of a small subset of genes, termed a "spatial reference atlas." Several recent computational methods have been developed to impute spatial data onto existing scRNA-seq datasets to achieve individual-cell resolution while retaining the spatial arrangement. In Chapter 2, we discuss a novel deep learning-based, system-adaptive method (DEEPsc) of integrating non-spatial scRNA-seq data with spatial imaging data. DEEPsc and other mapping methods rely on a high quality reference atlas which must be compiled from raw images into a useable form. In Chapter 3, we introduce AtlasGeneratorOT, a novel software suite which uses techniques in optimal transport theory to more fully automate the creation of a spatial reference atlas for use with DEEPsc and other integration methods. In Chapter 4, we extend AtlasGeneratorOT with additional capabilities for three-dimensional biological systems imaged in serial slices, allowing for alignment of and interpolation between slices to provide a more cohesive, comprehensive atlas than previously available