Statistical methods for spatial transcriptomics data

Abstract

Spatial transcriptomics is a set of techniques that enables the quantification of gene expression within intact tissue sections, preserving the spatial context of where specific genes are active. The main goal of this master thesis is to review state-of-the-art statistical methods for analyzing spatial transcriptomics data and to draw parallels with traditional spatial statistical analysis, providing a robust theoretical and mathematical foundation for these methodologies. Section 1 provides an introduction to basic concepts in genomics and an overview of spatial transcriptomics technologies, which are divided between sequencing-based and imaging-based technologies. Section 2 presents various spatial statistical concepts such as spatial autocorrelation, kriging, Gaussian-Markov random fields and point processes. Section 3 discusses normalization methods for different types of RNAseq data: bulk RNA-seq, single-cell RNA-seq and spatial transcriptomics, highlighting its similiarities and differences. Section 4 covers dimensionality reduction methods, ranging from nonspatial approaches such as PCA to spatially-aware methods that extend PCA by incorporating spatial information. Section 5 explores the identification and analysis of spatial domains, distinguishing between non-spatial approaches, such as the Louvain and Leiden methods, and spatially-aware methods specifically designed for spatial transcriptomics data. Section 6 examines various approaches for detecting spatially variable genes, based on different strategies, including methods that use Gaussian process regression, marked point process theory, and other nonparametric techniques. Section 7 addresses cell type deconvolution strategies.