Abstract:
Traditional RNA sequencing methods, such as bulk RNA-seq, fail to capture cellular heterogeneity. In contrast, single-cell RNA-seq addresses this limitation by profiling individual cells, but it loses spatial context. Spatial transcriptomics overcomes these constraints by preserving the spatial arrangement of gene expression, enabling researchers to map cellular organization and explore intercellular interactions within the tissue microenvironment. This spatial context is essential for understanding complex biological processes such as tumor progression, where cell-cell communication plays a pivotal role. However, analyzing spatial transcriptomics data presents significant challenges, including high data sparsity, the complexity of cell type annotation, and the difficulty of accurately inferring cell-cell communication. In this thesis, we address these challenges by implementing a computational framework that integrates dataset pre-processing, cell type annotation, cell state identification, and spatial neighborhood inference using tissue coordinates. For cell type annotation, we developed a novel method using UniPath, a normalization-free gene set enrichment approach, and the pre-existing method scType, which leverages curated marker gene sets for accurate classification of cell identities. To investigate gene-gene interactions and cell-cell communication, Spearman correlation was employed to identify transcriptional associations across nearest-neighbor cell clusters. Ligand-receptor interaction analysis was performed using curated databases via CellPhoneDB and stLearn. At the same time, Bayesian modeling was applied to validate the consistency and significance of the observed correlations and ligand-receptor interactions. Our pipeline revealed challenges in a breast tumor microenvironment, such as overlapping expression profiles across stromal and immune cells, the difficulty of resolving rare cell types, and the integration of noisy single-cell and spatial signals. Through pathway and Cancer Hallmark enrichment using UniPath, we identified biologically meaningful interactions within the tumor microenvironment.
