Probabilistic Graphical Models for Mapping Tumor Clones in Cancerous Tissues and Single Cells

Spatial, genomic, and phenotypic heterogeneity are crucial for understanding cancer progression, treatment, and survival. However, identifying cancer clones and their gene expression profiles alongside their location in the tumor tissue is challenging. This thesis is devoted to a comprehensive modeling of different aspects of tumor heterogeneity and builds upon three projects. In the first project, we focused on the genomic heterogeneity of the tumor and developed a probabilistic model that leverages independent genomic clustering of cells and scarce single-cell RNA sequencing data to map cells to given imperfect genotypes of tumor clones. In the second project, we explored all three aspects of heterogeneity with the main focus on spatial heterogeneity. We developed a complex probabilistic model to accurately infer the cancer clones and their localization in close to single-cell resolution by integrating pathological images, whole-exome sequencing, and spatial transcriptomics data. Expanding upon our previous project, in the third project, we focused on phenotypic heterogeneity. We proposed a probabilistic model that combines spatial transcriptomics and whole-exome sequencing data to accurately identify cancer clones and their gene expression profiles in tumor tissue. Our integrated approach provides a comprehensive understanding of the spatial, genomic, and phenotypic organization of tumors, opening new avenues to study the functional implications of tumor heterogeneity and the origins of resistance to targeted therapies.