Computational Characterization of the Tumor Microenvironment for Optimizing Immunotherapy Response
Immunotherapy has transformed cancer treatment, yet most patients do not achieve a durable benefit. A prevailing hypothesis is that differences in the tumor microenvironment (TME) influence therapeutic response, highlighting the need for accurate tools to characterize tumor cellular composition. In the first part of my PhD, I developed xCell 2.0, an improved computational method for inferring cell type enrichment from bulk transcriptomic data. The framework introduces a flexible training pipeline and automated handling of cell type dependencies, enabling robust performance across diverse reference datasets. In extensive benchmarking across human and mouse datasets, xCell 2.0 outperformed existing deconvolution methods and improved prediction of immunotherapy response in pan-cancer analyses. In the second part, I investigated how chemotherapy may reshape the TME to influence response to subsequent immunotherapy. I derived a transcriptomic signature that identifies tumors likely to undergo favorable immune remodeling following chemotherapy. Notably, this benefit is specific to patients receiving combined chemo-immunotherapy and not immunotherapy alone, consistent with a model of chemotherapy-mediated immune priming. Together, this work provides both a methodological advance for TME characterization and a mechanism-informed framework for identifying patients who may benefit from combined or sequential chemo-immunotherapy strategies.
