Probabilistic Multiscale Modeling of the Tumor Microenvironment - Abstract The research project proposed here addresses the pressing need for better statistical models and methods to analyze the spatial architecture of the tumor microenvironment (TME). TME data has demonstrated there is clear clinical and biological importance in the spatial architecture, e.g. as a determinant of response to treatment and metastasis. Given the recognized importance of the TME in cancer, technology has advanced at pace to profile the spatial properties of tumors using high resolution measurements including spatial transcriptomics and proteomics. However, the requisite computational methods to fully interpret these measurements are lagging. Accordingly, in Aim 1 we will develop a statistical framework, which we call BayesTME to model the TME at multiple scales, ranging from the level of individual cells to top-level patient stratification. We will develop BayesTME as a suite of innovative statistical methods for Bayesian multiscale spatial modeling that would enable a new class of spatial statistical models to quantitatively evaluate the properties of the TME. We have gathered a diverse and scaled collection of spatial profiling datasets upon which to test, benchmark and evaluate BayesTME. In developing BayesTME, we will create modular tools in Aim 2 that can use the same statistical concepts across different technologies such as multiplexed immunofluorescence, imaging mass cytometry and spatial transcriptomics. This will permit statistical integration of datasets that may have been generated with diverse sets of technology. In addition, we will include an extension to the base BayesTME to identify recurrent spatial properties across a cohort of samples, enabling discovery and quantitative description of spatial communities that related to specific cancer phenotypes. Finally, in Aim 3 we propose a validation experiment that will generate parallel spatial profiling data–in the form of spatial transcriptomics and imaging mass cytometry from the same ovarian cancer specimens. This dataset will help to address a critical question in ovarian cancer which is how TME dynamics enable bowel metastasis - a major determinant of morbidity and mortality for women with ovarian cancer. In summary, the goals of this proposal are to develop a robust, new class of statistical methods for analyzing the spatial architecture of the TME, generate robust open-source software enabling application of our methods across multiple spatial profiling techniques, and validate our methods and software by using them to conduct large-scale data analyses investigating novel biological hypotheses regarding the spatial architecture of the TME. Accomplishing these goals will lead to new quantitative encoding of the properties of the TME that are statistically grounded that will in turn lead to a new class of spatial biomarkers to define malignant phenotypes in cancer.
