Mathematical Modeling And Computational Predictions In Oncoimmunology PDF Download

Mathematical modeling and computational biology have been used to understand, describe, and predict critical behaviors of cancer progression. Recent technological advancements in the acquisition of single cell resolution data by high-throughput micrographic imaging and by single cell genomics now enable new analyses of cancer cells at the individual cell and cell population levels. This dissertation focuses on the development of math modeling frameworks capable of integrating and improving our utilization of these novel data types. First, we investigate the relevance of deviations from the conventional exponential growth model via an ecological principle known as the Allee effect, in which cancer cells exhibit cooperative growth dynamics at low population densities relevant in tumor initiation and metastases. Using a large number of single cell resolution growth trajectories acquired at low cell densities, we apply a stochastic parameter estimation framework to systematically evaluate the relevance of an Allee effect in a controlled experimental setting. Our findings reveal evidence for cooperative growth even in the presence of optimal space and nutrients, giving us motivation to consider Allee effects in making predictions regarding treatment response and tumor initiation. The remainder of our work focuses on utilizing multimodal data sources to better understand the dynamics of resistance to chemotherapy. We utilize a mathematical model describing the effects of a treatment-induced resistance on a population of cancer cells and seek to utilize available snapshot and longitudinal data to identify the model parameters. Using lineage tracing technologies developed in the Brock lab, the transcriptomic data set is made actionable by developing a classifier capable of predicting whether a cell in a sample is sensitive or resistant to chemotherapy. We apply this to estimate the composition of the population at a few snapshots in time during treatment response and combine this with longitudinal data directly into our model calibration. The explicit incorporation of molecular level data with population-size dynamics data improves the identifiability and predictive power of the mathematical model. We intend this work to be exemplary of ways in which novel methods can improve the use of data to describe, evaluate, predict, and optimize cancer treatments