Integrated Mechanistic and Data-Driven Computational Modeling of Kinase-Regulated Cell Signaling

Computational modeling has become a mainstay approach for deciphering the regulatory mechanisms of complex cell signaling networks. As experimental methods have evolved to generate increasingly rich datasets with high temporal and spatial resolution across many signaling nodes, the need for more sophisticated computational models has grown. In particular, the now-ubiquitous presence of high-dimensional datasets characterizing signaling processes has driven a massive increase in the application of data science models. The value of those models is limited, however, by their inability to capture the biophysical processes that govern the propagation of signaling, motivating a need to constrain the interpretation of data science models through integrated mechanistic modeling. This dissertation research describes the development and application of novel mechanistic and data-driven computational models of kinase-regulated mammalian cell signaling, with the final example demonstrating an approach for integrating these two fundamentally distinct modeling types to connect perturbations in the biophysical processes governing signaling initiation to quantitative changes in cell phenotypes. In the first study, we generate a spatiotemporal model of the reaction and diffusion mechanisms that enable membrane-bound epidermal growth factor receptor (EGFR), a tyrosine kinase, to maintain the assembly of functional complexes of the adaptor GRB2-associated binder 1 (GAB1) and the cytosolic phosphatase SH2 domain-containing phosphatase 2 (SHP2) throughout the cytoplasm. The results of the coupled partial and ordinary differential equation-based model are robust to parameter perturbations, consistent with independent order of magnitude estimates, and supported by imaging analysis in mammalian cells. In the second study, data-driven models of publicly available patient tumor data are used to identify the kinases responsible for driving epithelial-mesenchymal transition (EMT) in the hypoxic tumor microenvironment of pancreatic ductal adenocarcinoma (PDAC). Clustering and multivariate linear models based on bulk and single-cell omics data confirmed a significant relationship between hypoxia and EMT and systematically identified MAPK signaling as the most likely driver of EMT in hypoxic PDAC tumors and cells. Finally, we develop a novel integrated mechanistic/data-driven model that explains the ability of EGFR ligands with different receptor binding affinities to drive distinct cellular decisions through common effector kinase signaling pathways. The experimentally trained and validated integrated model bridges the gap between reliable mechanistic model predictions of signaling dynamics and phenotypes by using a partial least squares regression modeling approach with signaling dynamics as inputs and predicts how quantitative changes in ligand-receptor binding kinetics propagate to altered cell fate decisions. Collectively, these analyses add new fundamental understanding of the spatiotemporal regulation of signal transduction, regulation of cell phenotypes by coordinated kinase signaling, and methods to integrate different computational approaches to model the complete signaling cascade from input to cell fate decision.