Data-Driven Computational Modeling of the State and Architecture of the Breast Cancer Kinome

The complex signaling in the kinome provides a unique insight into breast cancer, which is heterogeneous with many disease states or subtypes. The kinome has been implicated in many cancers and is highly targeted by inhibitor therapies because of its importance in cell proliferation and differentiation. High-throughput data sets using proteomics help characterize the kinome and allow quantification of the baseline and perturbed states of the kinome. These high-throughput experimental methods allow for quantification of kinases that are not well-studied, or are understudied. In this thesis, I employ machine-learning techniques to distinguish between breast cancer subtypes using a functional proteomics data set and to demonstrate that the state of the kinome looks different in proteomic and sequencing data sets. Characterized, as well as understudied, kinases are identified as important features in stratifying unperturbed breast cancer subtypes. In addition, both understudied and characterized kinases respond dynamically across breast cancer subtypes in response to kinase inhibitor therapy treatment. Further, I developed computational methodologies to characterize the architecture of the kinome network and an optimization method for choosing effective combination therapies for cancer treatment. Public protein-protein interaction databases are compiled to create the comprehensive kinome network, consisting of only kinase to kinase interactions. The comprehensive kinome network is clustered to identify functional modules, or subnetworks, and some of these subnetworks are significantly enrichment for understudied and targeted kinases. In addition, the optimization proposed here provides a computational framework for choosing effective sets of inhibitors to use concurrently, i.e. combination therapies.Doctor of Philosoph