Computational methodologies and resources for discovery of phosphorylation regulation and function in cellular networks

Abstract

Post-translational modifications (PTMs) regulate cellular signaling networks by modifying activity, localization, turnover and other characteristics of proteins in the cell. For example, signaling in receptor tyrosine kinase (RTK) networks, such as those downstream of epidermal growth factor receptor (EGFR) and insulin receptor, is initiated by binding of cytokines or growth factors, and is generally propagated by phosphorylation of signaling molecules. The rate of discovery of PTM sites is increasing rapidly and is significantly outpacing our biological understanding of the function and regulation of those modifications. The ten-fold increase in known phosphorylation sites over a five year time span can primarily be attributed to mass spectrometry (MS) measurement methods, which are capable of identifying and monitoring hundreds to thousands of phosphorylation sites across multiple biological samples. There is significant interest in the field in understanding these modifications, due to their important role in basic physiology as well as their implication in disease. In this thesis, we develop algorithms and tools to aid in analysis and organization of these immense datasets, which fundamentally seek to generate novel insights and testable hypotheses regarding the function and regulation of phosphorylation in RTK networks. We have developed a web-accessible analysis and repository resource for high-throughput quantitative measurements of post-translational modifications, called PTMScout. Additionally, we have developed a semi-automatic, high-throughput screen for unsupervised learning parameters based on their relative ability to partition datasets into functionally related and biologically meaningful clusters. We developed methods for comparing the variability and robustness of these clustering solutions and discovered that phosphopeptide co-clustering robustness can recapitulate known protein interaction networks, and extend them. Both of these tools take advantage of a new linear motif discovery algorithm, which we additionally used to find a putative regulatory sequence downstream of the highly tumorigenic EGFRvIII mutation that indicates casein kinase II (CK2) activity may be increased in glioblastoma.