Modelling human cell protein phosphorylation networks

Defects in cell signalling networks are linked to over 400 human diseases. My thesis research aimed to model these networks in more detail to facilitate understanding of their architecture and operations under normal and pathological conditions. The various protein levels in diverse normal human cell and tissue types were inferred from their mRNA expressions, and their up/down-regulation was also investigated in about 300 human cancer cell lines and 50 types of human cancers. This was based on meta-analyses of gene microarray measurements deposited in the USA National Center for Biotechnology Information's Gene Expression Omnibus database. I identified proteins that were commonly or uniquely expressed in normal and cancerous human cells and tissues. The co-expression patterns of proteins were used to predict potential interactions, but there was not a strong correlation between high co-expression and actual direct protein-protein interactions documented in the scientific literature. With respect to the post-translational regulation of proteins, my research efforts primarily targeted protein phosphorylation, which is the most predominant type of reversible covalent modification of proteins. Complex protein phosphorylation networks emerge through the interplay of protein kinases, protein phosphatases, phosphorylation site-dependent binding proteins, and their phosphoprotein substrates. I modelled the interactions of protein kinases with substrate proteins and inhibitory compounds. Nearly a million human phosphosites were predicted, and each of these was tested in silico as substrates for 500 human protein kinases. The interactions of over 550 known protein kinase inhibitory drugs with the 500 protein kinases were also tested. These predicted interactions were compared with empirical data from other on-line protein-protein and protein-drug interaction databases. The human phosphosites were also analysed with respect to their protein conservation in over 20 other diverse species, and it was found that threonine phosphosites in protein kinases that were activatory were particularly well conserved in evolution. Finally, probabilistic graphical models were developed to model the most probable structure of substrate phosphosites for specific protein kinases. The discussed probabilistic graphical model, gave more theory justifications for the protein-protein interaction modelling that was presented in the earlier parts of the thesis.Science, Faculty ofComputer Science, Department ofGraduat