Exploiting patterns in genomic data for personalised cancer treatment and new target discovery

In response to a global requirement for improved cancer treatments a number of promising novel targeted cancer therapies are being developed that exploit vulnerabilities in cancer cells that are not present in healthy cells. In this thesis I explore different ways of identifying the vulnerabilities of cancer cells, with the ultimate aim of providing personalised therapies to cancer patients on an individual basis. I first investigate approaches that utilise the concept of synthetic lethality. Therapies that exploit synthetic lethality are suitable where a specific tumour suppressor has been inactivated by a cancer and an identified synthetic lethal (SSL) pair for that gene may be therapeutically targeted. Mainly due to the constraints of the experimental procedures, relatively few human SSL interactions have been identified. Here I describe computational systems approaches for predicting human SSL interactions by identifying and exploiting conserved patterns in protein-protein interaction (PPI) network topology both within and across model species. I report that my classifiers out-perform previous attempts to classify human SSL interactions. Experimental validation of my predictions suggest they may provide useful guidance for future SSL screenings and ultimately aid targeted cancer therapy development. All predictions from this study have been made available via a new online database that I designed, built and published. As an extension to this approach I used similar network features to predict gene dependencies, otherwise known as acquired essential genes, in specific cancer cell lines. Genetic alterations found in each individual cell line were modelled using the novel approach of removing protein nodes to reflect loss of function mutations and changing the weights of edges in each protein-protein interaction network to reflect gain of function mutations and gene expression changes. I report that base PPI networks can be used to successfully classify human cell line specific gene dependencies within individual cell lines, between cell lines and even across tissue types. Furthermore, my personalised PPI network models further improve prediction power and show improved sensitivity to rarer gene dependencies, an improvement which offers opportunities for personalised therapy. In a therapeutic context these essential genes would be suitable as individual drug targets for each specific patient. Finally, I analyse copy number variance and ploidy in a set of cancers from kidney patients. Using clustering algorithms I investigate patterns in cancer cell line arm-wise ploidy and identify factors that may be driving this genomic instability