A data-driven integrative framework for the identification of cancer driver pathways and their mode of action.
Network-based methods (NBM) are ideally suited to leverage in-house data with expanding public resources. These methods guide their analysis by a gene interaction network prior. Currently used prior networks are at same time context-independent and incomplete, which negatively affects the performance of NBM. In addition, current NBM rely on oversimplified network models that cannot exploit the specificities of the individual genetic aberrations nor exploit the full potential of the available data. For instance, information on functional effects other than differential expression are not accounted for (splice variants, post-translational modifications). Therefore, in this project we aim at 1) developing an innovative framework that allows converting publicly available data in a context-dependent network prior. This method, based on network representation learning, will allow summarizing relevant omics information in respectively a graph and vector based prior representation of the interaction network. 2) Improving NBM to use the probabilistic graph based prior and to accommodate the intricacies of the data. 3) Developing an alternative integrative method that operates directly on the vector rather than graph based representation. Our methods will be tuned in the context of cancer systems genetics and applied on a unique set of tumor data to gain insight into pathways driving the early onset of breast cancer and into the molecular mechanisms driving metastatic prostate cancer.