Like other cancers, glioblastoma is cancer caused by genomic perturbations.The Cancer Genome Atlas (TCGA) project has holistically characterized a large number of glioblastoma tumors at a molecular level.The seminal study by the TCGA groups has identified four subtypes of glioblastomas, which may have different underlying disease mechanisms and showed differential response to treatment.In this study, we address the task of identifying signaling pathways whose perturbation lead to differential clinical outcomes of glioblastomas.Using the TCGA data, we developed an approach integrating knowledge mining and graph-based data mining to identify non-disjoint, functionally related gene modules that are predictive.of clinical outcome of glioblastoma patients.We further investigated the genomie perturbation, somatic mutations and CNAs, that drive the differential expression of gene modules.This body of information shed light on different disease mechanisms that lead to different clinical outcomes, which can be used to guide personalized treatment of glioblastoma patients.