Cancer is a disease mainly caused by somatic genome alterations (SGAs) that perturb cellular signalling systems. Furthermore, the combination of pathway aberrations in a tumour defines its disease mechanism, and distinct disease mechanisms underlie the inter-tumour heterogeneity in terms of disease progression and responses to therapies. Discovering common disease mechanisms shared by tumours would provide guidance for precision oncology but remains a challenge. Here, we present a novel computational framework for revealing distinct combinations of aberrant signalling pathways in tumours. Specifically, we applied the tumour-specific causal inference algorithm (TCI) to identify causal relationships between SGAs and differentially expressed genes (DEGs) within tumours from the Cancer Genome Atlas (TCGA) study. Based on these causal inferences, we adopted a network-based method to identify modules of DEGs, such that the member DEGs within a module tend to be co-regulated by a common pathway. Using the expression status of genes in a module as a surrogate measure of the activation status of the corresponding pathways, we divided breast cancers (BRCAs) into five subgroups and glioblastoma multiformes (GBMs) into six subgroups with distinct combinations of pathway aberrations. The patient groups exhibited significantly different survival patterns, indicating that our approach can identify clinically relevant disease subtypes.