We have made great strides in diagnosing and treating cancer. However, cancer treatment is still often fraught with uncertainty. There is evidence that similar cancers have many variations at the molecular level, each with its own clinical course; this can be described as the heterogeneity of a tumor. A signal transduction pathway (STP) is a network of information flow in the cells that initiates with a signal outside the cell and results in a cellular response. Cancer is a disease in which aberrations to STPs can lead to altered protein activity, which in turn can lead to abnormality of cellular behavior. Cancer is a complex disease due to the following factors: 1) Cancer can result from perturbations in multiple biological processes; 2) A tumor of the same subtype can result from different genetic alterations; and 3) The same STP can be modified by distinct genetic mutations in different tumors. Bayesian networks (BN) represent probabilistic relationships among variables. We learn the deviated protein activity on an STP by a BN approach. Specifically, our method determines the STPs likely to be altered in the tumor in a group of patients. In a preliminary study, we applied our method to detect the aberrated signaling transduction pathways in breast cancer patients using TCGA gene expression data.