Great efforts have been devoted to alleviate uncertainty of detected cancer genes as accurate identification of oncogenes is of tremendous significance and helps unravel the
biological behavior of
tumors. In this paper, we present a differential network-based framework to detect biologically meaningful
cancer-related genes. Firstly, a gene regulatory network construction algorithm is proposed, in which a boosting regression based on likelihood score and informative prior is employed for improving accuracy of identification. Secondly, with the algorithm, two gene regulatory networks are constructed from case and control samples independently. Thirdly, by subtracting the two networks, a differential-network model is obtained and then used to rank differentially expressed hub genes for identification of
cancer biomarkers. Compared with two existing gene-based methods (t-test and lasso), the method has a significant improvement in accuracy both on synthetic datasets and two real
breast cancer datasets. Furthermore, identified six genes (TSPYL5, CD55, CCNE2, DCK, BBC3, and MUC1) susceptible to
breast cancer were verified through the literature mining, GO analysis, and pathway functional enrichment analysis. Among these oncogenes, TSPYL5 and CCNE2 have been already known as prognostic
biomarkers in
breast cancer, CD55 has been suspected of playing an important role in
breast cancer prognosis from literature evidence, and other three genes are newly discovered
breast cancer biomarkers. More generally, the differential-network schema can be extended to other complex diseases for detection of disease associated-genes.