Node-based learning of differential networks from multi-platform gene expression data.

Recovering gene regulatory networks and exploring the network rewiring between two different disease states are important for revealing the mechanisms behind disease progression. The advent of high-throughput experimental techniques has enabled the possibility of inferring gene regulatory networks and differential networks using computational methods. However, most of existing differential network analysis methods are designed for single-platform data analysis and assume that differences between networks are driven by individual edges. Therefore, they cannot take into account the common information shared across different data platforms and may fail in identifying driver genes that lead to the change of network. In this study, we develop a node-based multi-view differential network analysis model to simultaneously estimate multiple gene regulatory networks and their differences from multi-platform gene expression data. Our model can leverage the strength across multiple data platforms to improve the accuracy of network inference and differential network estimation. Simulation studies demonstrate that our model can obtain more accurate estimations of gene regulatory networks and differential networks than other existing state-of-the-art models. We apply our model on TCGA ovarian cancer samples to identify network rewiring associated with drug resistance. We observe from our experiments that the hub nodes of our identified differential networks include known drug resistance-related genes and potential targets that are useful to improve the treatment of drug resistant tumors.