Incorporating prior information into differential network analysis using non-paranormal graphical models

• DOI: 10.1093/bioinformatics/btx208
• Publication Date: 2017-04-05T19:57:27Z
• Authors: Xiao-Fei Zhang; Le Ou-Yang; Hong Yan
• Keywords: Graphical model; Gene regulatory network

Understanding how gene regulatory networks change under different cellular states is important for revealing insights into network dynamics. Gaussian graphical models, which assume that the data follow a joint normal distribution, have been used recently to infer differential networks. However, distributions of omics are non-normal in general. Furthermore, although much biological knowledge (or prior information) has accumulated, most existing methods ignore valuable information. Therefore, new statistical needed relax normality assumption and make full use information.We propose analysis method address above challenges. Instead using we employ non-paranormal model can assumption. We develop principled take account following information: (i) edge less likely exists between two genes do not participate together same pathway; (ii) changes driven by certain regulator perturbed across (iii) estimated from multi-view expression share common structures. Simulation studies demonstrate our outperforms other model-based algorithms. apply identify platinum-sensitive platinum-resistant ovarian tumors, proneural mesenchymal subtypes glioblastoma. Hub nodes rediscover known cancer-related contain interesting predictions.The source code at https://github.com/Zhangxf-ccnu/[e-mail] available Bioinformatics online.