MSIGNET: A Bayesian Approach for Disease-associated Gene Network Identification

Abstract

The analysis of gene networks and signalling pathways plays a key role in understanding gene functions, i.e., their effects on the development of a particular disease. Yet, for many heterogeneous diseases, the number of known disease-associated genes is limited. Identifying disease-associated genes is still an open challenge. To understand the functions of genes associated with a disease, we develop a Metropolis-Hastings sampling based SIGnificant NETwork (MSIGNET) identification approach. MSIGNET integrates disease gene expression data and human protein-protein interactions in a Bayesian network, and identifies interactions of genes specifically expressed under the disease condition. We applied MSIGNET to simulation and benchmark data. Results demonstrated its superior performance over conventional network identification tools on disease-associated gene network identification when multiple local gene modules existed. To learn genes and functional signalling pathways associated with ovarian cancer recurrence, we identified a gene network using TCGA ovarian cancer gene expression data and further validated results using an independent gene expression data set. Genes in the identified network were significantly enriched with cellular processes relevant to ovarian cancer development, and as