Inference of genetic regulatory networks with regulatory hubs using vector autoregressions and automatic relevance determination with model selections

Inference of gene networks using gene expression data with applications

Gene networks (GNs) use graphs to represent the interaction relationships between genes. Large-scale GNs are often sparse and contain hub genes that interact with many other genes. In this paper, we propose a novel method called NetARD, which utilizes Automatic Relevance Determination (ARD) to estimate partial correlations, to infer GNs with the hub genes from gene expression data. We test NetARD on simulated GNs and in silico GNs, and it outperforms existing methods. In our high-throughput gene expression data analysis, we integrate the NetARD into a method called GN Co-expression Extension (GNCE). This approach infers the GNs of co-expressed genes, with genes from a predefined GN serving as hub genes. We validate this approach by extending the core GN of transcription factor genes of E. coli using microarray data. In an application example, we identify biological process (BP) Gene Ontology (GO) terms that are significantly involved in cancer progression. This task is accomplished by analyzing the GN inferred through GNCE using the core GN associated with the colorectal cancer pathway and RNA-seq data.