Gene set-level network analysis using a toxicogenomics database

Toxicogenomics data sets on rat livers covering 118 compounds were subjected to inference of a gene set-level, not individual gene-level, network structure. Expression changing levels for 58 gene sets was used for network inference with a Gaussian graphical model algorithm. The established network contained reasonable relationships, such as ones between the blood glucose level and glycolysis-related genes or the blood transaminase level and cellular injury-related genes, indicating that the gene set-level network inference successfully highlighted biological pathway-level interactions. In addition, the robustness of the inferred network structure was investigated using microarray data on bromobenzene-treated rat livers, where the gene set-level activation exhibited time-dependent propagation through neighbored nodes (i.e. gene sets) on the network, indicating that the network structure was robust and comparable with an external microarray data set. Accumulating such robust gene sets with toxicity-associated subnetwork structures would lead to a better understanding of the molecular mechanisms of drug-elicited toxicities.