Figure 1

The schematic overview of learning contextual gene set interaction networks and identifying condition specificity. From the gene expression matrix, contextual gene sets are identified through the context-mining process. The expression values of genes in each contextual gene set for each sample are summarized into one major representative value, and a contextual gene set expression matrix is built as a result. Multiple Bayesian networks are learned from this matrix and their consensus network (undirected dependency likelihood matrix) is built while ignoring the direction of connections. For each condition, a subset of data is built by discarding the samples of the condition from the original data and a new dependency likelihood matrix is built from it. If the dependency likelihood of the interaction between G _i and G _j from all samples is significantly larger than the dependency likelihood from a data without a condition I, the interaction is specific to the condition I.