Simulation study results. A) Genes grouped into global gene clusters, marked 1-4, are further grouped locally within respective biological contexts, marked X, Y, and Z. Conversely, biological samples are in the same biological context if they have the same groupings of co-clustered genes. Differential co-expression score (DCS) is displayed in the right hand side-panel. B) Average ROC curves were obtained for repeatedly simulated data with noise levels ranging from σ = 0.4 0.8, with σ = 0.5 displayed here, by averaging the FPRs (incorrectly co-clustered pairs of samples) and TPRs (correctly co-clustered pairs) for each distinct tree cut level. C) To summarize ROC curves over all simulations at a given noise level σ, we compute the area under the curve (AUC) for each simulation and plot the average AUC against σ. D) The context structure is same as in A, but expression patterns within each context are modified. E) and F) same as B and C but for the clustering structure in D. Groupings based on traditional similarity measures no longer corresponded to the underlying context structure. DCIM algorithm continues to correctly identify the underlying contexts.