When information is continuous, but you only observe binary data, it can be difficult to detect association between variables. This method proposes a new model based on thresholding, and implements a search procedure that identifies groups of latently associated variables.
This method was developed to identify sets of genes that are more inter-correlated under one condition than under another; genes that are correlated among patients with a certain disease, but uncorrelated among those without the disease. The DCM method relies on statistical testing principles to iteratively search until a significantly differential set is discovered.