Mechanism-based disease similarity: with random walks starting from a set of disease genes, traveling through the protein-protein interaction network, and then back.*

Diseases have been traditionally classified based on the similarity between the symptoms they cause. One may wonder, however, could two or more diseases with drastically different symptoms have the same or similar underlying causes? To answer this question, it is essential to have a measure of similarity between diseases that reflects the underpinning molecular interactions. Such a measure can then be used for mechanism-based disease classification and for construction of a molecular-based human disease network. To achieve these goals, a bioinformatic tool called DeCoaD, which uses both genomic and proteomic data, has been developed. DeCoaD in turn utilizes ITMProbe, another in-house bioinformatic tool that employs damped random walks to analyze the flow of information in protein-protein interaction networks. In this talk the similarity measure and classification algorithm used by DeCoaD are presented and the resulting human disease network is explored. It is shown that the answer to the aforementioned question is positive. For example, DeCoaD demonstrates a significant similarity between dilated cardiomyopathy, a heart muscle disease, and infant diabetes mellitus. Several other examples of such seemingly unrelated pairs that have significant pair-wise similarities according to DeCoaD and the supporting experimental evince for these similarities are given. Finally, the utility of DeCoaD and the disease network for creation of a metadatabase for diseases is discussed. Such a metadatabse can be used for discovery of possible new disease relationships and validation of such relationships.