The mutual information graph of a distribution over a product of a $n$ sets is a weighted graph which is complete and whose edge weights are given by the mutual information between the marginals over pairs of components.