Data comparison using Gaussian Graphical Models

This paper focuses on estimated Gaussian Graphical Models (GGM) from sets of experimental data. Some extension of known Bayesian methods are proposed, allowing to introduce score functions to measure the relevance of the obtained GGM structure to describe the data. These score functions form the basic measurement to derive a new dissimilarity matrix based on the GGM structure. This latter is then exploited for classification purpose. Examples are provided using both simulated and real experimental functional Magnetic Resonance Imaging (fMRI) data.