Our paper on ridge estimation of mixtures of Gaussian graphical models (GGMs) appeared online in the Biometrical Journal. To learn GGMs from a possibly heterogeneous sample comprising multiple groups of patients, we present there a penalized EM algorithm that identifies these hypothesized groups and fits one GGM per such group. This approach is extended to deal with data from cross-sectional studies in which patients originate from various stages. It aims to unravel the within-stage heterogeneity of the patients but also to link the identified groups across stages. Read more.