Mixture of Conditional Gaussian Graphical Models for Unlabelled Heterogeneous Populations in the Presence of Co-factors

Conditional correlation networks, within Gaussian Graphical Models (GGM), are widely used to describe the direct interactions between components of a random vector. In case an unlabelled Heterogeneous population, Expectation Maximisation (EM) algorithms for Mixtures GGM have been proposed estimate both each sub-population’s graph and class labels. However, we argue that, with most real data, affiliation cannot be described Mixture Gaussian, which mostly groups data points according their geometrical proximity. particular, there often exists external co-features whose values affect features’ average value, scattering across feature space belonging same sub-population. Additionally, if co-features’ effect on features is Heterogeneous, then estimation this separated from sub-population identification. article, propose (CGGM) that subtracts heterogeneous effects regroup into corresponding clusters. We develop penalised EM algorithm graph-sparse model parameters. demonstrate synthetic how method fulfils its goal succeeds in identifying sub-populations where disrupted by co-features.