In this thesis, we develop the combination of two concepts. The first one, the Model-X Framework is a set of results and algorithms related to variable selection in the supervised setting. By shifting the burden of assumptions from the response to the covariables, it guarantees finite sample False Discovery Rate (FDR) control in the high-dimensional case. The second coverage is the theory of conditional Gaussian distributions which serve as statistical models for associations between variables, some of which are quantitative and some qualitative. We reveal, how both of these concepts can be combined to form powerful instruments for the structure learning problem of conditional Gaussian graphical models. In the end of the day, a selection rule will be presented which controls the expected proportion of falsely discovered edges by a prechosen target FDR Level.     «

In this thesis, we develop the combination of two concepts. The first one, the Model-X Framework is a set of results and algorithms related to variable selection in the supervised setting. By shifting the burden of assumptions from the response to the covariables, it guarantees finite sample False Discovery Rate (FDR) control in the high-dimensional case. The second coverage is the theory of conditional Gaussian distributions which serve as statistical models for associations between variables,...     »