Fig. 3

Illustration of data generation and analysis procedure in simulations. For each precision matrix \({\varvec{\Theta }}\) we consider, we modify it to create a prior precision matrix \({\varvec{\Theta }}_{\text{prior}}\). A prior data set \({\varvec{X}}_{\text{prior}}\) is generated from the resulting multivariate Gaussian distribution. The graphical lasso tuned by StARS is then used on this data to obtain a prior precision matrix estimate \(\widehat{{\varvec{\Theta }}}_{\text{prior}}\), and the absolute values of the corresponding partial correlation estimates are used as weights in the tailored and weighted graphical lasso to get a precision matrix estimate \(\widehat{{\varvec{\Theta }}}_q\) for the data \({\varvec{X}}\) of interest