Enhancing the Predictive Performance of Bayesian Graphical Models