Beyond normality: Learning sparse probabilistic graphical models in the non-Gaussian setting