Describir: Multimodal graphical models via Group Lasso