Graph Neural Networks For Improved Interpretability And Efficiency