Recent advances in interpretable machine learning using structure-based protein representations