Fig. 5: Comparison of test MAE results on QM9 datasets.

This figure compares the Test MAE of the QM9 datasets among various models: DenseGNN, DenseNGN, MEGNet, SchNet, enn-s2s, ALIGNN, and DimeNet++ (DN++). Among these, DenseNGN is a nested graph network that incorporates angle information, allowing the model to learn more accurate local chemical environment information. DenseNGN implements the DCN strategy within the nested graph networks framework of coNGN. The best results are highlighted in bold. A dash (-) indicates that the MAE results were not provided for certain comparisons.