Figure 3.

RMSE (top) and R² (bottom) values of predictions obtained by MDM and GNN for different data sets using different data set splitting methods. The data sets on a white background were evaluated using cross-validation and those on a blue background were evaluated using a fixed train-test split. Letters (a–i) correspond to the previous work that report current/previous best prediction accuracies for these data sets. (a): Tang et al.,¹⁸ (b): Lusci et al.,¹⁵ (c): Wu et al.,⁶⁷ (d): Boobier et al.,⁵⁴ (e): Cui et al.,³⁸ (f): Boobier et al.,⁵¹ (g): Llinas et al.,³⁹ and (h): Hopfinger et al.⁶⁸