Fig. 9

Dissolution rate predicted by a “topology-blind” machine learning (Model III) and b “topology-informed” machine learning (Model IV) as a function of the measured dissolution rate—wherein the dissolution data of Glasses A ((Na₂O)_0.25(Al₂O₃)_x(SiO₂)_0.75–x, training set) are used as a training set to predict the dissolution kinetics of Glasses B ((Na₂O)_x(Al₂O₃)_x(SiO₂)_1–2x, test set). c Distribution of prediction error for the training (solid line) and test sets (dash line) offered by Models III (black) and IV (red), respectively. The error is defined as the difference between predicted and measured dissolution rate