Fig. 5. Results on the prospective test set and best models overall.

Model performance was judged from a single run (n = 1) (nuclear magnetic resonance = NMR). Source data for each bar chart can be found in the source data Excel file. A Experimental results. Color-coded by reagent-specific reactivity. Split circles imply more than one reagent functionalized that position. B MPNN_LSF (best retrospective model) predictions on the prospective test set. C Fukui predictions on the prospective test set. D F-score, accuracy, and AUROC (area under the receiver operating curve) reported for MPNN_LSF (highlighted in light blue) NMR transfer learn (Fukui) is the transfer learned model with Fukui-index augmentation. E Comparison of the best models on each of the 3 test sets (n = 1). MPNN_P450 = Best P450-only test set model. Color coding corresponds to the test set used for evaluation.