Fig 1

(A) The amide synthesis reaction was used as a model for data collection, including four dimensions of aromatic amine, carboxylic acid, solvent, and coupling reagent. R1: H, OCH3, NO2, CF3, Cl, Br, CH3; R2: CH3, C2H5, CH = CH2, C3H8, C5H11, CH2CHCOCH3; R3: H, Cl, CH3, OCOCH3; the corresponding molecular structures for aromatic amine and carboxylic acid are provided in Fig. S1 and the detail information for solvent, and coupling reagent can be found in Table S1 (Supporting Information). (B) The high-throughput experiments' reaction outcomes were obtained from the reactions of aromatic amines, carboxylic acids, organic solvents, and coupling reagents in different combinations. The first line of horizontal coordinate in the plot is the aromatic amine, the second line is the organic solvent, and the last line is the coupling reagent (represented by Roman numerals, Ⅰ: HBTU, Ⅱ: PyBOP, Ⅲ: DPP-Cl, Ⅳ: BOP-Cl). The longitudinal coordinate is the type of carboxylic acid, the first six are fatty acids, and the last six are aromatic acids (represented by Arabic numerals, 1: Acetic acid, 2: Propionic acid, 3: Acrylic acid, 4: Butyric acid, 5: N-hexanoic acid, 6: Levulinic acid, 7: Benzoic acid, 8: p-chlorobenzoic acid, 9: p-toluic acid, 10: Niacin, 11: Isonicotinic acid, 12: Acetylsalicylic acid)."

Fig 2

(A) The process of calculation, evaluation, and selection of the molecular atoms and vibrational descriptors is demonstrated by using aromatic amine molecules as an example. SA: surface area; V30 to V40, vibrational modes of atoms; *, the atoms in the structure shared by the same molecules in the reaction participants (Note: The detailed procedure is referenced from online instructions 34). (B) Data structure diagram for machine learning. (C) Plot of the basic operational process of ML."

Fig 3

(A) Evaluation plot of six algorithmic models. The horizontal coordinate is the predicted conversion of the model, and the longitudinal coordinate is the experimentally measured conversion. Blue solid line: partially weighted regression best-fit curve; black dashed line: y = x line. The slope of the fitted line for these points should be close to 1 and the intercept close to 0. (B) The evaluation plot versus different ratios of training sets. The prediction performance of the random forest model was evaluated using small sample data. The training sets were selected randomly from the original training data."

Fig 4

(A) The most important 11 descriptors given by the trained random forest model are determined by rescaling the values of the given descriptors and measuring the percentage increase in mean squared error when retraining the model. E: energy; HOMO: highest occupied molecular orbital; V: the vibration of an atom in a reacting molecule. *, the atoms in the structure shared by the same molecules in the reaction participants. (B) An analytical plot of the molecular structures of representative reactants selected from aromatic amines, carboxylic acids, and solvents, combined with featured descriptors produced by the model."

Fig 5

(A) Aromatic amine species in the training set (1 to 9) and the unknown aromatic amine species in the test set (10 to 12). (B) The high-throughput experiments' reaction outcomes were obtained from the reactions of aromatic amines, carboxylic acids, organic solvents, and coupling reagents in different combinations. For the detailed explanation, please refer to Fig. 1B."

Fig 6

(A) The predictive performance of the random forest for the unknown aromatic amine species; a–c, the predictive performance of the random forest model before optimization; d–f, the predictive performance of the random forest model after optimization. (B) Model predictive performance plot versus reaction data of unknown aromatic amines with different fatty acids added to the training set."

Fig 7

(A) The prediction effect of the random forest model trained by one-hot coding approach to characterize the descriptors of all reaction participants. (B) The prediction effect of the random forest model trained by the hybrid method (one-hot/DFT-mixing), The descriptors of the amines in the reaction were using one-hot encoding, and the descriptors of the rest participants were using DFT."