Fig. 1. Machine learning prediction of synthetic performance and molecular property.

a Representative strategy of synthetic prediction by concatenating the molecular encodings of reactant 1 (orange), reactant 2 (blue), additive (green), and product (yellow). b Previous work of quantum chemistry-augmented graph model for synthetic prediction using the WLN (Weisfeiler-Lehman network) encoder and QM (Quantum Chemistry) descriptors. The green background represents the input compounds and the architecture of the model. The orange background represents four quantum chemical descriptors. α, β, γ, and α’ indicate the reaction sites. c Chemical knowledge-based design of graph model for synthetic performance prediction (this work).