'Building-block coupling effect' in dimethylpyrazine formation: Mechanistic insights and predictive modeling.

This study investigated the patterns and mechanisms of dimethylpyrazines (DMPs) formation based on hydroxyacetone (HA) and established a machine learning predictive model. We found the HA-ammonium acetate system as optimal, and the formation of DMPs followed a zero-order kinetic model. Under the proposed 'building-block coupling effect', DMP isomers are selectively generated through heterotypic C₃-C₃ coupling to yield 2,6-dimethylpyrazine, asymmetric C₄-C₂ assembly leading to 2,3-dimethylpyrazine, and homotypic C₃-C₃ coupling to form 2,5-dimethylpyrazine. The critical intermediate 2-amino-3-hydroxy-4-oxopentanoic acid was tentatively identified in the system, which is involved in three reaction pathways. Machine learning models were further applied to predict DMP isomers content, with the ExtraTrees regressor showing the highest predictive accuracy (R2 ≈ 0.98). Feature importance and SHAP analyses consistently identified temperature as the primary driving factor. Correlation analysis confirmed pathway-specific dependencies on reaction conditions. This work refines the mechanistic for DMP formation and provide an integrated mechanistic and data-driven framework for controlling DMPs formation and tailoring pyrazine profiles for rational flavor design.