Simultaneous formation and prediction of bound and free heterocyclic amines in roasted beef patties using integrated generalized kinetics and fusion machine learning.

The formation patterns of heterocyclic amines (HAs), carcinogens formed during meat cooking, in real food matrices remain incompletely understood. This study developed generalized kinetic and machine learning (ML) models to characterize bound and free HAs in roasted beef patties under different heating conditions. Bound HAs predominated, peaking at 225 °C (662.31 ± 69.30 ng g-1) but slightly decreasing at 250 °C, while free HAs followed a linear pattern, peaking at 250 °C (27.80 ± 3.28 ng g-1). Arrhenius analysis revealed higher activation energies for bound HAs, indicating greater temperature sensitivity. Rapid glucose and creatine degradation, along with protein and lipid oxidation, were closely associated with HAs accumulation. The artificial neural network (ANN)-extreme gradient boosting (XGBoost) fusion model achieved high predictive accuracy (test R2 > 0.9175), reproducing time-temperature-HAs complex nonlinear relationships. This study integrates kinetic modeling with ML, providing a practical framework for optimizing processing conditions and mitigating HAs exposure.