Fulkerson Andrew, Bayram Ipek, Decker Eric A, Parra-Escudero Carlos, Lu Jiakai, Corvalan Carlos M
Transport Phenomena Laboratory, Department of Food Science, Purdue University, West Lafayette, IN 47906, USA.
Department of Food Science, University of Massachusetts, Amherst, MA 01003, USA.
Foods. 2025 Jun 18;14(12):2135. doi: 10.3390/foods14122135.
Accurately modeling the degradation of food antioxidants in oils is essential for understanding oxidative stability and improving food shelf life. This study presents an innovative machine learning approach integrating neural differential equations and sparse symbolic regression to derive a parsimonious differential equation for myricetin degradation in stripped soybean oil. Despite being trained on a small experimental dataset, the model successfully predicts degradation trends across a wide range of initial concentrations and extrapolates beyond the learning data. This capability demonstrates the robustness of machine learning for uncovering governing equations in complex food systems, particularly when experimental data is scarce. Our findings provide a framework for improving antioxidant efficiency in food formulations.
准确模拟油脂中食品抗氧化剂的降解对于理解氧化稳定性和延长食品保质期至关重要。本研究提出了一种创新的机器学习方法，该方法整合了神经微分方程和稀疏符号回归，以推导脱去大豆油中杨梅素降解的简洁微分方程。尽管该模型是在一个小的实验数据集上进行训练的，但它成功地预测了各种初始浓度下的降解趋势，并能在学习数据范围之外进行外推。这种能力证明了机器学习在揭示复杂食品系统中的控制方程方面的稳健性，特别是在实验数据稀缺的情况下。我们的研究结果为提高食品配方中抗氧化剂的效率提供了一个框架。
