Machine Learning Identifies a Parsimonious Differential Equation for Myricetin Degradation from Scarce Data.

Area of Science:

• Food Science
• Chemical Kinetics
• Machine Learning

Background:

• Accurate modeling of food antioxidant degradation is crucial for oxidative stability and shelf-life prediction.
• Understanding degradation kinetics informs strategies to enhance food product longevity.

Purpose of the Study:

• To develop a machine learning model for predicting myricetin degradation in stripped soybean oil.
• To derive a parsimonious differential equation governing myricetin degradation using limited experimental data.

Main Methods:

• Integration of neural differential equations and sparse symbolic regression.
• Training a machine learning model on a small experimental dataset of myricetin degradation.

Main Results:

• The developed model accurately predicts myricetin degradation trends across various initial concentrations.
• The model demonstrates extrapolation capabilities beyond the training data, indicating robustness.
• A parsimonious differential equation governing myricetin degradation was successfully derived.

Conclusions:

• Machine learning offers a robust approach for uncovering governing equations in complex food systems, especially with scarce data.
• The findings provide a framework for optimizing antioxidant efficiency in food formulations.
• This methodology can be applied to other food systems requiring stability modeling.