Machine learning modeling for predicting growth dynamics of Listeria monocytogenes using ComBase database: A...

Area of Science:

• Food microbiology
• Computational biology
• Predictive modeling

Background:

• Traditional models for Listeria monocytogenes struggle with complex environmental interactions and nonlinear responses.
• Accurate prediction of pathogen behavior is crucial for ensuring food safety.

Purpose of the Study:

• To develop an advanced machine learning framework for Listeria monocytogenes prediction.
• To integrate stress physiology principles and interpretable mechanisms for improved accuracy.

Main Methods:

• Utilized a curated dataset of 2632 observations from ComBase under various conditions.
• Developed a stress physiology-informed feature engineering module and a SHAP-based interpretability module.
• Combined these with XGBoost for predictive modeling.

Main Results:

• Achieved R² values of 0.90 for growth and 0.88 for inactivation, outperforming baseline models.
• Identified key factors influencing different growth phases, such as Suitability Scores and Water Activity (Aw).
• Demonstrated significantly improved prediction accuracy and generalization across diverse food matrices.

Conclusions:

• The stress physiology-informed machine learning framework enhances predictive accuracy for Listeria monocytogenes.
• The model shows superior performance and generalization capabilities compared to previous methods.
• This framework offers a robust tool for improving food safety assessments.