Research and Economic Development, Office of


Date of this Version

Spring 4-11-2022

Document Type



Copyright © 2022 by the authors


Prevention of the growth of harmful microorganisms in food products is an important requirement for ensuring food safety and quality. Mathematical models to predict the quantitative changes in microbial populations in food to the variations of environmental conditions are useful tools in this regard. Current approaches that use empirical formulation generate arbitrary forms of model equations, impeding systematic analysis towards identifying key factors governing microbial growth and inactivation in food. To address this challenge, we present a data-driven modeling pipeline that enables automatic discovery of model equations (through parsimonious selection of relevant terms in a pre-built library) without having to assume specific functional forms of equations a priori. Through case studies using literature data, we showed how one can systematically build and analyze microbial inactivation models using the pipeline to predict the changes in D-value (i.e., the time taken to reduce microbial population to 10% of the initial level) as a function of given input variables. We used Akaike information criterion to avoid overfitting without hurting model accuracy. The final model was integrated with global sensitivity analysis to evaluate the impacts of individual factors on target variables. We highlight that, besides enhanced performance in data fit, the ability to generate models of varying complexity by accounting for a trade-off between accuracy and interpretability is a unique feature with our approach, not shared by empirical methods. Due to its generality, the pipeline presented in this work is readily applicable to many other related non-linear systems without being limited to microbial inactivation datasets