In an article published on July 11, 2023, in the journal Food and Chemical Toxicology, Shan-Shan Wang from Kaohsiung Medical University and National Health Research Institutes, Kaohsiung, Taiwan, and co-authors present a nonlinear machine learning method for modeling the migration potential of food contact chemicals (FCCs) from packaging materials into foods, i.e., determining packaging-food partition coefficients (log Kpf).
The scientists used nine existing models to build theirs as well as a data set of 1847 records of their previous work to train and validate their model. One record included information on the tested chemical, the material type of the packaging (e.g., polyethylene, silicone), ethanol equivalence (i.e., ethanol concentration), and temperature as factors that may influence migration. Concerning chemical information, the records only the compound’s included octanol-water partition coefficient (log Kow) even though other chemical characteristics can influence migration. Therefore, Wang et al. used “molecular descriptors and fingerprints” as additional chemical descriptors in their current model. A background article by the Food Packaging Forum on migration modeling explains factors influencing migration and provides an overview of the different approaches to mathematically modeling the migration of chemicals (FPF reported).
The authors reported that their nonlinear prediction model outperforms linear models which don’t consider the nonlinear relationship between the ethanol equivalent and log Kpf. Furthermore, it could “leverage multiple models to make a more accurate prediction.” However, they also pointed out that their generic model can only be applied to a limited set of compounds because necessary information to run the model is often missing (e.g., material types, temperature, log Kow). If all that data is available their model “showed high performance.”
Wang and co-authors applied their model to evaluate the migration of FCCs of high concern due to being toxic to reproduction, development, or carcinogenic. The 47 carcinogenic FCCs were prioritized in a previous study using in silico models (FPF reported). The researchers reported that the carcinogen diethyl sulfate (CAS 64-67-5) “tends to migrate into all five food simulants” tested which was also the case for methanol (CAS 67-56-1) and ethylene oxide (CAS 75-21-8) that are associated with reproductive and developmental toxicity. The remaining FCCs of concern “showed distinct migration potential under different ethanol equivalent levels.”
The authors emphasized that combining their approach of prioritizing FCC of concern (FPF reported) with the proposed migration predictions “could be a useful method for accelerating the risk assessment FCCs.” As well as Wang et al., also a study published in 2022 evaluated a model to predict chemical migration and found it suitable for regulatory applications (FPF reported).
In an article published on July 11, 2023, in the Journal of Food Engineering, Francesco Petrosino and co-authors from the University of Calabria, Rende, Italy, proposed two approaches for modeling migration of selected chemicals from packaging and additionally from processing equipment.
As opposed to Wang et al. whose model was trained with and is thought to be applied to multiple compounds, Petrosino and colleagues selected two specific cases: (1) the migration of the additive octadecyl-3-(3,5-di-tert-butyl-4- hydroxyphenyl) (CAS 2082-79-3) from a 100 µm thick polyethylene film into a refrigerated sandwich over time and (2) of chromium (CAS 7440-47-3), magnesium (CAS 7439-95-4), and nickel (CAS 7440-02-0) from an industrial stainless-steel piece used in production equipment into an aqueous solution at 60°C. The first they modeled in MATLAB based on the geometry of the food object and the polymer characteristics while the software Comsol Multiphysics was used to implement the second model. To validate the latter, the researchers further performed experimental migration testing in 3% acetic acid (60 °C, 5 and 30 min).
The modeling showed that the migration of octadecyl-3-(3,5-di-tert-butyl-4- hydroxyphenyl) into the sandwich was within its specific migration limit (SML) of 6 mg/kg (according to European regulation) until day 10 (5.9 mg/kg food) but exceeded it after. The authors further found “a very good agreement between the experimental data and the model prediction” for the migration from the steel article. While nickel migrated in concentrations of 0.290 mg/L within 30 min, magnesium and chromium did in levels of 0.046 and 0.226 mg/L, respectively.
References
Wang, S.-S. et al. (2023). “Machine learning for predicting chemical migration from food packaging materials to foods. ” Food and Chemical Toxicology. DOI: 10.1016/j.fct.2023.113942.
Petrosino, F. et al. (2023). “Modeling of specific migration from food contact materials.” Journal of Food Engineering. DOI: 10.1016/j.jfoodeng.2023.111652
