Location: Food and Feed Safety ResearchTitle: Gradient boosting and bayesian network machine learning models predict aflatoxin and fumonisin contamination of maize in Illinois – First USA case study
|BARNETT, KRISTIN - Illinois Department Of Agriculture|
|Rajasekaran, Kanniah - Rajah|
Submitted to: Frontiers in Microbiology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/13/2022
Publication Date: 11/10/2022
Citation: Castano-Duque, L., Vaughan, M., Lindsay, J., Barnett, K., Rajasekaran, K. 2022. Gradient boosting and bayesian network machine learning models predict aflatoxin and fumonisin contamination of maize in Illinois – First USA case study. Frontiers in Microbiology. https://doi.org/10.3389/fmicb.2022.1039947.
Interpretive Summary: Aflatoxin and fumonisin are fungal toxins that cause annual losses in the corn production sector and are a food safety concern as these toxins can cause severe health issues to humans and livestock. Although prior research has been done in modeling of mycotoxin contamination risk in European countries, our manuscript is the first one to do this research in US-corn. We modeled aflatoxin and fumonisin contamination risk of corn in Illinois, a major corn producing US state, by using machine learning approaches. In our research we used average monthly meteorological data for 14 historical years and corresponding toxin contamination data. Our models showed over 82% overall prediction accuracy for contamination risk in corn. Our research gives insights about the most influential weather features through the year that lead to higher probability of aflatoxin and fumonisin contamination at the end of the corn growing season. Furthermore, our modeling approaches can be applied to other US states which will be useful to farmers and stakeholders.
Technical Abstract: Mycotoxin contamination of corn results in significant agroeconomic losses and poses serious health hazards worldwide. This is one of the first reports of modeling for mycotoxin risk in a major corn producing US state using feature engineering and machine learning approaches. Monthly meteorological data for 14 historical years and corresponding aflatoxin and fumonisin contamination data from the state of Illinois were used to perform feature engineering in combination with mechanistic models to predict aflatoxin and fumonisin contamination risk in corn. We generated gradient boosting and bayesian network machine learning models based on monthly weather data that can predict aflatoxin and fumonisin contamination with 94% and 82% confidence, respectively. Analyses using both gradient boosting and bayesian network models for aflatoxin and fumonisin showed that meteorological and satellite-acquired vegetative index data during March significantly influenced grain contamination levels of both mycotoxin classes. Gradient boosting model prediction of aflatoxin was highly influenced by aflatoxin risk index and vegetative index in March and April. For fumonisin, precipitation in January/February and vegetative index in February showed high influence in the prediction. Temperatures during flowering time significantly impacted fumonisin contamination while aflatoxin risk index during flowering time significantly impacted aflatoxin. We suggest that creating models from mycotoxin levels and meteorological data for each specific state, rather than from generalized US data, significantly enhances the model’s accuracy. Similar modeling approaches for other crops will be very useful to farmers and stakeholders when developed and implemented.