Recently, the Grain and Oil Loss Reduction and Mycotoxin Control Innovation Team at the Institute of Food Science and Technology, Chinese Academy of Agricultural Sciences (IFST-CAAS), successfully developed a quantitative monitoring and early-warning technology for mold contamination levels and aflatoxin B1 (AFB1) concentrations in maize and peanuts. The breakthrough is based on a high-sensitivity whole-cell biosensor array combined with AI machine learning regression models. The related research findings have been published in the international journal Journal of Hazardous Materials (IF = 11.3). Sun Lu, a 2022 China–Belgium joint PhD student, and Dr. Ma Junning are co-first authors of the study. Prof. Xing Fuguo is the corresponding author. The study was financially supported by the CAAS Major Scientific Research Task (CAAS-ZDRW202414) and the National Agricultural Science and Technology Innovation Program.

Mold growth during the storage and processing of grains and oilseeds is a major cause of postharvest losses and poses a serious threat to global food security. During fungal contamination, highly toxic and carcinogenic mycotoxins are produced—among them AFB1, a potent toxin that can induce primary hepatocellular carcinoma, posing severe risks to food safety and human health. Accurate and quantitative monitoring of mold development and AFB1 levels has long been a technical bottleneck in the industry.

In this study, the research team focused on the signature volatile organic compounds (VOCs) produced by Aspergillus flavus infection in maize and peanuts. Using transcriptomic analysis, they identified eight novel inducible genes and constructed eight bioluminescent E. coli strains using their promoter sequences, forming a new whole-cell biosensor array. By integrating this biosensor array with a suite of artificial intelligence regression models—including Random Forest, XGBoost, CatBoost, Stochastic Gradient Boosting (SGB), Support Vector Machine (SVM), and Sparse Partial Least Squares Discriminant Analysis (SPLSDA)—the team achieved precise quantitative prediction of mold growth days and AFB1 concentrations in maize and peanuts.

Among these models, XGBoost, a high-performance ensemble learning algorithm based on gradient boosting, showed the best performance. In internal validation using maize samples, the XGBoost model achieved 94% prediction accuracy for mold growth days and 98% for AFB1 content. It also demonstrated strong generalization ability in external validation using independent A. flavus strains, significantly outperforming previous whole-cell biosensors based on general stress promoters. Similar results were obtained for peanuts, where XGBoost achieved 94% and 97% prediction accuracy for mold levels and AFB1 content, respectively, confirming its robustness across different matrices and fungal strains. Further feature importance analysis revealed early host response mechanisms during fungal infection.

This study presents an innovative solution for the real-time monitoring of mold contamination and mycotoxin risk in grains, providing robust technological support for safeguarding national food and feed safety.

Link: https://www.sciencedirect.com/science/article/abs/pii/S030438942502802X?dgcid=author

 

Graphical Abstract