Early Warning and Risk Prediction of Aspergillus flavus Contamination and Aflatoxin B₁ Production in Maize and Peanut Kernels Using Volatile and Whole-Cell Biosensors

Mycotoxin contamination remains a major challenge for food and feed safety, particularly in postharvest systems where fungal growth may occur before visible spoilage or measurable toxin accumulation becomes evident. Among toxigenic fungi, Aspergillus flavus is of particular concern because of its ability to infect important agricultural commodities and produce aflatoxin B₁ (AFB₁), one of the most toxic and strictly regulated mycotoxins. This thesis aimed to develop early-warning and risk-monitoring strategies for A. flavus contamination by focusing on volatile organic compounds (VOCs), whole-cell biosensing, and regression-based prediction.
First, the thesis reviewed the occurrence and control of major mycotoxins in Asian food and feed systems, highlighting the need for preventive monitoring rather than late-stage intervention. It then examined VOC-based detection as a promising non-destructive approach for identifying postharvest deterioration. Building on this framework, experimental work was conducted to characterize early volatile changes in maize kernels infected with A. flavus. By integrating electronic nose analysis, gas chromatography–ion mobility spectrometry, gas chromatography–mass spectrometry, and exploratory GC×GC–MS screening, distinct VOC profile changes were observed during the asymptomatic stage of infection. These changes allowed infected maize samples to be differentiated from controls before visible mold development, indicating that early fungal colonization is accompanied by detectable metabolic disturbance in the headspace environment.
The second experimental part extended this early-warning concept toward toxin-related risk estimation. A transcriptome-guided whole-cell biosensor array was constructed by selecting promoter elements from Escherichia coli genes responsive to VOCs emitted by A. flavus-infected maize and peanut kernels. When combined with machine-learning regression models, especially XGBoost and CatBoost, the biosensor responses enabled accurate prediction of infection stages and AFB₁ accumulation in both commodities. The use of transcriptome-informed promoter selection improved not only predictive performance but also biological interpretability, linking biosensor outputs to cellular responses induced by mold-associated volatile environments.
Overall, this thesis demonstrates that VOC changes can provide early-warning information during asymptomatic A. flavus infection, while transcriptome-guided biosensors coupled with regression models can further translate these biological signals into quantitative estimates of infection progression and AFB₁-related risk. The proposed platform is better positioned as a front-line screening and risk-triage tool for repeated, low-cost, and minimally invasive monitoring in postharvest systems. Further validation under complex storage conditions, mixed fungal backgrounds, and broader commodity systems will be necessary before practical deployment. Nevertheless, this work provides a biologically informed approach for moving food safety monitoring from endpoint detection toward earlier, risk-oriented intervention.