A novel approach for predicting aflatoxin B₁ production using regression models and whole-cell biosensors in moldy...

Aspergillus flavus is a major cause of post-harvest losses in maize and peanuts through aflatoxin B₁ (AFB₁) contamination, highlighting the urgent need for sensitive and scalable early detection strategies. In this study, we developed a transcriptome-guided whole-cell biosensor array by integrating eight infection-induced promoters, identified from E. coli transcriptomic responses to volatile organic compounds, into calcium alginate-immobilized bioreporters coupled with machine learning regression models. Time-resolved bioluminescence signals were used to train ensemble regressors for the first time, including XGBoost, CatBoost, and RandomForest, for quantitative prediction of infection stages and AFB₁ levels. XGBoost consistently achieved superior performance with R² values of 0.94 and 0.98 in internal validation for maize infection staging and AFB₁ quantification and maintained strong generalization in external validation with independent A. flavus isolates (R² = 0.92 and 0.91). Comparable results were observed in peanuts, where XGBoost achieved R² = 0.94 and 0.97 internally and 0.92 and 0.86 externally, confirming robustness across different substrates and fungal strains. Direct comparison with our previous biosensors constructed from 14 general stress-responsive promoters revealed that the novel biosensors based on 8 new transcriptome-guided promoters yielded markedly higher predictive accuracy, particularly under external validation conditions. Feature importance analysis revealed that early host responses, including transcriptional regulation and biofilm formation, served as key predictive features, thereby providing mechanistic interpretability not attainable with conventional optical or chemical assays. Together, these findings establish a biologically informed, non-invasive, and cost-efficient biosensing platform that integrates promoter-level transcriptomic insights with ensemble learning, offering a versatile approach for real-time aflatoxin risk assessment and scalable food safety monitoring across diverse agroecosystems.