Modeling and Investigation of Deoxynivalenol Reduction in Wheat Flour After Cold Atmospheric Plasma Treatment Using Artificial Neural Networks

The aim of this study was to explore the effectiveness of cold atmospheric plasma (CAP) treatments for reducing the deoxynivalenol (DON) content in spiked white wheat flour samples containing 750 μg kg^-1 DON. The flour samples were treated with plasma generated in air for durations of 30 s, 60 s, 90 s, 120 s, 150 s, and 180 s and at four distances from the cold plasma source: 6 mm, 21 mm, 36 mm, and 51 mm. An artificial neural network (ANN) model with three layers utilizing the Broyden-Fletcher-Goldfarb-Shanno (BFGS) iterative algorithm was developed to predict the reduction in deoxynivalenol (DON) content, moisture content, and temperature in wheat flour samples following cold atmospheric plasma (CAP) treatment. The model accounted for two key variables: the distance from the plasma source and the treatment duration. The ANN model exhibited excellent predictive performance, achieving coefficient of determination (<i>r</i>²) values of 0.999, 0.996, and 0.996 for DON reduction, moisture content, and temperature, respectively, during the training phase. The ANN model successfully identified the experimental optimal CAP conditions (51 mm distance and 150 s treatment), resulting in a 71% reduction in DON content. Multi-objective optimization (MOO) using the ANN further predicted the same level of reduction but at 168 s while maintaining acceptable moisture and temperature levels, representing the model-derived optimal treatment within the investigated design space. The study highlights the potential of ANNs to model complex relationships and optimize CAP treatment for efficient mycotoxin reduction in wheat flour.