The Research Of Near Infrared Spectroscopic Qualitative And Quantitative Analysis Of Ver A Content In Maize Based On Chemometrics

Objective:Hazardous aflatoxins?AFs?widely contaminated in agricultural products,and it is urgently requested to prognosticate the risk of AFs contamination before and during storage and transportation.The previous investigations discovered that monitoring the content of Versicolorin A?Ver A?,a precursor of AFB₁?a representative of AFs?,are meaningful in evaluation the risk of potential AFs contamination?even when AFs is undetectable?of stored grain.However,there is currently a lack of rapid and easy detection means of Ver A for stored grain.In this work,a method of detecting Ver A in maize using near-infrared spectroscopy?NIR?coupled with multivariate statistical methods.On the basic of this,Ver A has the role of early warning AFs,through the establishment of binary classifier to achieve the safety sorting for aflatoxin contamination of grain?maize?.Methods:An entire of 92 naturally contaminated maize samples,their Ver A were detected by high performance liquid chromatography?HPLC?technique,which considered to be a classic and reliable method.The pretreatment was achieved by using a one-step process,the solid-phase extraction affinity column based on Ver A aptamer.The obtained data of Ver A values by HPLC were basically used as the reference for further model building.Near-infrared spectroscopy data were collected from the same samples.The collected data sets were then processed chemometrics analysis with partial least squares?PLS?,machine learning algorithm XGBoost and support vector machine algorithms to built up the Ver A prediction model of maize sample.Based on the critical parameters of penalty parameter c?a compromise cost minimizing both of training error and model complexity?and the nuclear parameter g?defining the nonlinear mapping from the input space to some high-dimensional feature space?the models was optimized.Results:?1?The partial least-squares algorithm?PLS?model was used to predict the Ver A content of maize samples.The optimal PLS model was obtained by the second derivative pretreatment with an RPD of 1.50.?2?Based on the PLS model,the variables selection was carried out with choosing multiple scatter correction?MSC?as the pretreatment method.The optimal model was obtained with RPD of 1.54.?3?By using with the machine learning algorithm XGBoost,a quantitative model of Ver A of maize samples was successfully established with RPD of 5.4.That high RPD indicates that the model is adequate for the quality control of grain?maize?.?4?A binary classfier for assessing the security of maize?risk of further contamination by AFs during storage and tranportation?was established based on Ver A content using with the support vector machine algorithm.The sensitivity of the classfier was found to be 100.00%with the specificity of 52.94%and the accuracy rate of 74.19%,when used to evaluate 31 maize samples.Conclusion:Ver A determination with NIR technique was established by coupling used with chemometrics.With variables selection and optimization with conference of parameters of c and g,the finally obtained quantitative model of Ver A of maize was satisfied in quality control of grain?maize?.Because of the fast,easy carrying-out,low-cost with accurate of the established NIR method,this method is applicable with in-situ analysis.The established secutity classifier is a convenient tool?required of grinding?in early-knowing the risk of AFs contamination before and during the storage and transportation of grain.