Detection Of Fungal Contamination In Several Agricultural Products Based On Electronic Nose And HS-GC-IMS

Rice,peanut and wheat are important agricultural crops consumed in China and around the world.However,they are prone to be contaminated with spoilage or toxigenic fungi in the field as well as during the storage.Conventional methods for detecting fungal contamination are generally time-consuming and sample-destructive,making them impossible for large-scale nondestructive detection and real-time analysis.Moldy spoilage of agricultural products is a complex procedure from microorganisms by consuming large amounts of nutritional substances,accompanied by the odor changes.Thus,the advent of the electronic nose(E-nose)and headspace-gas chromatography ion-mobility spectrometry(HS-GC-IMS)enabled early identification of mold contamination through the changes in odor.In this study,it was aimed to establish the early discriminative and quantitative models of fungal contamination in rice,peanuts and wheat by using E-nose and HS-GC-IMS technologies.The changes of volatiles in fungi-contaminated agricultural products were analyzed,the mechanism of E-nose and HS-GC-IMS technologies to distinguish different fungal species and predict the fungal colony counts was preliminarily explained and verified by practical application.The main research contents and results are as follows:1)The detection of Aspergillus spp.contamination in rice kernels was investigated by electronic nose(E-nose)and gas chromatography-mass spectrometry(GC-MS).The volatiles released by rice kernels infected by different aspergillus fungi were analyzed,the response signals of E-nose in each treatment group were analyzed,and logistic model was applied to simulate the growth of individual fungus.Principle component analysis(PCA),partial least squares regression(PLSR),back-propagation neural network(BPNN),support vector machine(SVM)and learning vector quantization(LVQ)were employed for qualitative classifcation and quantitative regression.It could be found that there was a signifcant correlation between the volatile compounds(such as 1-octyl alcohol and tetradecane)and total amounts/species of fungi,as well as E-nose signals.X-axis barycenters(PC1 scores)of PCA after E-nose signals visualized were significantly correlated with fungal counts,and PC1 scores could be employed to simulate the growth of A.candidus,A.fumigatus,and A.clavatus.The classification model based on E-nose signals combined with BPNN had higher classification accuracy for the fungi-infected rice during the storage of 2 days.2)The performances of HS-GC-IMS and E-nose for the classification of rice kernels with different Aspergillus spp.infection levels(healthy or moldy)were investigated.Principal component analysis(PCA),K-nearest neighbor(KNN)and PLSR were employed to compare the performances of two technologies.Meanwhile,the difference of volatiles detected by HS-GC-IMS was analysis.The results indicated that both HS-GC-IMS and E-nose technologies combined with KNN achieved good classification accuracies,which were suitable for rapid screening of fungal contamination in rice samples.In the PLSR prediction models,both HS-GC-IMS and electronic nose technologies show prediction performances,whereas better prediction performance was achieved by HS-GC-IMS.On the first day of storage,the species and contents of volatiles released from rice samples infected by different fungi varied greatly.The peak intensities of 2-pentanone monomer and dimer were the highest in rice samples infected with A fumigatus,while the peak intensities of 2-butanone monomer,2-butanone dimer and pentanol monomer were the highest in the samples infected with A.clavatus3)Rapid determination of potential aflatoxigenic fungi contamination on peanut kernels based on HS-GC-IMS coupled to fluorescence spectroscopy were investigated.Data-level and feature-level fusion strategies were introduced to integrate HS-GC-IMS and fluorescence spectra,and the performance of classification and regression models constructed by different data fusion methods were compared.The results indicated that the application of feature-level data fusion using first 10 PCs coupled with orthogonal partial least squares discriminant analysis(OPLS-DA)offered the best classification rate for aflatoxigenic and non-aflatoxigenic fungal infection on peanut samples.In addition,the HS-GC-IMS-based method allowed the two classes to be distinguished after only 3 incubation days.The prediction performance of PLSR model based on the feature-level fusion using first 10 PCs was also the best,and the performance of feature-level data fusion model was better than that using single data.Moreover,eight substances released from peanut kernels contaminated with aflatoxin-producing and non-aflatoxin producing fungi were significantly different,which could be served as biomarkers to distinguish aflatoxin-producing species contamination in peanut kernels at early stage4)HS-GC-IMS technology combined with two chemometric strategies including nontargeted spectral fingerprinting and targeted specific markers were employed to distinguish different fungal species in wheat samples,and to quantitatively predict the colony counts of A flavus,A.niger,A.tamarii,and P.islandicum in wheat kernels.Genetic algorithm optimized support vector machine(GA-SVM),random forest(RF)and PLSR algorithms were used to establish the classification and regression models of fungi-infected wheat samples,and their performances were compared.In addition,the validation experiment of detecting the percent A flavus infection presented in simulated field samples was carried out.Results showed that satisfying results for the differentiation of fungal species were obtained based on both strategies by GA-SVM,and better values were obtained based on nontargeted spectral fingerprinting strategy.Likewise,the GA-SVM model based on nontargeted spectral fingerprinting strategy achieved the best prediction performances of colony counts in fungal infected samples.The results of validation experiment showed that GA-SVM models based on nontargeted spectral fingerprinting strategy could still provide satisfactory classification and prediction performances for percent A.flavus infection presented in simulated field samples at day 4.The results indicated the feasibility of HS-GC-IMS-based approaches for the early detection of fungal contamination in wheat kernels.