Detection Of Clostridium Botulinum By Confocal Raman Microspectroscopy

The botulinum toxin is released during the propagation process,it is the strongest biological toxinit in the natural and synthetic toxins,and it is often found in the canned food,vacuum packaging,sealed pickled food,etc.The traditional detection method is to cultivate and detect laboratory flora.However,it is complex,the detection cycle is generally more than 10days,which is not conducive to the timely control of the risk.Therefore,it is of great significance to find a fast,non-destructive,and simple method.The Confocal Raman microspectroscopy（CRM）is a method of rapid,reliable,lossless and no sample pretreatment,it can overcome the problem of Raman signal weakness,and detect single bacterium in theμm level at high resolution.This paper aims to establish a rapid identification of the three levels（Bacteria,clostridia and serotype）based on the CRM and chemometrics.532 nm laser was selected as excitation light,the objective lens is 50 times,the output power is 40 m W,the spectral resolution is 0.6 cm^-1,the grating is 1200 mm,the spectral collection range is 600～1800 cm^-1,the number of scan is 6,the integration time is 10 s.At the bacterial level,6 kinds of food-derived pathogenic bacteria were chose（Salmonella typhimurium,Shigella flexneri,Listeria monocytogenes,Vibrio cholerae,Staphylococcus aureus and Clostridium botulinum）.They were investigated based on four spectrum pretreatment methods,spectral analysis,unsupervised learning and supervised learning.Principal Component Analysis（PCA）was used to study the differences of bacteria and the dimensionality reduction of the data,there are three kinds of supervised learning models,including decision tree（DT）,artificial neural network（ANN）and Fisher discriminant analysis（FDA）.The effect of four spectrum pretreatment methods on models were discussed.The results showed that by comparing the characteristic peaks and peak ratio values can distinguish between different pathogenic bacteria,but the automation capacity is poor,and the detection of the large sample cannot be met.The preliminary classification of the sample can be achieved in the non-supervised PCA model.In the supervisory analysis,the CART algorithm in the DT model combined with the pre-treatment method SG 1st der is the best DT model.The correct rate of training set was 100.00%,and the correct rate of the verification set was 98.1%（157/160）.The MLP algorithm in the ANN model combines the pre-treatment method MSC is the best ANN model,the correct rate of the training set is 100.0%,the CRR of the verification set is 99.4%（159/160）.FDA model combines SG 1st der pretreatment Methods is the best FDA model,the correct rate of training sets and verification sets is 100.0%,and the performance is best in three models.At the clostridia level,three kinds of foodborne pathogenic bacteria were selected（Clostridium botulinum,Clostridium perfringens,Clostridium difficile）.At the serotype level,two kinds of botulinum clostridia were selected（type A and Type B）,They were associated with the human.Based on four different spectral pretreatment methods,PCA and Principal component analysis-Linear discriminating analysis（PCA-LDA）were used.The results showed that there is a difference in model accuracy under different pretreatment methods.In the three kinds of the samples in the same bacterium but different clostridia,based on SG 1st der,the sample could be completely distinguished in the PCA model.In the three kinds of the samples of different serotypes,based on the SNV,the sample could be mostly distinguished in the PCA model,but the problem of the sample overlap was not avoided,and in the PCA-LDA model,the sample was distinguished when not using any preprocessing method.The samples were obviously distinguish,and the correct rate of cross-confirmation was 100%.In summary,the CRM in this paper could realize the distinction between the three levels of Bacteria,clostridia and serotype compared to traditional detection methods.It has the advantages of convenient operation and rapid operation,and could provide a reference for the investigation of food safety criminal cases and the research on food safety supervision.