基于近红外光谱的橙汁中酸土脂环酸芽孢杆菌的生长预测

张嘉雯; 刘佳元; 封雨桐; 孙佳怡; 周彬静; 屠康; 潘磊庆

doi:10.13386/j.issn1002-0306.2022050024

基于近红外光谱的橙汁中酸土脂环酸芽孢杆菌的生长预测

Growth Prediction of Alicyclobacillus acidoterrestris in Orange Juice Based on Near-infrared Spectroscopy

摘要

摘要: 酸土脂环酸芽孢杆菌（Alicyclobacillus acidoterrestris）是引起橙汁劣变的主要微生物，为研究酸土脂环酸芽孢杆菌在橙汁中的生长规律，利用近红外光谱获取橙汁中酸土脂环酸芽孢杆菌含量的信息，采用标准化（autoscale）、多元散射校正（multiplicative scatter correction，MSC）、标准正态变换（standard normal variate，SNV）、去趋势化（detrend）对光谱进行预处理，结合化学计量学，构建近红外光谱与酸土脂环酸芽孢杆菌含量预测模型。在此基础上，将近红外光谱转换为酸土脂环酸芽孢杆菌预测菌落数据，并采用“一步法”直接基于预测菌落数构建橙汁中酸土脂环酸芽孢杆菌的生长模型。结果表明，利用标准化进行光谱预处理建立的偏最小二乘（partial least squares，PLS）模型对橙汁中酸土脂环酸芽孢杆菌含量的预测效果相对较好，其预测决定系数（prediction determination coefficient，R_p²）与预测均方根误差（root mean square error of prediction，RMSEP）分别为0.733和0.242 lg CFU/mL，相对分析误差（relative percent deviation，RPD）为1.919。4种预测模型的均方误差（mean square error，MSE）介于0.0046~0.0300 lg CFU/mL之间；均方根误差（root mean square error，RMSE）介于为0.068~0.173 lg CFU/mL之间；赤池信息准则（akaike information criterion，AIC值）介于−66.383~−53.944之间，且Huang-full模型的3种指标相较更小，较适合描述橙汁中酸土脂环酸芽孢杆菌的生长。将近红外光谱获得预测菌落数构建的4种生长模型与平板计数法构建的生长模型分别进行相关性分析，发现4种模型的相关系数（r）均大于0.900，且Huang-full模型的拟合效果最优。所构建的模型通过准确因子（accuracy factor， A_f ）和偏差因子（bias factor， B_f ）进行验证，证实模型均具有良好的可靠性。因此，利用近红外光谱分析结合适当的化学计量学方法描述酸土脂环酸芽孢杆菌生长预测是可行的。

Abstract: Alicyclobacillus acidoterrestris is the dominant spoilage bacteria resulting the deterioration of orange juice. To simulate the growth of Alicyclobacillus acidoterrestris in orange juice, near-infrared (NIR) spectroscopy technique was used to predict the content of Alicyclobacillus acidoterrestri in orange juice. Different spectral pre-processing methods, including autoscale, multiplicative scatter correction (MSC), standard normal variate (SNV) and detrend, coupled with chemometric regression were used to build the prediction model of Alicyclobacillus acidoterrestris in orange juice by NIR spectroscopy. Based on that, the NIR predicted colony data of Alicyclobacillus acidoterrestris was used to develop the growth model of Alicyclobacillus acidoterrestris in orange juice by one-step approach. Results showed that, PLS model established by spectral pretreatment after Autoscale had relatively good prediction effect on the content of Alicyclobacillus acidoterrestri in orange juice, with the prediction determination coefficient (R_p²), root mean square error of prediction (RMSEP) and relative percent deviation (RPD) of 0.733, 0.242 lg CFU/mL and 1.919, respectively. Four different growth simulation models gave satisfactory predictions, with MSE values from 0.0046 to 0.0300 lg CFU/mL, RMSE values from 0.068 to 0.173 lg CFU/mL, AIC values from -66.383 to -53.944, respectively. Correlation analysis was performed between the four developed growth models based on the NIR prediction of colony number and the growth model constructed by plate counting method, and all of their correlation coefficients (r) were higher than 0.900. Particularly, the Huang-full model had the best ability to describe the growth of Alicyclobacillus acidoterrestris in orange juice and showed the best fitting results. Besides, the good reliability of all developed models was verified by accuracy factor ( A_f ) and bias factor ( B_f ). Accordingly, this study indicated the potential to use NIR spectroscopy combined with advanced chemometrics to describe the growth prediction of Alicyclobacillus acidoterrestris in orange juice.