Detection of adulteration of non-transgenic soybean oil with transgenic soybean oil by integrating absorption, scattering with fluorescence spectroscopy

Abstract

In this study, three distinct brands of soybean oils with varying proportions of transgenic and non-transgenic were subjected to analysis. The fluorescence intensity (F) was obtained via a fluorescence spectrophotometer, while the absorption (µ_a) and reduced scattering coefficients (µ’_s) were obtained by through a self-developed double integrating sphere (DIS) system. A quantitative detection method for the adulteration ratio based on fluorescence spectroscopy was proposed which considered the entangling effect of absorption and scattering. The method entails initially conducting principal component analysis (PCA) on the F, µ_a and µ’_s spectra in the range of 350–700 nm, thereby obtaining the first five principal components (PCs) of each kind of spectrum were obtained. Furthermore, the three brands of oil exhibited a discernible clustering tendency when subjected to a three-dimensional PCA mapping approach. The distribution positions of the three spectra in the three plots indicated that they could be considered to complement each other. Following further normalization processing, the PCs were fused and quantitative models were calibrated by using multiple linear regression (MLR), partial least squares regression (PLSR), and support vector regression (SVR). The results indicated that, in comparison to the utilisation of individual spectral characteristics, the fusion of F and µ_a can effectively mitigate the impact of fluorescence internal filtering, thereby improving the prediction accuracy. Furthermore, the combination of F, µ_a and µ’_s can effectively eliminate the interference of scattering on fluorescence, and achieve optimal prediction results. Among them, the MLR model based on F, µ_a and µ’_s could reach the best performance, with determination coefficients of calibration (R²_c) and validation sets (R²_v) reaching 0.959 and 0.947, respectively, while the root mean square error of calibration (RMSEC) and validation sets (RMSEV) were as low as 2.970% and 3.429%, respectively. In comparison, the MLR model based solely on F yielded unsatisfactory results, with R²_c and R²_v were 0.571 and 0.595. It can be concluded that it can greatly improve the accuracy of predicting the adulteration of transgenic in non-transgenic soybean oil by integrating F, µ_a and µ’_s spectroscopy.