Rapid detection of the authenticity and adulteration of sesame oil using excitation-emission matrix fluorescence and chemometric methods

Journal Pre-proof Rapid detection of the authenticity and adulteration of sesame oil using excitationemission matrix fluorescence and chemometric methods Yuan-Yuan Yuan, Shu-Tao Wang, Jun-Zhu Wang, Qi-Cheng, Xi-Jun Wu, De-Ming Kong PII:

S0956-7135(20)30061-X

DOI:

https://doi.org/10.1016/j.foodcont.2020.107145

Reference:

JFCO 107145

To appear in:

Food Control

Received Date: 26 October 2019 Revised Date:

15 January 2020

Accepted Date: 27 January 2020

Please cite this article as: Yuan Y.-Y., Wang S.-T., Wang J.-Z., Qi-Cheng , Wu X.-J. & Kong D.M., Rapid detection of the authenticity and adulteration of sesame oil using excitation-emission matrix fluorescence and chemometric methods, Food Control (2020), doi: https://doi.org/10.1016/ j.foodcont.2020.107145. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2020 Published by Elsevier Ltd.

Credit Author Statement: Yuan-Yuan Yuan: Conceptualization, Methodology, Software, Writing - Original Draft, Writing - Review & Editing. Shu-Tao Wang: Validation, Data Curation, Resources. Jun-Zhu Wang: Writing - Review & Editing. Qi Cheng: Formal analysis, Writing - Review & Editing. Xi-Jun Wu: Resources, Data Curation, Investigation. De-Ming Kong: Supervision.

1

Rapid detection of the authenticity and

2

adulteration of sesame oil using

3

excitation-emission matrix fluorescence and

4

chemometric methods

5 6 7 8

Yuan-Yuan Yuan, Shu-Tao Wang*, Jun-Zhu Wang, Qi-Cheng, Xi-Jun Wu, De-Ming Kong Measurement Technology and Instrument Key Lab of Hebei Province, Yanshan University, Qinhuangdao 066004, China * Correspondence: [email protected]

9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Abstract: Sesame oil (SO) is a high-quality oil that is more expensive than other edible oils, and therefore becomes a target of economically motivated adulteration. An approach based on excitation-emission matrix (EEM) fluorescence and chemometric methods was applied for the rapid classification and determination of the authenticity of SO. First, a five-factor alternating trilinear decomposition (ATLD) model roughly completed the characterization of the fluorescent components in the edible oil samples, providing meaningful chemical information. Then, four chemometric methods, including linear discriminant analysis (LDA), partial least squares–discriminant analysis (PLS-DA), support vector machine (SVM) and unfolded partial least-squares discriminant analysis (UPLS-DA), were used to establish the models for the classification of SO and other edible oils (Model 1), and determine the authenticity of SO and adulterated SOs (Model 2). All models achieved good classification results. The combination of the second-order calibration algorithm (ATLD) and the pattern recognition algorithm (LDA, PLS-DA, or SVM) not only achieved the characterization of the components in edible oils but also realized the rapid detection of adulterated SOs. The proposed method is rapid, accurate, requires a simple sample pre-treatment and can be used to determine the authenticity and adulteration of high-quality edible oils.

25 26 27

Keywords: Excitation-emission matrix fluorescence, Sesame oil adulteration, Alternating trilinear decomposition, Chemometric methods, Classification

28

1.

29 30 31 32 33 34 35 36 37 38 39 40 41

Sesame oil (SO) is a high-quality oil with high nutritional value that is rich in unsaturated fatty acids, sesamol, and vitamin E, and has been attracting interest among consumers (Elleuch, Besbes, Roiseux, Blecker, & Attia, 2007). Unlike any other edible oil, SO possesses a pleasant odour and good taste; therefore, it is used as a flavour enhancer in many Asian and African countries (Anilakumar, Pal, Khanum, & Bawa, 2010). Others reported health benefits of SO, such as reducing high blood pressure, lowering hyperglycaemia and improving the plasma lipid profile (Devarajan et al., 2016). With these merits, SO is more expensive than other edible oils and therefore becomes a target of economically motivated adulteration (Cserháti, Forgács, Deyl, & Miksik, 2005). Chinese national standards (GB/T 8233-2018) clearly stipulate that SO should not be mixed with other edible oils and non-edible oils; no essence or perfume should be added. However, some unscrupulous traders usually achieve SO adulteration by adding less expensive edible oils or directly adding sesame oil essence (SOE) to cheaper edible oils. The most common cheap edible oils present in

Introduction

1

42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92

adulterated SO are mainly corn oil (CO), soybean oil (SBO), rapeseed oil (RSO), cottonseed oil and other oils (Lee, Lee, Kim, Kim, & Kim, 2001), which are much cheaper (8–30 times) than SO. In particular, SOE is a chemically synthesized food additive that mainly contains ethyl maltol (2-ethyl pyromeconic acid), cheap edible oils, etc., and has the characteristics of strong volatility and rich aroma (Yue et al., 2012), similar to the special aroma of SO. However, a high intake of SOE in the diet may cause vomiting, headaches, nausea, even affect liver and kidney functions (Ni, Zhang, & Kokot, 2005). In general, SO adulteration for economic motivation is a potential threat harming consumer rights and human health. Thus, excellent analytical techniques are required to discriminate the difference between authentic SO and adulterated SOs. Over the years, various techniques for detecting edible oil adulteration have been developed. The traditional techniques include a separation step, typically using gas chromatography (GC) (Peng et al., 2015) or liquid chromatography (LC) (Lee, Su, Lee, & Lin, 2013); however, these techniques are time-consuming due to the lengthy sample preparation procedures required. Additionally, an electronic nose apparatus (González Martı́n, Luis Pérez Pavón, Moreno Cordero, & Garcı́a Pinto, 1999) and DNA-based markers (Vietina, Agrimonti, & Marmiroli, 2013) have been developed to detect the authenticity and adulteration of edible oils and achieved good results. Currently, spectroscopic techniques, such as ion mobility spectrometry (IMS) fingerprints (Contreras, Arroyo-Manzanares, Arce & Arce, 2019), nuclear magnetic resonance (NMR) (Nam et al., 2014) or any type of vibrational spectroscopy (i.e., infrared (IR) spectroscopy (Uncu & Ozen, 2019), Fourier transform infrared (FT-IR) spectroscopy (Rodríguez, Gagneten, Farroni, Percibaldi, & Buera, 2019), Raman spectroscopy (Woo Park et al., 2017) or laser-induced breakdown spectroscopy (LIBS) (Gazeli, Bellou, Stefas, & Couris, 2020), require fewer sample preparation procedures and are rapid and non-destructive, and these advantages have triggered the use of spectroscopic techniques as alternative approaches to detect the adulteration of edible oils. Three-dimensional excitation-emission (EEM) fluorescence spectroscopy is considered an effective method for analysing food because it possesses several advantages in terms of characterizing the chemical composition, a high sensitivity, simplicity, low cost, and rapid and minimal sample preparation procedures. Because edible oils contain intrinsic fluorophores, including vitamins, phenolic compounds, chlorophyll, and oxidation products, fluorescence spectroscopy has a greater appeal in the detection of edible oils. Recently, the combination of EEM fluorescence spectroscopy and chemometrics has successfully been utilized for the classification and determination of the authenticity of camellia oil, honey, coffee, dairy products, vinaigrette and other foods (Wang et al., 2019; Lenhardt, Bro, Zeković, Dramićanin, & Dramićanin, 2015; Botelho, Oliveira & Franca, 2017; Kamal, & Karoui, 2015; Peng et al., 2019). With the improvement of the standard of life of many people, high-quality edible oil has become a highly sought product. The use of EEM fluorescence technology to study the adulteration of high-quality edible oils has become a research hotspot. Durán Merás et al. (Durán Merás et al., 2018) reported the potential of EEM fluorescence spectroscopy coupled with multiple chemometric methods, such as a parallel factor analysis-linear discriminant analysis (PARAFAC-LDA) and unfolded partial least-squares discriminant analysis (UPLS-DA), to detect the adulteration of extra virgin olive oil with other olive oils. Guimet et al. (Guimet, Ferré, Boqué, & Rius, 2004) reported the application of an unfolded principal component analysis (U-PCA) and PARAFAC to the EEM spectroscopy to distinguish virgin olive oils and pure olive oils, and achieved good classification results. Jiménez-Carvelo et al. (Jiménez-Carvelo, Lozano, & Olivieri, 2019) successfully developed EEM spectroscopy coupled with a principal component analysis (PCA) and N-way partial least squares discriminant analysis (NPLS-DA) to confirm the authenticity and geographical origin of Argentinean extra virgin olive oils. To the best of our knowledge, no study has reported the classification and determination of the adulteration of SO based on EEM fluorescence. The 2

93 94 95 96 97 98 99 100 101

aim of this study is to combine EEM fluorescence spectroscopy and four different chemometric analyses, including alternating trilinear decomposition-linear discriminant analysis (ATLD-LDA), alternating trilinear decomposition-partial least squares–discriminant analysis (ATLD-PLS-DA), alternating trilinear decomposition-support vector machine (ATLD-SVM) and UPLS-DA, to evaluate the difference between SO and other cheap edible oils (including CO, SBO and RSO), as well as the adulteration of SO. The combination of the second-order correction algorithm (ATLD) and the pattern recognition algorithms (LDA, PLS-DA, and SVM) not only achieves the characterization of the components in edible oil but also achieves the rapid classification of adulterated SOs.

102

2.

103

2.1 Samples

104 105 106 107 108 109 110 111 112 113 114 115 116

In the present study, 96 edible oil samples, including 51 SOs from 17 different brands, 15 COs from 5 different brands, 15 SBOs from 5 different brands, and 15 RSOs from 5 different brands, were purchased in a large local supermarket. All purchased edible oils are produced by well-known international manufacturers, and conform to the national standard of vegetable oil (GB 2716-2018) and sesame seed oil (GB/T 8233-2018) of China. These samples were used to establish a classification model of pure edible oil (Model 1). Adulterated SOs were prepared by adding SOE to the cheaper edible oils CO, SBO, and RSO at 1:9, 2:8, 3:7, 4:6, 5:5, 6:4, 7:3, 8:2, and 9:1 ratios (v/v). Three brands each of CO, SBO, and RSO were randomly selected for the repeated configuration. Therefore, 81 adulterated SOs sample (including adulterated SO1 (CO + SOE), adulterated SO2 (SBO + SOE) and adulterated SO3 (RSO + SOE)) were obtained. One hundred thirty-two samples (including 81 adulterated SOs and 51 pure SO samples) were used to establish an identification model for SO adulteration (Model 2).

117

2.2 EEM measurements

118 119 120 121 122 123 124 125 126 127

All EEM fluorescence spectra of these above samples were acquired using an FS920 steady-state fluorescence spectrometer (Edinburgh instruments, Edinburgh, UK) equipped with a Xe900 450W Xenon arc lamp, single-photon counting, Czerny-turner monochromator, and F900 advanced software. The EEM fluorescence data were collected under the following conditions: the excitation wavelengths ranging from 330 to 550 nm (10 nm steps); emission wavelengths ranging from 350 to 750 nm (2 nm steps), an integration time of 0.1 s, and slit widths of excitation and emission monochromators of 10 nm. In the measured spectral range, the measurement time of a single sample is about 400s. Furthermore, the sample chamber temperature was maintained at 20°C using a TC125 temperature controller (Quantum Northwest, Inc., Spokane, WA, USA).

128

2.3 Chemometric methods

129

2.3.1 ATLD

130 131 132 133 134

The alternating trilinear decomposition (ATLD) algorithm (Wu, Shibukawa, & Oguma, 1998) is an improvement in the traditional PARAFAC that is insensitive to the number of components. In addition, ATLD uses the trilinear component model of the slice matrix, which reduces the memory required for calculation, improves the efficiency of the operation, and has the advantage of fast convergence.

Materials and methods

3

135 136

In this study, each sample was measured to obtain an excitation-emission matrix (EEM) dataset ( I × J ) , where I and J represent the number of excitation wavelengths and

137 138

emission wavelengths. A series of EEMs is arrayed in a three-way data array X ( I × J × K ), where K is the number of samples. Each element xijk of X was decomposed by the ATLD

139

model using the following equation: N

140

xijk = ∑ ain b jn ckn + eijk（i = 1, 2,L , I ; j = 1, 2,L , J ; k = 1, 2,L , K）

141

where xijk is the fluorescence intensity of sample k at the excitation wavelength i and

142

emission wavelength j ; N is the total number of responsive factors; a in , b jn and ckn represent

143

the elements of the normalized excitation spectral matrix A ( I × N ) , normalized emission

n =1

144

spectral matrix B ( J × N ) , and relative concentration matrix C ( K × N ) (score matrix),

145

respectively; and eijk represents an element

146

After ATLD decomposition, the score matrix

147 148 149 150

methods, such as LDA, PLS-DA and SVM, were further used to classify the samples. Furthermore, the meaningful chemical compositions of a sample were identified from the decomposed spectral profiles of A and B .

151

2.3.2 LDA

152 153 154 155 156 157 158 159 160 161 162 163 164 165 166

Linear discriminant analysis (LDA), also known as Fisher LDA, is one of the classic pattern recognition methods developed on discriminant analysis (DA) (Belhumeur, Hespanha, & Kriegman, 1997). This method makes full use of the category information of the known training samples, and transforms the original problem into the eigenvalue problem of the inter-class and the intra-class matrix. The core idea of this method is to project high-dimensional data into a low-dimensional vector space and strive to maximize the inter-class dispersion and the minimum intra-class dispersion in the new model space to extract the classification information and reduce dimensionality. Principal component analysis (PCA) selects a projection direction that maximized the variance of the original data in each dimension of the projection subspace, whereas LDA selects a dimension to ensure that the different types of data are separated as much as possible after the original data are projected on that dimension. Compared with the neural network method, LDA does not require parameters to be adjusted, which effectively avoids the problem of parameter selection. LDA even exhibits comparable performance to support vector machine (SVM), but its computational efficiency is much better than SVM.

167

2.3.3 PLS-DA and UPLS-DA

168 169 170 171 172 173 174 175 176

Partial least squares-discriminant analysis (PLS-DA) (Indahl, Martens, & Næs, 2007) is one of the most commonly used classification techniques in chemistry. This method is based on the PLS regression algorithm that converts the observed data into a set of latent variables (LVs) with a maximum covariance and is used to predict the dependent variables, which are called class variables. The PLS-DA corresponds to the inverse-least-squares approach of LDA, and the results are approximately the same, but with the advantages of PLS noise reduction and variable selection (Barker, & Rayens, 2003). The optimal number of LVs is usually selected through a cross-validation procedure, which minimizes the error in classification. In this study, we used Venetian blind cross-validation, which selects every sth sample from the

( i, j, k ) in the residual tensor E ( I × J × K ) . C ( K × N ) combined with pattern recognition

4

177 178

dataset and performs s data splits to ensure that all samples are retained only once. In the present study, for PLS-DA, a classification model was built using the C ( K × N )

179 180 181 182 183 184

score matrix of ATLD. However, UPLS-DA is an unfolding strategy for second-order data. The advantage of unfolding is that it uses all of the information in the second-order data, and almost all classification methods available for two-way data are allowed. However, unfolding the array data presents some drawbacks, such as the very large data matrix obtained by unfolding. Therefore, the number of unrelated variables becomes greater than the number of truly meaningful variables, resulting in complex and time-consuming calculations.

185

2.3.4 SVM

186 187 188 189 190 191 192

SVM (Cortes, & Vapnik, 1995) was widely used in data classification tasks, because it has obvious advantages in solving practical problems that are particularly represented by the small sample size, nonlinearity, and high-dimensional datasets. An SVM is applied with the aims of mapping the original input data to a high-dimensional feature space, defining a maximum margin of separation between classes, and establishing a classification hyperplane in the central area of the maximum margin (Pradhan, 2013). The general idea of achieving classification is summarized below. First, the kernel function ϕ ( x ) is introduced and used to

193 194 195 196

map the training set data to a high-dimensional feature space. Then, the classification hyperplane that constructs the largest interval in the feature space is identified, and the points of different labels are separated. In the present study, the radial basis function (RBF) with great adaptability was chosen as a kernel function ϕ ( x ) implemented in SVM. The penalty

197 198 199 200 201 202

factor C and kernel function parameter g of RBF-SVM must be determined. Since the values of these two parameters directly affect the classification accuracy of the SVM, a parameter optimization method was introduced to obtain the optimal parameters. Particle swarm optimization (PSO) (Kennedy, & Eberhart, 1995), which is a population-based stochastic optimization technique, was introduced to optimize these two parameters of SVM and identify their optimal combination within a certain range.

203

2.4 Evaluation of performance

204 205 206 207 208 209 210 211 212 213 214 215

The performance parameters, such as correct classification rate (CCR), sensitivity, specificity, and precision, are used to characterize the performance of the analytical method (López, Callao, & Ruisánchez, 2015). Usually, when evaluating a class-modelling approach using sample test results, samples belonging to the class are designated as a true positive (TP) (the number of positive samples that are correctly identified as positive samples), false negative (FN) (the number of positive samples that are misclassified as negative samples), false positive (FP) (the number of negative samples that are incorrectly identified as positive samples), and true negative (TN) (the number of negative samples that are correctly identified as negative samples). The CCR is an important parameter used to evaluate classification models, and was used to assess the overall performance of the multi-class classification model. CCR is also regarded as the overall accuracy and was calculated using the following formula:

216

C CR =

TP + TN × 100% TP + T N + FP + F N

217

where ( TP + TN ) is the number of samples belonging to class and correctly assigned to class

218

and ( TP + TN + FP + FN ) is the total number of samples 5

219 220

Sensitivity is defined as the ability of the model to correctly recognize samples that resulted in the true positive assignation and is calculated as follows:

221 222 223

Sensitivit y =

Conversely, specificity characterizes the ability of the true class to refuse the samples from all the other classes, and is calculated as follows:

224 225 226 227

TP TP + FN

Specificity =

TN TN + FP

Precision indicates the ability of a classification model to exclude samples from other classes in the class under consideration. It is typically measured by the ratio of the correctly classified category to the total number of samples assigned to that category:

228

Precision =

TP TP + FP

229

3. Results and discussion

230

3.1 Analysis of the EEM spectra

231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256

EEM fluorescence spectra are presented as contour maps of all pure edible oil samples (SO, CO, SBO, and RSO) after removing Rayleigh scattering, which was performed using the new Delaunay triangulation method (Zepp, Sheldon, & Moran, 2004), as shown in Fig. 1. Spectra were recorded at excitation wavelengths ranging from 330 to 550 nm and emission wavelengths ranging from 350 to 750 nm, and the size of each sample was 23 × 201. These EEM spectra reflect the specificity of intrinsic fluorophores of pure edible oils, enabling their classification. Within the measured spectra for SO (Fig. 1(a)), two distinct fluorescence emission regions are clearly visible. The first fluorescence region is located at excitation wavelengths ranging from 450–500 nm and emission wavelengths ranging from 500–600 nm, and it is mainly attributed to vitamin E (Kyriakidis, & Skarkalis, 2000; Zandomeneghi, Carbonaro, & Caffarata, 2005), which also verified that SO is rich in vitamin E. Another less intense fluorescence band detected at emission wavelengths of 660–690 nm (excitation wavelength range with 400-550 nm) is characteristic of the fluorescent pigments, mainly chlorophylls and pheophytins (Galeano Díaz, Durán Merás, Correa, Roldán, & Rodríguez Cáceres, 2003). In the EEM spectrum of CO (Fig. 1(b)), the first fluorescent peak is located at emission wavelengths ranging from 400–500 nm (excitation wavelengths ranging from 330 to 400 nm), and is as attributed to oxidation and degradation products (Milanez et al., 2017). A less intense fluorescence band with an excitation wavelength of approximately 470 nm and an emission wavelength of approximately 530 nm is ascribed to vitamin E. The EEM spectrum of SBO is shown in Fig. 1(c). Similar to the EEM spectrum of CO, the main fluorescence peak is located at emission wavelengths of 400–500 nm (excitation wavelengths ranging from 330 to 400 nm). Another less intense fluorescence band with emission at 660–690 nm is attributed to chlorophylls. As shown in Fig. 1(d), the spectrum of RSO presents multiple fluorescent peaks, located in two fluorescence emission bands at 500-550 nm and 660-690 nm, which are mainly attributed to vitamin E, chlorophylls and pheophytins.

257

3.2 ATLD results

258 259 260 261 262

The datasets included in Model 1 and Model 2, which are mentioned in Section 2.1, were arranged as two three-way tensors X1 (23 × 201 × 96) and X2 (23 × 201 × 132), respectively. The first dimension of this tensor is the number of excitation wavelengths, the second dimension refers to the number of emission wavelengths, and the third dimension is the number of samples. The ATLD method (which employs DTLD vectors) has been applied to 6

263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291

these two tensors. Because both the resolved excitation and emission spectra must be positive, non-negative constraints was applied on ATLD. Usually, the number of factors should be estimated before the multivariate calibration. The core consistency diagnostic (CORCONDIA) (Bro, & Kiers, 2003) was applied to select the number of spectral factors in these three-way tensors. For Model 1, a five-factor N = 5 ATLD model was chosen (the core consistency ranged from 77% to 100%). For Model 2, a five-factor N = 5 ATLD model was also chosen (the core consistency ranged from 75% to 100%). Using the ATLD algorithm (N = 5) to decompose the EEM spectrum of Model 1, normalized excitation spectra (a1), normalized emission spectra (b1) and the relative concentration, also called ATLD scores (c1) were obtained, as shown in Fig. 2. Within the measured spectra, the fluorescence peak position of the first factor (blue) with the excitation/emission wavelengths λ ex / λem = 350 nm/400 nm, second factor (green) with the

292

3.3 Classification according to the chemometric algorithms

293 294 295 296 297 298 299 300 301 302 303 304 305 306 307

The main goal of the classification method is to identify a mathematical relationship between a set of descriptive variables (such as spectral measurements, chemical measurements, etc.) and a qualitative variable (defined category). In the present, four classification methods, including LDA, PLS-DA, SVM, and UPLSDA, were used to build classification models for pure edible oils (SO, CO, SBO, and RSO) (Model 1), and pure SO and adulterated SOs (Model 2). For the 96 samples included in Model 1, approximately 70% of each type of pure edible oil samples was used to construct training set, and the remaining 30% of each type samples were used to build a test set to assess the predictive power of the established model. Sixty-six training samples and 30 test samples were analysed using these classification methods. Similarly, for the 132 samples in Model 2, approximately 70% of each type of pure SO samples and adulterated SO samples were used to build the training set of the model, and the remaining samples were used as the test set. Therefore, for Model 2, the numbers of samples in the training set and test set were 93 and 39, respectively. We used the Kennard-Stone algorithm to select the training and test samples (Kennard, & Stone, 1969), which selected representative samples distributed homogeneously in the multivariate space.

excitation/emission wavelengths λ ex / λem = 380 nm/450 nm and third factor (sky blue) with the excitation/emission wavelengths λ ex / λem = 425 nm/500 nm should be related to oxidation and degradation products, which are present in all four pure edible oils (SO, CO, SBO and RSO). The fluorescence peaks of the fourth factor (red) with the excitation/emission wavelengths λ ex / λem = 420-480 nm/500-550 nm corresponds to vitamin E, which is present at relatively high levels in SO. The fifth factor (purple) and third factor (sky blue) with fluorescent emission peaks at 680 nm (multiple excitation peaks) are considered typical signatures of native chlorophyll and pheophytins. Fig. 2 also shows the profiles resolved using the ATLD algorithm (N = 5) for Model 2. As mentioned above, Fig. 2 presents normalized excitation spectra (a2), normalized excitation spectra (b2) and the ATLD scores (c2), respectively. The scores for each factor in SO exhibited relatively small changes. However, with the continuous increase in the content of SOE, the scores for factors in adulterated SO1 (CO + SOE), adulterated SO2 (SBO + SOE) and adulterated SO3 (RSO + SOE) changed. The difference in analyte scores provides the basis for subsequent chemometric classification. However, it is worth to note that the ATLD method was used as a blind resolution algorithm, and one factor may contain several structurally similar fluorescent components, which are called “co-factor” in mathematics.

7

308

3.3.1 Classification of pure edible oils - Model 1

309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349

The exploratory analysis of data included in Model 1 was performed using the ATLD scoring method to observe the class structure. Data scaling is required for pre-processing to eliminate the difference in sample feature attributes; as a result, each sample feature quantity value is on the same order of magnitude. In the present study, autoscaling was used to perform certain types of processing of the original feature quantity of the sample in advance. As shown in Fig. S1, the scaled score data for the five factors were calculated for every pure edible oil. After data autoscaling, the dispersion of different types of oil samples was relatively high. In particular, SO samples had a higher content of factor 4 than other edible oil samples; SBO samples showed a higher content of factor 1. Factor 5 was present at higher levels in RSO. For Model 1, all spectra data were arranged in a three-way tensor X1 (23 × 201 × 96), where 96 represents the sum of 66 training sets and 30 test set samples. Four classification methods, including ATLD-LDA, ATLD-PLS-DA, ATLD-SVM and UPLS-DA, were used separately to develop classification and identification models for different pure edible oil samples. The first three methods are based on the ATLD scores to establish a discriminant model (the feature data for each sample is a 5), and the last method is designed to directly build a discriminant model based on the unfolded high-dimensional data array ( IJ × K ) (the feature data for each sample is a 4623). According to the optimal CCRs for the cross-validation models, training set and test set, the best PLS-DA and UPLS-DA models were obtained using 3 and 3 LVs, respectively. The optimal number of LVs was chosen based on 10-fold cross-validation. On the other hand, the penalty factor C and kernel parameter g in SVM kernel function will affect the predicted

350

3.3.2 Classification of SO and adulterated SOs - Model 2

351 352

Compared to the classification of pure edible oils (Model 1), the samples included in Model 2 were arranged in a three-way tensor X2 (23 × 201 × 132) and labelled with four

CCR of the model. Since the traditional methods readily encounter the dilemma of a local optimal solution when searching for the combination of C and g , in this study, the PSO algorithm was used to identify the optimal values of C and g ( C & g ). The parameters were set to the default values: maximum number of iterations: 200; the size of the number of particles: 20; personal and social learning factors: c1  = 1.5,   c 2 = 1.7; and inertia weight: w  = 0.6. The values of C and g were limited in the range of [0.01, 10] and [1, 10], respectively. The final fitness curve is shown in Fig. S2(a). Using PSO-SVM, the optimal parameter penalty factor C and kernel parameter g were obtained as 1.00 and 0.62, respectively. The optimal parameters for Model 1 are shown in Table 1. In the present study, the CCR was used to assess the overall performance of the multi-class classification model. As shown in Table 1, the CCR (%) for the training set, cross-validation, and test set were all 100%. Considering the large differences in the fluorescence of pure edible oil samples measured, each method achieved good classification results. The plots of the classification results based on the scores of the first two canonical variables for the LDA model (a), the first two LVs for the PLS-DA model (b) and UPLS-DA model (c) are shown in Fig. 3. All samples were successfully classified into the appropriate classes with a high level of precision, considering that only two canonical variables and LVs were used for this visualization. Although the visual classification plots of ATLD-SVM are not displayed, it still showed a strong classification ability with 100% CCR.

8

353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402

categories, including pure SO, adulterated SO1 (CO + SOE), adulterated SO2 (SBO + SOE) and adulterated SO3 (RSO + SOE). Similarly, 132 represents the sum of the 93 training sets and 39 test set samples. ATLD-LDA, ATLD-PLS-DA, ATLD-SVM and UPLS-DA, were also applied to Model 2. For the PLS-DA and UPLS-DA models, the optimal CCRs were obtained when using 3 and 7 LVs, respectively. For SVM, the fitness curve of PSO is shown in Fig. S2(b). The use of the PSO algorithm to optimize SVM parameters effectively accelerated the speed of convergence, and basically converged after approximately 3 iterations, and the optimal fitness reached 98.9%. The penalty factor C = 9.78 and kernel parameter g = 2.50 of SVM model were obtained. Similarly, the CCR was initially calculated to evaluate the performance of the classification algorithms. The CCRs of the training set (Tra), cross-validation (CV) and test set (Test) are shown in Table 1. The optimal CCRs of the LDA and PLS-DA methods were approximately 90% and 90%, respectively. For this complex system, the results are acceptable. Moreover, based on the confusion matrix (shown in Table 2), these two methods achieved a CCR of 100% for pure SO and adulterated SOs, but different types of adulterated SOs were misidentified, which may be due to the higher SOE contents in some adulterated SO samples or presence of similar components in cheaper edible oils. Fortunately, the latter two methods both produced the best results (CCR = 100%). This finding precisely represents the unique advantages of SVM in solving small sample, nonlinear and high dimensional pattern recognition problems. UPLS-DA directly models the original data matrix of the sample and uses the richer spectral information than simply the scores for ATLD decomposition, but the processing time is slightly longer than other methods. Subsequently, sensitivity, specificity, and precision for training and test sets were calculated to evaluate the classification performance of these algorithms, and the results are shown in Table 3. Regarding the use of the ATLD-LDA method for the test set, an acceptable CCR (90%) was obtained, with the sensitivity, specificity, and precision ranging from 0.50 to 1.00, 0.87 to 1.00, and 0.67 to 1.00, respectively. In addition, the ATLD-PLS-DA method also produced acceptable results (CCR = 90%), with the sensitivity, specificity, and precision ranging from 0.50 to 1.00, 0.87 to 1.00 and 0.67 to 1.00, respectively. Notably, the ATLD-SVM and UPLS-DA models obtained encouraging results, including sensitivity, specificity, and precision values of 1.00. The predicted results of the ATLD-PLS-DA and UPLS-DA models for each class are shown in Fig. 4. For ATLD-LDA (Fig. 4(a1-d1)), pure SO and adulterated SO3 (RSO + SOE) were correctly classified both in the training and the test sets. Meanwhile, for adulterated SO1 (CO + SOE) and adulterated SO2 (SBO + SOE), different degrees of class overlap were observed, and the algorithm was unable to correctly distinguish all samples. More interestingly, as shown in Fig. 4(a2-d2), UPLS-DA achieved an accurate classification of SO and three adulterated SOs in both the training and test sets. In addition, ATLD-SVM also achieved a CCR of 100%, but the graph depicting the prediction result is not displayed. The comparison of these supervised techniques using the ATLD scores, showed that the SVM model is superior to the LDA and PLS-DA models in the present study. The UPLS-DA model still exhibits strong classification performance, but the processing time is slightly longer, potentially because it directly models the original data matrix of the sample. On the one hand, it uses all of the spectral information to improve the classification result. On the other hand, the original data matrix of each sample was unfolded into a large 4623-feature vector, and some redundant information was present, resulting in a longer processing time. Overall, these findings were rationalized by considering that the combination of chemometrics and pattern recognition algorithms is a promising and reliable tool with great potential, not only for characterizing the components in a sample but also for achieving the accurate classification of samples. 9

403

4. Conclusion

404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422

In this report, EEM fluorescence spectroscopy combined with chemometric methods (including ATLD-LDA, ATLD-PLS-DA, ATLD-SVM and UPLS-DA) successfully characterized meaningful chemical components in edible oils, achieved the classification of pure SO and other cheap edible oils, and identified adulterated SOs. An ATLD five-factor model of EEM spectra characterized the composition and contents of various components of different edible oils and performed meaningful chemical decomposition, including oxidation products, vitamin E, chlorophyll and other products. For the classification of pure SO, CO, SBO and RSO (Model 1), all methods achieved a satisfactory CCR (100%) with high sensitivity, specificity and precision. However, for the classification of pure SO and adulterated SOs (including CO + SOE, SBO + SOE and RSO + SOE) (Model 2), the CCRs of the LDA and PLS-DA methods were approximately 90% and 90%, respectively. For this complex system, the result is acceptable. Moreover, these two methods achieved 100% CCRs for pure SO and adulterated SOs, and only misidentified different types of adulterated SOs. In this case, the highest classification CCRs (100%) were provided by the ATLD-SVM and UPLS-DA methods. Overall, this study provides a promising demonstration of consumer fraud in pure SO. The combination of EEM fluorescence spectroscopy with chemometric methods represents a powerful tool for preventing potential adulterations of high-quality edible oils. Moreover, this method is considered a simple, rapid, green and non-destructive determination method because it does not require pre-treatments or the use of organic solvents.

423

Declarations of interest

424

The authors declare no conflict of interest.

425

Acknowledgments

426 427 428

The authors gratefully acknowledge the National Natural Science Foundation of China (Nos. 61771419) and Natural Science Foundation of Hebei Province of China (Nos. F2017203220) for financial supports.

429

Appendix A. Supplementary data

430

References

431 432 433 434 435 436 437 438 439 440 441 442 443 444 445

Anilakumar, K. R., Pal, A., Khanum, F., & Bawa, A. S. (2010). Nutritional, Medicinal and Industrial Uses of Sesame (Sesamum indicum L.) Seeds - An Overview. Agriculturae Conspectus Scientificus, 75, 159. Barker, M., & Rayens, W.S. (2003). Partial least squares for discrimination. Journal of Chemometrics, 17, 166-173. Belhumeur, P. N., Hespanha, J.P., & Kriegman, D. J. (1997). Eigenfaces vs. fisherfaces: recognition using class specific linear projection. IEEE Transaction Pattern Analysis Machine Intelligence, 19, 711-720. Botelho, B. G., Oliveira, L. S., & Franca, A. S. (2017). Fluorescence spectroscopy as tool for the geographical discrimination of coffees produced in different regions of Minas Gerais State in Brazil. Food Control, 77, 25-31. Bro, R., & Kiers, H.A.L. (2003). A new efficient method for determining the number of components in PARAFAC models, Journal of Chemometrics, 17, 274–286. Contreras, M. D. M., Arroyo-Manzanares, N., Arce, C., & Arce, L. (2019). HS-GC-IMS and chemometric data treatment for food authenticity assessment: Olive oil mapping and classification through two different devices as an example. Food Control, 98, 82-93. Cortes, C., & Vapnik, V. (1995) Support-vector networks. Machine Learning, 20:273-297. 10

446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489

Cserháti, T., Forgács, E., Deyl, Z., & Miksik, I. (2005). Chromatography in authenticity and traceability tests of vegetable oils and dairy products: a review. Biomedical Chromatography, 19, 183-190. Devarajan, S., Chatterjee, B., Urata, H., Zhang, B., Ali, A., Singh, R., & Ganapathy, S. (2016). A Blend of sesame and rice bran oils lowers hyperglycemia and improves the lipids. The American Journal of Medicine, 129, 731-739. Durán Merás, I., Domínguez Manzano, J., Airado Rodríguez, D., & Muñoz De La Peña, A. (2018). Detection and quantification of extra virgin olive oil adulteration by means of autofluorescence excitation-emission profiles combined with multi-way classification. Talanta, 178, 751-762. Elleuch, M., Besbes, S., Roiseux, O., Blecker, C., & Attia, H. (2007). Quality characteristics of sesame seeds and by-products. Food Chemistry, 103, 641-650. Galeano Díaz, T., Durán Merás, I., Correa, C. A., Roldán, B., & Rodríguez Cáceres, M. I. (2003). Simultaneous Fluorometric Determination of Chlorophylls a and b and Pheophytins a and b in Olive Oil by Partial Least-Squares Calibration. Journal of Agricultural and Food Chemistry, 51, 6934-6940. Gazeli, O., Bellou, E., Stefas,D., & Couris, S. (2020) Laser-based classification of olive oils assisted by machine learning. Food Chemistry, 302, 125329. González Martín, Y., Luis Pérez Pavón, J., Moreno Cordero, B., & García Pinto, C. (1999). Classification of vegetable oils by linear discriminant analysis of Electronic Nose data. Analytica Chimica Acta, 384, 83-94. Guimet, F., Ferré, J., Boqué, R., & Rius, F. X. (2004). Application of unfold principal component analysis and parallel factor analysis to the exploratory analysis of olive oils by means of excitation–emission matrix fluorescence spectroscopy. Analytica Chimica Acta, 515, 75-85. Indahl, U. G., Martens, H., & Næs, T. (2007). From dummy regression to prior probabilities in PLS DA. Journal of Chemometrics, 21, 529–536. Jiménez-Carvelo, A. M., Lozano, V. A., & Olivieri, A. C. (2019). Comparative chemometric analysis of fluorescence and near infrared spectroscopies for authenticity confirmation and geographical origin of Argentinean extra virgin olive oils. Food Control, 96, 22-28. Kamal, M., & Karoui, R. (2015). Analytical methods coupled with chemometric tools for determining the authenticity and detecting the adulteration of dairy products: A review. Trends in Food Science and Technology, 46, 27-48. Kennard, R.W., & Stone, L.A. (1969). Computer Aided Design of Experiments. Technometrics, 11, 137-148. Kennedy, J., & Eberhart, R. (1995). Particle swarm optimization. Proc. IEEE Int. Conf. Neural Networks, 4: 1942-1948. Kyriakidis, N. B., & Skarkalis, P. (2000). Fluorescence spectra measurement of olive oil and other vegetable oils. Journal of AOAC International, 83, 1435-1439. Lee, D., Lee, E., Kim, H., Kim, S., & Kim, K. (2001). Reversed phase liquid chromatographic determination of triacylglycerol composition in sesame oils and the chemometric detection of adulteration. Analytica Chimica Acta, 429, 321-330. Lee, W., Su, N., Lee, M., & Lin, J. (2013). Assessment of authenticity of sesame oil by modified Villavecchia Test and HPLC-ELSD analysis of triacylglycerol profile. Food Research International, 53, 195-202. Lenhardt, L., Bro, R., Zeković, I., Dramićanin, T., & Dramićanin, M. D. (2015). Fluorescence spectroscopy coupled with PARAFAC and PLS DA for characterization and classification of honey. Food Chemistry, 175, 284-291. 11

490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533

López, M. I., Callao, M. P., & Ruisánchez, I. (2015). A tutorial on the validation of qualitative methods: From the univariate to the multivariate approach. Analytica Chimica Acta, 891, 62-72. Milanez, K. D. T. M., Nóbrega, T. C. A., Nascimento, D. S., Insausti, M., Band, B. S. F., & Pontes, M. J. C. (2017). Multivariate modeling for detecting adulteration of extra virgin olive oil with soybean oil using fluorescence and UV–Vis spectroscopies: A preliminary approach. LWT - Food Science and Technology, 85, 9-15. Nam, Y. S., Noh, K., Roh, E. J., Keum, G., Lee, Y., & Lee, K. (2014). Determination of Edible Vegetable Oil Adulterants in Sesame Oil Using 1H Nuclear Magnetic Resonance Spectroscopy. Analytical Letters, 47, 1190-1200. Ni, Y., Zhang, G., & Kokot, S. (2005). Simultaneous spectrophotometric determination of maltol, ethyl maltol, vanillin and ethyl vanillin in foods by multivariate calibration and artificial neural networks. Food Chemistry, 89(3), 465-473. Peng, D., Bi, Y., Ren, X., Yang, G., Sun, S., & Wang, X. (2015). Detection and quantification of adulteration of sesame oils with vegetable oils using gas chromatography and multivariate data analysis. Food Chemistry, 188, 415-421. Peng, T., Yin, X., Sun, W., Ding, B., Ma, L., & Gu, H. (2019). Developing an Excitation-Emission Matrix Fluorescence Spectroscopy Method Coupled with Multi-way Classification Algorithms for the Identification of the Adulteration of Shanxi Aged Vinegars. Food Analytical Methods, 12, 2306-2313. Pradhan, B. (2013). A comparative study on the predictive ability of the decision tree, support vector machine and neuro–fuzzy models in landslide susceptibility mapping using GIS. Computers and Geosciences. 51, 350–365. Rodríguez, S. D., Gagneten, M., Farroni, A. E., Percibaldi, N. M., & Buera, M. P. (2019). FT-IR and untargeted chemometric analysis for adulterant detection in chia and sesame oils. Food Control, 105, 78-85. Uncu, O., & Ozen, B. (2019). A comparative study of mid-infrared, UV–Visible and fluorescence spectroscopy in combination with chemometrics for the detection of adulteration of fresh olive oils with old olive oils. Food Control, 105, 209-218. Vietina, M., Agrimonti, C., & Marmiroli, N. (2013). Detection of plant oil DNA using high resolution melting (HRM) post PCR analysis: A tool for disclosure of olive oil adulteration. Food Chemistry, 141, 3820-3826. Wang, T., Wu, H., Long, W., Hu, Y., Cheng, L., Chen, A., & Yu, R. (2019). Rapid identification and quantification of cheaper vegetable oil adulteration in camellia oil by using excitation-emission matrix fluorescence spectroscopy combined with chemometrics. Food Chemistry, 293, 348-357. Woo Park, C., Lee, I., Kwon, S., Lee, K., Jo, C., & Ko, D. (2017). Authentication of adulterated edible oil using coherent anti-Stokes Raman scattering spectroscopy. Journal of Raman Spectroscopy, 48, 1330-1336. Wu, H., Shibukawa, M., & Oguma, K. (1998). An alternating trilinear decomposition algorithm with application to calibration of HPLC–DAD for simultaneous determination of overlapped chlorinated aromatic hydrocarbons. Journal of Chemometrics, 12, 1-26. Yue, Y., Liu, J., Yao, M., Yao, X., Fan, J., & Ji, H. (2012). The investigation of the binding behavior between ethyl maltol and human serum albumin by multi-spectroscopic methods and molecular docking. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 96, 316-323. Zandomeneghi, M., Carbonaro, L., & Caffarata, C. (2005). Fluorescence of Vegetable Oils: Olive Oils. Journal of Agricultural and Food Chemistry, 53, 759-766. Zepp, R. G., Sheldon, W. M., & Moran, M.A. (2004). Dissolved organic fluorophores in southeastern US 12

534 535

coastal waters: correction method for eliminating Rayleigh and Raman scattering peaks in excitation– emission matrices. Marine Chemistry, 89, 15-36.

13

Rapid detection of the authenticity and adulteration of sesame oil using excitation-emission matrix fluorescence and chemometric methods Yuan-Yuan Yuan, Shu-Tao Wang*, Jun-Zhu Wang, Qi-Cheng, Xi-Jun Wu, De-Ming Kong Measurement Technology and Instrument Key Lab of Hebei Province, Yanshan University, Qinhuangdao 066004, China * Correspondence: [email protected] Table 1 The CCR (%) of the training set (Tra), cross-validation (CV) and test set (Test) for different classification models, as well as the optimal number of LVs for PLS-DA and UPLS-DA models, optimal parameter penalty factor C and kernel parameter g for SVM.

ATLD-LDA

ATLD-PLS-DA

ATLD-SVM

UPLS-DA

Model

Tra

CV

Test

LVs

Tra

CV

Test

C

g

Tra

Test

LVs

Tra

CV

Test

Model 1

100

100

100

3

100

100

100

1.00

0.78

100

100

3

100

100

100

Model 2

91

91

90

3

90

90

90

9.78

2.50

100

100

7

100

100

100

Table 2 Confusion matrices for the ATLD-LDA, ATLD-PLS-DA, ATLD-SVM and UPLS-DA models for training set and test set (the green colour represents correct classification and red donates a false classification; Adu-SO1 represents adulterated SO1 (CO + SOE), Adu-SO2 represents adulterated SO2 (SBO + SOE) and Adu-SO3 represents adulterated SO3 (RSO + SOE).)

Predicted Method Actual

ATLD-LDA SO Adu- AduSO1 SO2

AduSO3

ATLD-PLS-DA SO Adu- AduSO1 SO2

AduSO3

ATLD-SVM / UPLS-DA SO Adu- Adu- AduSO1 SO2 SO3

Training set SO

36

0

0

0

36

0

0

0

36

0

0

0

Adu-SO1

0

Adu-SO2

0

14

5

0

0

18

1

0

0

19

0

0

0

19

0

Adu-SO3

0

8

11

0

0

0

19

0

0

0

3

16

0

0

0

19

0

0

0

19

SO

15

0

0

0

15

0

0

0

15

0

0

0

Adu-SO1

0

4

4

0

0

8

0

0

0

8

0

0

Adu-SO2

0

0

8

0

0

4

4

0

0

0

8

0

Adu-SO3

0

0

0

8

0

0

0

8

0

0

0

8

Test set

Table 3 The sensitivity, specificity, and precision of the LDA, PLS-DA, SVM, and UPLS-DA models for the training and test sets. (Adu-SO1 represents adulterated SO1 (CO +SOE), Adu-SO2 represents adulterated SO2 (SBO + SOE) and Adu-SO3 represents adulterated SO3 (RSO + SOE).)

Sensitivity

Specificity

Precision

Models

Validation

SO

AduSO1

AduSO2

AduSO3

SO

AduSO1

AduSO2

AduSO3

SO

AduSO1

AduSO2

AduSO3

LDA

Training

1.00

0.74

1.00

0.84

1.00

1.00

0.89

1.00

1.00

1.00

0.70

1.00

Test

1.00

0.50

1.00

1.00

1.00

1.00

0.87

1.00

1.00

1.00

0.67

1.00

Training

1.00

0.95

0.58

1.00

1.00

0.89

0.99

1.00

1.00

0.69

0.92

1.00

Test

1.00

1.00

0.50

1.00

1.00

0.87

1.00

1.00

1.00

0.67

1.00

1.00

Training

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

Test

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

PLS-DA SVM / UPLS-DA

Rapid detection of the authenticity and adulteration of sesame oil using excitation-emission matrix fluorescence and chemometric methods Yuan-Yuan Yuan, Shu-Tao Wang*, Jun-Zhu Wang, Qi-Cheng, Xi-Jun Wu, De-Ming Kong Measurement Technology and Instrument Key Lab of Hebei Province, Yanshan University, Qinhuangdao 066004, China * Correspondence: [email protected]

Fig. 1. The contour plots of EEM spectra for the pure edible oils (a) SO, (b) CO, (c) SBO and (d) RSO after removing scattering.

Fig. 2. The profiles resolved by the five-factor ATLD analysis for Model 1: (a1) normalized excitation spectra, (b1) normalized emission spectra and (c1) relative concentration (ATLD scores). The profiles resolved by five-factor ATLD analysis for Model 2: (a2) normalized excitation spectra, (b2) normalized emission spectra and (c2) relative concentration (ATLD scores). (Adu-SO1 represents adulterated SO1 (CO +SOE), Adu-SO2 represents adulterated SO2 (SBO + SOE) and Adu-SO3 represents adulterated SO3 (RSO + SOE).)

Fig. 3. Plots of the classification results based on the scores of (a) the first two canonical variables for the LDA model, (b) the first two LVs for the PLS-DA model, and (c) the first two LVs for UPLS-DA model in training samples included in Model 1. (EV represents the explained variance; blue circles represent samples of pure SO, red circles represent samples of pure CO, green circles represent samples of pure SBO and purple circles represent samples of RSO).

Fig. 4. PLS-DA predictions for each class: (a1) pure SO, (b1) Adu-SO1 (CO + SOE), (c1) Adu-SO2 (SBO + SOE) and (d1) Adu-SO3 (RSO + SOE). UPLS-DA predictions for each class: (a2) pure SO, (b2) Adu-SO1 (CO + SOE), (c2) Adu-SO2 (SBO + SOE) and (d2) Adu-SO3 (RSO + SOE). Vertical dashed lines separate training and test samples. Blue circles and inverted triangles represent samples of pure SO, red circles and inverted triangles represent samples of Adu-SO1, green circles and inverted triangles represent samples of Adu-SO2, and purple circles and inverted triangles represent samples of Adu-SO3.

Highlights 1.

A simple and rapid EEMs fluorescence was developed for adulteration of sesame oil for the first time.

2.

ATLD decomposed the meaningful chemical composition of edible oils.

3.

Discriminant model for detection was built by ATLD-LDA, ATLD-PLS-DA, ATLD-SVM and UPLS-DA.

4.

All methods successfully identified various pure edible oils and adulterated sesame oils