The Combination of Vibrational Spectroscopy and Chemometrics for Analysis of Milk Products Adulteration
Table 1
The application of vibrational spectroscopy (Raman, near infrared, and mid infrared) combined with chemometrics for milk authentication.
Adulteration issues
Type of vibrational spectroscopy
Chemometrics
Results
Ref.
Cow milk
Addition of sucrose to cow milk
Normal mid IR spectra at wavenumbers of 1070–980 cm−1
PCA and SIMCA for classification. PCR and PLS for quantification
The levels of sucrose cold be quantified with Cal: 0.996; Val: 0.993, RMSE (Cal: 0.15% ; Val: 0.20% ), RE% (Cal: 4.9% ; Val: 5.1% ), and RPD (13.40). SIMCA was able to classify test samples with a classification efficiency of 100%
First derivative spectra at wavenumbers of 800-1800 cm_1
PCA and PLS-DA for classification
FTIR spectroscopy has great potentials in quality control of milk and their related products because the PLS-DA model yielded satisfactory separation of the two spectral fingerprints
MIR: 1373, 1454, and 956 cm-1 Raman: 1005, 1154, and 1551 cm-1
SIMCA for classification and PLSR for prediction of milk adulteration
SIMCA result showed the β-carotene band at 1373, 1454, and 956 cm-1 (MIR spectra) and 1005, 1154, and 1551 cm-1 (Raman spectra) as a biomarker for classification of cow milk in goat milk PLSR using MIR and Raman spectra were used to predict goat and cow milk mixtures with 0.32 SECV, 0.98 Cal, 0.57 SEP, and 0.98 Val (MIR) and 0.46 SECV, 0.96 Cal, 0.57 SEP, and 0.94 Val (Raman)
2D IR/NIR heterospectra range of 1400-1704 cm-1 and 4200-4800 cm-1
NPLS-DA for classification of pure milk and adulterated milk
Results showed that, for the samples in the prediction set, the rate of correct classification was 96.2% using synchronous 2D heterospectra IR/NIR correlation spectra versus 88.5% using synchronous 2D homospectral IR/IR and NIR/NIR correlation spectra. Comparison of the results showed that 2D heterospectra IR/NIR correlation spectra and NPLS-DA could give better classification between adulterated milk and pure milk
Addition of five common adulterants (water, starch, sodium citrate, formaldehyde and sucrose) in raw cow milk
MID infrared-ATR spectra range of 4000-600 cm-1
PLS-DA
The method was able to detect the presence of the adulterants water, starch, sodium citrate, formaldehyde, and sucrose in milk samples containing from one up to five of these analytes, in the range of 0.5–10%
Addition of pseudo protein (urea, melamine, and ammonium nitrate) and thickeners (dextrin and starch)
First derivative NIR spectra at wavenumbers of 4000-10.000 cm-1
Nonlinear supervised pattern recognition methods of improved support vector machine (I-SVM) and improved and simplified K nearest neighbours (IS-KNN)
Both methods (I-SVM and IS-KNN) exhibit good adaptability in discriminating adulterated milks from raw cow milks at the concentration of adulteration solutions which equals or exceeds 5%
Normal NIR spectra at wavenumbers 4000-10.000 cm-1
One class partial least square (OCPLS)
The combination of NIR spectroscopy and OCPLS can serve as a potential tool for rapid and on-site screening melamine in milk samples with the total accuracy of 89%, the sensitivity of 90%, and the specificity of 88%
Infant formula (powder), milk powder, and milk liquid
Addition of melamine
NIR spectra range of 9000-4500 cm-1 MIR spectra range of 500-4000 cm-1
Partial least square (PLS), orthogonal projection to latent structures (OPLS), polynomial partial least squares (Poly-PLS), artificial neural networks (ANN), and support vector machine (SVM)
Linear calibration methods (PLS and OPLS) show a much larger prediction error, exceeding 1 ppm. The average error of the PLS/OPLS methods is ppm, while the error of the Poly-PLS, ANN, and SVM-based methods is almost 5 times smaller () The relationship between the MIR/NIR spectrum of milk product and melamine content is nonlinear. Thus, nonlinear regression methods, such as Poly-PLS, ANN, SVR, or LS-SVM, are needed to correctly predict the melamine
Milk adulterated with formaldehyde, hydrogen peroxide bicarbonate, carbonate, chloride, citrate, hydroxide, hypochlorite, starch, sucrose, and water
MIR region at wavenumbers of 1000-4000 cm-1
Multiplicative scatter correction (MSC) for spectra preprocessing; PCA for visualization of the sample distribution, SIMCA for classification milk
In the first step, a one-class model was developed with unadulterated samples, providing 93.1% sensitivity. Four poorly assigned adulterants were discarded for the following step (multiclass modelling). Then, in the second step, a multiclass model, which considered unadulterated and formaldehyde, hydrogen peroxide, citrate, hydroxide, and starch as adulterated samples, was implemented, providing 82% correct classifications, 17% inconclusive classifications, and 1% misclassifications
Tetracycline’s residue (tetracycline, chlortetracycline, and oxytetracycline)
MID FTIR spectra at wavenumber of 4000-550 cm-1
SIMCA for classification, PLS and PCR for quantification of tetracycline residue
SIMCA could be used for classification of pure milk and milk adulterated with the confidence level of 99%. The calibration models developed with three algorithms (PLS1, PLS2 and PCR) to predict tetracycline, chlortetracycline, and oxytetracycline concentrations in milk revealed values of of 0.999, 0.998, and 0.997, respectively
FT-MIR spectra at wavenumber of 1550-1725 and 2800-2981 cm-1, while FT-NIR used raw and first derivative spectra at the region of 3500-8000 cm-1
PLS for quantification of tetracycline hydrochloride in milk
FT-MIR: the optimum number of factors using PLS method was 15, and the between the predicted and actual values was 0.89, the SEC value was 385 ppb, and the repeatability value was 163 FT-NIR: PLS-first derivative calibration method gave an value of 0.76, SEC value of 431 ppb, and the repeatability value of 73 ppb Results indicated that FT-MIR spectroscopy could be used for rapid detection of tetracycline hydrochloride residues in milk
ANN for quantification and PLS for corrected prediction
A high-capacity prediction model was obtained using ANN, with of 0.9999. Alternatively, ANN can be replaced by a linear model adjusted using PLS, which also exhibited reasonable results for the prediction of percentage of whey added ()
PCA and the data driven soft independent modeling of class analogy (DD-SIMCA) for classification, PLS for quantification
Preliminary PCA performed on the whole data revealed that both big similarities and differences between pure and adulterated milk samples were collected from a variety of dairy farms The DD-SIMCA approach achieved satisfactory classification. By the PLSR model, standard error of prediction (SEP) values of 4.35, 0.34, 4.74, and 5.56 g/L and Val value of 0.94, 0.87, 0.93, and 0.89 were obtained for water, urea, starch, and goat milk, respectively
PLSR for quantification of fat, protein, and lactose
The obtained prediction models were thoroughly tested on all the remaining samples not included in the calibration sets (, respectively, 846 and 857). For the post hoc prediction models, this resulted in an overall prediction error (RMSEP) smaller than 0.08% (all % are in ) for milk fat (range 1.5-6.3%), protein (2.6-4.3%), and lactose (4-5.1%), while for the real-time prediction models, the RMSEP was smaller than 0.09% for milk fat and lactose and smaller than 0.11% for protein
Second derivative NIR spectra (whole region, 1100-1850, 2048-2500, and combination of 1100-1850, 2048-2500 nm)
DPLS and SIMCA for classification, PLSR for quantification
The best DPLS classification model for natural milk, milk adulterated with water and milk adulterated with whey was developed using the MSC and second derivative spectra in the whole region of 1100–2500 nm with a PLS factor of 7 and classification performance of 100% The best prediction result is obtained for water adulterated in natural milk, when the model is developed by using the MSC spectra over the whole region of 1100–2500 nm. Its statistical results are the lowest value of the root mean square error of prediction (RMSEP) of 2.159% () with a PLS factor of 4, while the best calibration model for milk adulteration by mixing whey yields the prediction result with a RMSEP value of 0.244% () by a PLS factor of 4. This model was built using the MSC pretreated spectra of the combination regions of 1100–1850 and 2048–2500 nm
PCA perfectly classified between pure milk and milk adulterated with water. PLS was successfully used to predict the concentration of water in milk samples with more than 0.9 and RMSEC lower than 0.04
Artificial neural network (ANN) for classification and multiple linear regression (MLR) for quantification
Chemometrics of ANN could classify pure and adulterated milk samples with hydrogen peroxide with high accuracy. Quantification of hydrogen peroxide could be obtained using MLR with of calibration 0.80 and RMSEC value of 0.15
