Abstract
Quality control of plant-based medicine and supplements must be carried out to ensure uniformity in quality and safety in their use, resulting in the need for effective and accurate analytical methods. Infrared spectroscopy is a method of qualitative and quantitative analysis that is fast, time-saving, cost-effective,accurate, and nondestructive. This method has been applied for quantitative analysis of compounds in complex matrices such as plant-based medicine and supplements supported by chemometrics techniques. The success of infrared spectroscopy applications for quantitative analysis of phytochemicals and adulterants content in plant-based medicine and supplement can happen by several factors. This article highlights the effect of spectral preprocessing and variable selection on quantitative analysis of phytochemical and adulterant in plant-based medicine and supplements using infrared spectroscopy. Literature search was conducted with PubMed, Google Scholar, and Science Direct by selecting quantitative analysis research on plant-based medicines and supplements that utilize spectral preprocessing techniques and variable selection in processing data analysis. The preprocessing spectra and variables selection can affect the accuracy and precision of infrared spectroscopy methods. The variable selection can be done using the wavenumber point technique, the wavenumber interval, or a combination thereof. Variable selection is more commonly used for near-infrared data than for IR data. The optimization of the preprocessing spectra and variables selection technique will be useful in increasing the ability of infrared spectroscopy in predicting compound levels.
Key words: Adulterant, herbal, phytochemical, preprocessing, variable selection
INTRODUCTION
The use of plant-based medicine and supplements had been increased in the past few decades. This is in line with the evolution of self-medication so that there is a tendency to return to traditional and natural products. Consumers choose to use herbal and natural products because natural herbs are safe and more likely to minimize the side effects of using chemical drugs, improve health, and reduce treatment costs.[1] However, the use of herbal products is not yet acceptable in some countries due to counterfeit products, uneven quality, and safety of their use. It may cause negative effects to the consumer, product quality assurance can be assessed from both qualitative and quantitative aspects. Quantitative assessment of herbal products quality is focused on phytochemical components that are naturally contained in sample or adulterants that should not be contained therein.[2,3]
Various analytical methods can be used to examine phytochemical content and adulterants in herbal medicines such as high performance liquid chromatography (HPLC), ultra-high-HPLC, liquid chromatography (LC)-mass spectrometry (MS), gas chromatography (GC)-MS, nuclear magnetic resonance (NMR), and thin-layer chromatography. However, the methods have several disadvantages. HPLC method is still costly and unpracticed in sample handling, needs long-time analysis, and has a high consumption of solvents. NMR method is insensitive and requires a relatively large amount of sample to make a measurement.[4] Furthermore, in herbal medicines analysis, one of the obstacles is the efficiency of the analysis, because the complex components of herbal medicines can complicate the process. In addition, the analysis process can also damage the material.[5]
Infrared spectroscopy can be examined by qualitative and quantitative analysis. This method provides information about the compound content in complex samples with small levels. The complexity of the information provided can be resolved by chemometric techniques. Chemometric techniques can assist in extracting information from spectra through multivariate analysis. Spectroscopy is considered fast, time-saving, cost-effective, accurate, and nondestructive analytical tool.[6] Infrared spectroscopy has been used successfully for quantitative analysis in various fields including pharmaceutical, food industry, agriculture, and biological evaluation. The success of this method is assessed from the value of root mean square error of calibration (RMSEC), root mean square error of prediction (RMSEP), root mean square error of cross-validation (RMSECV), and determination coefficient (R2).[7] Good linearity and accepted standard errors of a model of good linearity and standard errors of a prediction model are influenced by several factors. The preprocessing spectra and variables selection can affect the accuracy and precision of infrared spectroscopy methods. We performed the PubMed, Google Scholar, and Science Direct databases to find articles providing information on the effect of preprocessing spectra and variables selection. Some of the research done on quantitative analysis using infrared spectroscopy has been presented in Table 1. This article will discuss several preprocessing techniques and variable selection that contribute to improving the quantitative analysis results of chemical components (in the form of adulterant) and phytochemicals (second metabolites) in plant-based medicine and supplements to standardize products with green analytical chemistry methods.
Table 1.
Quantitative analysis using infrared spectrometry coupled with chemometrics
| Quantitative analysis by infrared spectroscopy | Description |
|---|---|
| Morfin and Thebain in Poppy head (flower or fruit of family Papaveraceae) by FTIR[9] | Mean-centred scalling technique as preprocessing spectra followed by PCA as variable selection gives a better predictive ability, but the model still needs an optimation due to high % RSD for morfin prediction |
| Quersetin, Quersitrin, Rutin, and total flavonoid in genus Fagopyrum by FTIR[10] | Several spectra pretreatments are employed. Wavelet transform gives the best result to preprocessed high resolution spectra, while wavelet transform together with derivative has good linearity for low resolution spectra. Spectra at wavenumber 500-1700 cm−1 and between 2950-3500 cm−1 are chosen to build a PLS calibration model |
| Total flavonoid in 20 leaves extract and commercial herbal supplement by FTIR and NIR[11] | PLS regression built by full spectra of NIR with no spectra pretreatment shows a better prediction than FTIR spectra |
| 3,4-dihydroxy-phenyl lactic acid (danshensu) in Pill containing Salviae miltiorrhizae Radix et Rhizoma Notoginseng Radix et Rhizoma and by NIR[12] | Spectral of wavenumber 7994-7258, 6135-5399 cm−1 is chosen to build PLS modeling after preprocessing of spectra. The 1st derivative spectra produce the best calibration model with a good evaluation result |
| Puerarin in Puerariae radix by NIR and UV spectrofotometry[13] | The best PLS model obtained from NIR-UV fusion data with MSC with the1st derivative and smoothing as preprocessing technique |
| Alkaloids BB, CO, PA, EP and JA in Coptidis rhizoma by NIR[14] | The best model for BB, CO and PA model built by ANN, while EP and JA by PLS. Different preprocessing spectra fits to different alkaloid |
| Isoflavon and saponin in Soybeans by NIR[15] | The MLR regression model of derivative spectra has a good evaluation value for total isoflavones, but the results are not satisfactory for total saponin |
| Stevioside, rebaudioside, and sum of them in Stevia rebaudiana Bertoni by NIR[16] | NIR spectra is preprocessed by smoothing, and then followed by the 1st and the 2nd derivative. The PLS regression is built in wavenumber region 4760-5016 cm−1. The results show that model evaluation values are not satisfactory due to low linearity and different value of RMSEC and RMSEP |
| Five total antraquinone (emodin, chrysophanol, rhein, aloeemodin, and physcion) in Rhei radix et rhizome by NIR[17] | Different preprocessing is applied to NIR spectra with 16 cm−1 resolution, and then the wavenumber regions at 4242-5581 cm−1, 5885-6233 cm−1, and 6394-7011 cm−1 are chosen to build a multivariate regression, namely SMLR, PCR, and PLS. The best model is obtained using PLS with the 1st derivative |
| Icariin in Epimedium species from Berberidaceae family by NIR[18] | NIR spectra with 8 cm−1 resolution is processed by the 1st derivative and smoothing, followed by CARS as variable selection technique. After that, the variable builds a PLS regression and results in good calibration and validation value |
| CO and SFO as adulterant in Extra virgin olive oil by FTIR[19] | The FTIR with resolution at 4 cm−1 spectra is processed by derivative. The best SFO model is obtained using PLS with the 1st derivative, while the best CO model is obtained using a PLS with no preprocessing. The models have good linearity, but significant differences in RMSEC and RMSEP |
| Millet (Eleusine cora-cana) and Buckwheat (Fagopyrum esculentum) as adulterant in Blackpepper (Piper nigrum) by NIR[20] | The best model is obtained by PLS regression of NIR with MSC, followed by the1st derivative preprocessing spectra. It has good R2 and RPD, but significant differences in RMSEC and RMSEP |
| Sibutramine and phenolptalein as adulterant in Weightloss herbal medicine by FTIR[21] | SNV and a Savitzky-Golay 2nd derivative are used as a preprocessing technique. The best model is obtained by FTIR. R2>0.93. Sibutamine’s model produces 0.8% RMSEC and RMSECV, and phenolphthalein 2.2% |
| Total flavone content in snow lotus (Saussurea involucrate) by NIR[22] | SNV is choosing as a preprocessing spectrum. Before constructing the calibration model, the spectral split into several intervals by iPLS than the variable of the split interval is choosing by GA. Comparing to other variable selection, iPLS-GA produces the lowest RMSEC and RMSEP |
| EGCG in green tea by FT-NIR[23] | SNV as preprocessing spectra and siPLS-GA as variable selection give the best result of a calibration model with few variables, lowest RMSEC and RMSEP, and highest R2 |
| Alpha-mangostin in mangosteen exctract by FTIR[24] | Wave number 3825-937 cm−1 was selected using TQ analystTM to build a prediction model using PLS. The validation model produces accurate and precise results for the determination of alpha-mangostin levels |
| Rosmarinic acid in rosmarini leaves by ATR-IR and NIR[25] | The whole spectrum is using to constructed PLS model pretreated by MSC followed with 2nd derivative method give the best evaluation value |
| BB in rhizoma coptidis by IR and NIR fusion[26] | SNV followed by 2nd derivative is the best spectra preprocessing method for analysis of BB with fusion strategy |
| Total flavonoid content in Ginkgo biloba leaf by NIR[27] | SNV is the superior pretreatment spectra technique among several other techniques applied in this research with the lowest RMSECV value |
PCA: Principal component analysis, RSD: Relative standard deviation, PLS: Partial least square, NIR: Near-infrared, UV: Ultraviolet, MSC: Multiplicative scatter correction, BB: Berberine, CO: Coptisine, PA: Palmatine, EP: Epiberberine, JA: Jatrorrhizine, ANN: Artificial neural network, MLR: Multiple linear regression, SMLR: Stepwise MLR, RMSEC: Root mean square error of calibration, RMSEP: Root mean square error of prediction, RMSECV: Root mean square error of cross validation, CARS: Competitive adaptive reweighted sampling, CO: Corn oil, SFO: Sunflower oil, SNV: Standard normal variate, iPLS: Interval PLS, GA: Genetic algorithm, EGCG: Epigallocatechin-3-gallate, siPLS: Synergy PLS, FTIR: Fourier Transform Infrared, RPD: Relative percent difference, IR: Infrared
QUANTITATIVE ANALYSIS USING INFRARED SPECTROSCOPY
Quantitative analysis with infrared spectroscopy is an indirect method because it still requires reference methods such as chromatographic (HPLC, GC and capillary electrophoresis), and spectroscopic (NMR, ultraviolet). To support the quantification of compound information in the sample, the relationship between Fourier transform infrared spectroscopy (FTIR) spectrum as a predictor variable, and the value of measurement level results with the reference method (response variable) will be translated by chemometrics techniques. In general, content analysis with infrared spectroscopy combined with chemometrics is carried out through several stages. First, the samples of spectra data are scanned using near-infrared (NIR) and mid-infrared (MIR) spectrometers. The sample is also analyzed by the reference method to get the actual value. This reference method is chosen based on the ability to quantify components in a sample accurately. Furthermore, infrared spectra can be treated further with preprocessing and variable selection. This aims to reduce the dimensions of the data to simplify calibration modeling without losing important information.[8]
Calibration is made by plotting the prediction variable (variable of infrared spectra) as x with the actual variable (compound content obtained from the reference method) as y. Calibration models for quantitative analysis are divided into two categories: linear and nonlinear. Generally, linear calibration that can be used is partial least square (PLS), principal component regression, and step-wise multiple linear regression, while nonlinear calibration is artificial neural network.[8] After that, the calibration model formed is evaluated to testify the ability in analyzing the levels of compounds in the sample.
QUANTIFICATION OF PHYTOCHEMICAL COMPOUND AND ADULTERANT
An analysis with infrared spectroscopy is ideal if the evaluation results show R2 values approaching a value of 1, low standard error (RMSEC, RMSECV, RMSEP, or predicted residual error sum square [PRESS]), and the relative percent difference (RPD) values should be at least 2.4 or greater.[28,29] R2 calibration values describe the linearity of the calibration curve formed from infrared spectrum data (x value) with the concentration data measured using the reference method (y value). R2 values close to 1 indicate that the infrared spectrum data are able to explain the concentration of compounds as the dependent variable. R2 validation value illustrates the accuracy of the measured concentration value by the infrared method. A validation value of R2 close to 1 indicates that the measurement of compound concentrations using the FTIR method yields a value that is proportional to the measurement results of the reference method. The RPD values were used to evaluate the fitting and prediction capacities of the models by calculating the standard deviation divided by standard error prediction. The RMSEC, RMSECV, RMSEP, or PRESS values are used to assess the precision. The analysis process using infrared spectroscopy can be seen in Figure 1.
Figure 1.
Quantitative analysis process using infrared spectroscopy
Spectra preprocessing
Spectra preprocessing in quantitative analysis aims to minimize the noise and the physical phenomena so that the resulting signal correlates with concentration that increases the predictive ability.[30] There are several ways of preprocessing the spectra. In this review, Savitzy-Golay-based derivatization, standard normal variate (SNV), multiplicative scatter correction (MSC), and scaling techniques are commonly used.[9,21,30]
Most of the analysis with plant materials using infrared spectroscopy is done on solid samples, making it susceptible to scattering.[30] The scattering causes irrelevant variations in the spectra data. If there are no preprocessing spectra, a mix of information and noise will occur, and this can cause a decrease in the predictive ability of a model. The scattering usually results in fewer MIR measurements than NIR. It is due to inhomogeneity of particle size in NIR measurement.[30] In research conducted by Otsuka, it was shown that the particle size was inversely proportional to the scattering coefficient.[31]
MSC and SNV are the examples of useful scattering correction techniques in reducing physical variability due to scattering. MSC and SNV often produce similar evaluation values.[13,17,20] MSC works to correct the spectrum by estimating the correction coefficient of the raw spectrum and calculating it with the average reference spectrum used in the calibration model. Since it uses the average spectrum, correction errors might occur if there is a dominant spectrum used as a reference spectrum.
Therefore, repeated application of MSC generally reduces spectrum correction errors. Nevertheless, it does not mean that the more MSC processes are corrected for the better, the repetition of MSC can reduce the difference between the spectra data sets used.[32] In using SNV, spectra correction is done by calculating the mean and deviation of the sample spectrum. Unlike MSC, spectra preprocessing using SNV does not require a reference spectrum to make corrections, so the process is simpler.[33] SNV works by normalizing the spectrum due to physical differences but not chemically.[34] SNV success occurs when the uniform scatters between samples in the full spectrum. If this condition does not occur, then the spectral correction will not be optimal.[35]
Savitzky-Golay is a derivative technique which has a mode for spectral smoothing.[36] The derivatization can provide a more detailed picture of the structure in the spectra to increase sensitivity. Spectra derivatization that is widely used in quantitative analysis of compounds in plants includes the 1st derivatives and the 2nd derivatives. The 1st derivative will remove baseline variation between the data, and the 2nd derivative will improve the resolution of spectra to remove any slope effect on the data.[18,21] Although derivatization generally provides a detailed description, not all analyzes with IR spectroscopy require derivatization for preprocessing spectra. Such is the case with research conducted by Kokalj Ladan et al., who conducted the derivatization of spectra with different resolutions. From these studies, it is known that derivatization is useful for improving prediction models in spectra with a resolution of 16 cm-1 but not at a resolution of 4 cm-1.[10] In this regard, lower resolutions can cause loss of useful information for quantitative analysis; therefore, derivatization will help in describing the spectra in more detail.[37] However, the spectra contain more noise in higher resolution so that when derivatization is carried out, it will cause too much noise in the data and decrease the signal to noise ratio. Thus, the standard error in making the model also does not improve or even increase.[10]
Variable selection
The selection of spectra preprocessing techniques is very important to determine a successful calibration model formation, because it can emerge or eliminate important information connected with the content measurement. However, the successful of spectra effect preprocessing on the calibration model formation can only be known after validating the model.[38] It is important to compare several spectral preprocessing techniques and their combinations to determine a more suitable technique for the analyzed data. Preprocessing techniques are appropriate if it decreases standard errors by reducing or minimally maintaining the complexity of the model.
Besides using spectral preprocessing, reducing the complexity of the model can also be done by selecting variables. From the summary of the research, it can be seen that the selection of variables using principal component analysis (PCA),[9] interval PLS (iPLS),[13] Competitive adaptive reweighted sampling (CARS),[18] or genetic algorithm (GA)[39] reduce the number of variables by eliminating irrelevant variables. PCA will change the original variables correlated into new variables that contain a combination of the original variables. This combination can reduce the number of variables, but still can explain most of the original variable information.[40,41]
CARS variable selection has three stages. The first step is a random sampling of data sets, and then a number of these variables will be reduced again using the exponential decrease function technique where the number of variables rapidly decreased. The remaining variables that produce low RMSECV will be selected as informative variables included in the calibration model.[42] This method can make the calibration model simpler and effective.[43] While GA is done by randomly selecting a subset of variables and then calibrating the PLS model to select the variable with the greatest influence. Furthermore, crossover and mutations are carried out to form new variables, and re-calibration is carried out to decide which variable is the most suitable.[44,45]
In addition to the selection based on wave number points, variable selection can also be done by separating the wave numbers at certain intervals. In this article, iPLS and synergy interval (siPLS) have been used. iPLS works by dividing the spectrum into several parts with the same interval. For each subinterval, submodels will be made and the standard error will be calculated. The submodel with the lowest standard error contains important information correlated with the objectives of the analysis. At this point, it will be chosen as a subinterval of wavenumbers, which will be used to form the calibration model.[46] While siPLS is a modification of iPLS to optimize variable selection with interval combinations containing informative variables.[47] siPLS can contain more informative variables because it does not only select one spectrum region but combines several.[48]
Variable selection can also be done by combining the selection of wave number points and wave number intervals such as siPLS-GA. Variable selection using combination techniques can produce a simpler model with fewer variables than without a combination. However, the smaller number of variables cannot be ascertained to have the best predictive ability, because of the loss of informative data.[49,50] Variable selection is useful for reducing noise and interfering variables in data so as to make prediction model formation more efficient with fewer variables but reliable prediction ability.[51]
A successful method in completing multivariate analysis is assessed by calibration and validation parameters. R2 values that are close to 1 indicate that the predictive variables used in making the model correlate linearly with the levels of compounds. In addition, a calibration model is good if there is no large difference between the RMSEC and root mean square error P values.[17] If this happens, it will be overfitting or underfitting. Overfitting is a situation where the calibration model produces a low standard error but unable to predict new samples and vice versa.
CONCLUSION
Infrared spectroscopy is useful in the quantification of phytochemical components and adulterants in plant-based medicines and supplements. This analysis can help further study to determine the quality of nutritious plants to ensure the safety and effectiveness of plant-based products. It is found that chemometrics can overcome the complexity of the chemical content in herbs. Optimization of preprocessing techniques and selection of variables and a combination of both are generally useful in improving predictive ability of models. This is indicated by an increase in linearity value (in calibration and validation) and a decrease in the standard error in the model by the spectrum that is experiencing preprocessing, variable selection, and a combination of both compared to raw spectra. The high R2 validation value (close to 1) indicates that the measurement of compound concentrations using the FTIR method yields a proportional accuracy to the measurement results of the reference method.
Financial support and sponsorship
Nil.
Conflicts of interest
There are no conflicts of interest.
REFERENCES
- 1.Zahn R, Perry N, Perry E, Mukaetova-Ladinska EB. Use of herbal medicines: Pilot survey of UK users' views. Complement Ther Med. 2019;44:83–90. doi: 10.1016/j.ctim.2019.02.007. [DOI] [PubMed] [Google Scholar]
- 2.Kucharska-Ambrożej K, Karpinska J. The application of spectroscopic techniques in combination with chemometrics for detection adulteration of some herbs and spices. Microchem J. 2020;153:1–9. [Google Scholar]
- 3.Chan Fung VS. Integrating traditional Chinese (herbal) medicines into risk based regulation - With focus on non-clinical requirements to demonstrate safety. J Tradit Chinese Med Sci. 2020;7:88–94. [Google Scholar]
- 4.Silva Elipe MV. Advantages and disadvantages of nuclear magnetic resonance spectroscopy as a hyphenated technique. Anal Chim Acta. 2003;497:1–25. [Google Scholar]
- 5.Upton R, David B, Gafner S, Glasl S. Botanical ingredient identification and quality assessment: Strengths and limitations of analytical techniques. Phytochem Rev. 2019:1–21. [Google Scholar]
- 6.Guo C, Luo X, Zhou X, Shi B, Wang J, Zhao J, et al. Quantitative analysis of binary polymorphs mixtures of fusidic acid by diffuse reflectance FTIR spectroscopy, diffuse reflectance FT-NIR spectroscopy, Raman spectroscopy and multivariate calibration. J Pharm Biomed Anal. 2017;140:130–6. doi: 10.1016/j.jpba.2017.02.053. [DOI] [PubMed] [Google Scholar]
- 7.Westad F, Marini F. Validation of chemometric models - A tutorial. Anal Chim Acta. 2015;893:14–24. doi: 10.1016/j.aca.2015.06.056. [DOI] [PubMed] [Google Scholar]
- 8.Yun YH, Li HD, Deng BC, Cao DS. An overview of variable selection methods in multivariate analysis of near-infrared spectra. TrAC - Trends Anal Chem. 2019;113:102–15. [Google Scholar]
- 9.Turner NW, Cauchi M, Piletska EV, Preston C, Piletsky SA. Rapid qualitative and quantitative analysis of opiates in extract of poppy head via FTIR and chemometrics: Towards in-field sensors. Biosens Bioelectron. 2009;24:3322–8. doi: 10.1016/j.bios.2009.04.030. [DOI] [PubMed] [Google Scholar]
- 10.Kokalj Ladan M, Straus J, Tavčar Benković E, Kreft S. FT-IR-based method for rutin, quercetin and quercitrin quantification in different buckwheat (Fagopyrum) species. Sci Rep. 7:7226. doi: 10.1038/s41598-017-07665-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Wulandari L, Retnaningtyas Y, Nuri, Lukman H. Analysis of flavonoid in medicinal plant extract using infrared spectroscopy and chemometrics. J Anal Methods Chem. 2016;2016:4696803. doi: 10.1155/2016/4696803. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Zhang S, Xiong H, Zhou L, Ju W, Yang Z, Yan K, et al. Development and validation of in-line near-infrared spectroscopy based analytical method for commercial production of a botanical drug product. J Pharm Biomed Anal. 2019;174:674–82. doi: 10.1016/j.jpba.2019.06.046. [DOI] [PubMed] [Google Scholar]
- 13.Wang Y, Yang Y, Sun H, Dai J, Zhao M, Teng C, et al. Application of a data fusion strategy combined with multivariate statistical analysis for quantification of puerarin in Radix puerariae. Vib Spectrosc. 2020;108:1–7. [Google Scholar]
- 14.Jintao X, Yufei L, Liming Y, Chunyan L, Quanwei Y, Weiying W, et al. Rapid and simultaneous analysis of five alkaloids in four parts of coptidis rhizoma by near-infrared spectroscopy. Spectrochim Acta A Mol Biomol Spectrosc. 2018;188:611–8. doi: 10.1016/j.saa.2017.07.053. [DOI] [PubMed] [Google Scholar]
- 15.Berhow MA, Singh M, Bowman MJ, Price NPJ, Vaughn SF, Liu SX. Quantitative NIR determination of isoflavone and saponin content of ground soybeans. Food Chem. 2020;317:126373. doi: 10.1016/j.foodchem.2020.126373. [DOI] [PubMed] [Google Scholar]
- 16.Martono Y, Trihandaru S, Rondonuwu FS. Determination of stevioside and rebaudioside a in stevia rebaudiana bertoni leaves using near infrared spectroscopy and multivariate data analysis. Indones J Chem. 2018;18:664–71. [Google Scholar]
- 17.Zhan H, Fang J, Wu H, Yang H, Li H, Wang Z, et al. Rapid determination of total content of five major anthraquinones in rhei radix et rhizoma by NIR spectroscopy. Chinese Herb Med. 2017;9:250–7. [Google Scholar]
- 18.Yang Y, Wu Y, Li W, Liu X, Zheng J, Zhang W, et al. Determination of geographical origin and icariin content of Herba Epimedii using near infrared spectroscopy and chemometrics. Spectrochim Acta A Mol Biomol Spectrosc. 2018;191:233–40. doi: 10.1016/j.saa.2017.10.019. [DOI] [PubMed] [Google Scholar]
- 19.Rohman A, Che Man YB. Quantification and classification of corn and sunflower oils as adulterants in olive oil using chemometrics and FTIR spectra. Sci World J. 2012;2012:250795. doi: 10.1100/2012/250795. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.McGoverin CM, September DJ, Geladia P, Manley M. Near infrared and mid-infrared spectroscopy for the quantification of adulterants in ground black pepper. J Near Infrared Spectrosc. 2012;20:521–8. [Google Scholar]
- 21.Rooney JS, McDowell A, Strachan CJ, Gordon KC. Evaluation of vibrational spectroscopic methods to identify and quantify multiple adulterants in herbal medicines. Talanta. 2015;138:77–85. doi: 10.1016/j.talanta.2015.02.016. [DOI] [PubMed] [Google Scholar]
- 22.Chen Q, Jiang P, Zhao J. Measurement of total flavone content in snow lotus (Saussurea involucrate) using near infrared spectroscopy combined with interval PLS and genetic algorithm. Spectrochim Acta A Mol Biomol Spectrosc. 2010;76:50–5. doi: 10.1016/j.saa.2010.02.045. [DOI] [PubMed] [Google Scholar]
- 23.Guo Z, Chen Q, Chen L, Huang W, Zhang C, Zhao C. Optimization of informative spectral variables for the quantification of EGCG in green tea using Fourier transform near-infrared (FT-NIR) spectroscopy and multivariate calibration. Appl Spectrosc. 2011;65:1062–7. doi: 10.1366/11-06287. [DOI] [PubMed] [Google Scholar]
- 24.Rohman A, Arifah FH, Irnawati, Alam G, Muchtaridi M. The application of FTIR spectroscopy and chemometrics for classification of mangosteen extract and its correlation with alpha-mangostin. J Appl Pharm Sci. 2020;10:149–54. [Google Scholar]
- 25.Kirchler CG, Pezzei CK, Beć KB, Henn R, Ishigaki M, Ozaki Y, et al. Critical evaluation of NIR and ATR-IR spectroscopic quantifications of rosmarinic acid in rosmarini folium supported by quantum chemical calculations. Planta Med. 2017;83:1076–84. doi: 10.1055/s-0043-107032. [DOI] [PubMed] [Google Scholar]
- 26.Qi L, Ma Y, Zhong F, Shen C. Comprehensive quality assessment for Rhizoma Coptidis based on quantitative and qualitative metabolic profiles using high performance liquid chromatography, Fourier transform near-infrared and Fourier transform mid-infrared combined with multivariate statistical analysis. J Pharm Biomed Anal. 2018;161:436–43. doi: 10.1016/j.jpba.2018.09.012. [DOI] [PubMed] [Google Scholar]
- 27.Zhao LJ, Liu W, Xiong SH, Tang J, Lou ZH, Xie MX, et al. Determination of total flavonoids contents and antioxidant activity of ginkgo biloba leaf by near-infrared reflectance method. Int J Anal Chem. 2018;2018:1–7. doi: 10.1155/2018/8195784. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Shenk JS, Landa I, Hoover MR, Westerhaus MO. Description and evaluation of near infrared reflectance spectro-computer for forage and grain analysis. Corp Sci. 1981;21:355–8. [Google Scholar]
- 29.Rohman A, Sudjadi Devi, Ramadhani D, Nugroho A. Analysis of curcumin in curcuma longa and Curcuma xanthorriza using FTIR spectroscopy and chemometrics. Res J Med Plant. 2015;9:179–86. [Google Scholar]
- 30.Rinnan A. Pre-processing in vibrational spectroscopy-when, why and how. Anal Methods. 2014;6:7124–9. doi: 10.1039/c3ay42270d. [DOI] [PubMed] [Google Scholar]
- 31.Otsuka M. Comparative particle size determination of phenacetin bulk powder by using Kubelka-Munk theory and principal component regression analysis based on near-infrared spectroscopy. Powder Technol. 2004;141:244–50. [Google Scholar]
- 32.Windig W, Shaver J, Bro R. Loopy MSC: A simple way to improve multiplicative scatter correction. Appl Spectrosc. 2008;62:1153–9. doi: 10.1366/000370208786049097. [DOI] [PubMed] [Google Scholar]
- 33.Alrezj OA, Patchava KC, Benaissa M, Alshebeili S. Coupling scatter correction with bandpass filtering for preprocessing in the quantitative analysis of glucose from near infrared spectra. Annu Int Conf IEEE Eng Med Biol Soc. 2017;2017:1800–3. doi: 10.1109/EMBC.2017.8037194. [DOI] [PubMed] [Google Scholar]
- 34.Sandak J, Sandak A, Meder R. Assessing trees, wood and derived products with near infrared spectroscopy: Hints and tips. J Near Infrared Spectrosc. 2016;24:485–505. [Google Scholar]
- 35.Bi Y, Yuan K, Xiao W, Wu J, Shi C, Xia J, et al. A local pre-processing method for near-infrared spectra, combined with spectral segmentation and standard normal variate transformation. Anal Chim Acta. 2016;909:30–40. doi: 10.1016/j.aca.2016.01.010. [DOI] [PubMed] [Google Scholar]
- 36.Savitzky A, Golay MJ. Smoothing and differrentiation of data by simplified least square procedure. Anal Chem. 1964;36:1627–39. [Google Scholar]
- 37.Chen Y, Thosar SS, Forbess RA, Kemper MS, Rubinovitz RL, Shukla AJ. Prediction of drug content and hardness of intact tablets using artificial neural network and near-infrared spectroscopy. Drug Dev Ind Pharm. 2001;27:623–31. doi: 10.1081/ddc-100107318. [DOI] [PubMed] [Google Scholar]
- 38.Jonsson H, Gabrielsson J. Evaluation of preprocessing methods. Compr Chemom. 2009;2:199–206. [Google Scholar]
- 39.Niazi A, Leardi R. Genetic algorithms in chemometrics. J Chemom. 2012;26:345–51. [Google Scholar]
- 40.Abdi H, Williams LJ. Principal component analysis. Wiley Interdiscip Rev Comput Stat. 2010;2:433–59. [Google Scholar]
- 41.Jolliffe IT, Cadima J. Principal component analysis: A review and recent developments. Philos Trans A Math Phys Eng Sci. 016;;74:1–16. doi: 10.1098/rsta.2015.0202. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Li H, Liang Y, Xu Q, Cao D. Key wavelengths screening using competitive adaptive reweighted sampling method for multivariate calibration. Anal Chim Acta. 2009;648:77–84. doi: 10.1016/j.aca.2009.06.046. [DOI] [PubMed] [Google Scholar]
- 43.Wang LL, Lin YW, Wang XF, Xiao N, Xu Y Da, Li HD, et al. A selective review and comparison for interval variable selection in spectroscopic modeling. Chemom Intell Lab Syst. 2018;172:229–40. [Google Scholar]
- 44.Jiang H, Xu W, Chen Q. Comparison of algorithms for wavelength variables selection from near-infrared (NIR) spectra for quantitative monitoring of yeast (Saccharomyces cerevisiae) cultivations. Spectrochim Acta A Mol Biomol Spectrosc. 2019;214:366–71. doi: 10.1016/j.saa.2019.02.038. [DOI] [PubMed] [Google Scholar]
- 45.Arakawa M, Yamashita Y, Funatsu K. Genetic algorithm-based wavelength selection method for spectral calibration. J Chemom. 2011;25:10–9. [Google Scholar]
- 46.Norgaard L, Saudland A, Wagner J, Nielsen JP, Munck L, B ES. Interval partial least-squares regression (iPLS): A comparative chemometric study with an example from near-infrared spectroscopy. Appl Spectrosc. 2000;54:413–9. [Google Scholar]
- 47.Li Y, Guo M, Shi X, Wu Z, Li J, Ma Q, et al. Online near-infrared analysis coupled with MWPLS and SiPLS models for the multi-ingredient and multi-phase extraction of licorice (Gancao) Chin Med. 2015;10:38. doi: 10.1186/s13020-015-0069-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Duarte LM, Paschoal D, Izumi CMS, Dolzan MD, Alves VR, Micke GA, et al. Simultaneous determination of aspartame, cyclamate, saccharin and acesulfame-K in powder tabletop sweeteners by FT-Raman spectroscopy associated with the multivariate calibration: PLS, iPLS and siPLS models were compared. Food Res Int. 2017;99:106–14. doi: 10.1016/j.foodres.2017.05.006. [DOI] [PubMed] [Google Scholar]
- 49.Zhu J, Ahmad W, Xu Y, Liu S, Chen Q, Hassan MM, et al. Development of a novel wavelength selection method for the trace determination of chlorpyrifos on Au@Ag NPs substrate coupled surface-enhanced Raman spectroscopy. Analyst. 2019;144:1167–77. doi: 10.1039/c8an02086h. [DOI] [PubMed] [Google Scholar]
- 50.Zhu J, Agyekum AA, Kutsanedzie FY, Li H, Chen Q, Ouyang Q, et al. Qualitative and quantitative analysis of chlorpyrifos residues in tea by surface-enhanced Raman spectroscopy (SERS) combined with chemometric models. Lwt Food Sci Technol. 2018;97:760–9. [Google Scholar]
- 51.Guyon I, Elisseeff A. An introduction to variable and feature selection. J Machine Learn Res. 2003;3:1157–82. [Google Scholar]

