Based on near-infrared spectroscopy and chemometrics to rapidly evaluate the adulteration of Ganoderma lingzhi powder

doi:10.21203/rs.3.rs-5308313/v1

Download PDF

Research Article

Based on near-infrared spectroscopy and chemometrics to rapidly evaluate the adulteration of Ganoderma lingzhi powder

https://doi.org/10.21203/rs.3.rs-5308313/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Ganoderma lingzhi, the dry fruiting bodies of G. lucidum or G. sinensis, is a microbial food of high nutritional and health value. It is expensive but in high demand. In pursuit of higher profits, counterfeit products adulterated with G. lingzhi, such as G. applanatum, have appeared in the market. To identify the authenticity and forecast the degree of adulteration in Ganoderma lingzhi powder rapidly and non-destructively, the combination of near-infrared spectroscopy (NIRS) and chemometrics was used. Partial least squares discriminant analysis (PLS-DA), back propagation neural network (BPNN), support vector machine (SVM), and random forest (RF) were adopted as qualitative identification of G. lingzhi authenticity model methods, and partial least-squares (PLS) was developed as a quantitative prediction of adulteration content. Preprocessing and feature variables selection methods were developed to optimize the model and screen the best model. Among these experimental approaches, PLS-DA + first-order derivatives (D1), SVM + D1 + Competitive adaptive reweighted sampling (CARS), RF + standard normal variate transform (SNV) and BPNN + D1 + Uninformative variable elimination (UVE) + CARS achieved 100% classification accuracy. SVM + second-order derivatives (D2) + CARS and BPNN + D2 + CARS identified all adulterated G. lucidum, PLS-DA + D1 + UVE + CARS, RF + D2 + Genetic algorithm (GA), SVM + D2 + GA, and BPNN + D2 + CARS could distinguish all adulterated G. sinensis effectively.

Chemometrics

First-order derivatives

Ganoderma lingzhi

Identification

Second-order derivatives

Ganoderma lingzhi is the dry fruiting bodies of G. lucidum or G. sinensis. As a traditional Chinese health food, G. lingzhi has a long history of consumption in China and has been regarded as a miraculous treasure for nourishing, strengthening, invigorating qi, calming nerves, and supporting the essence of the body by successive generations of medicine practitioners (Wu et al., 2019; Yao et al., 2023). G. lingzhi mainly distributes in China, Japan, and the Korean Peninsula (He et al., 2023). G. lingzhi is expensive and usually sold in powder form for better absorption when consumed, but the powdered product is difficult to distinguish from the authenticity in terms of appearance. As a consequence, many illicit merchants mix other cheap powders of similar appearances, such as G. applanatum, G. bamboos, and Coriolus versicolor with G. lingzhi powder to reduce costs (Lai, 2018). Among these forgeries, G. applanatum is the most common. This seriously damaged the legitimate rights and interests of consumers and disturbed the market order. Therefore, the qualitative and quantitative detection of G. lingzhi adulteration is very necessary. At present, microscopic identification, character identification, physicochemical identification, thin layer chromatography, nuclear magnetic resonance, and biomolecular methods (Lai, 2018; Kozarski et al., 2012; Shao et al., 2019) can be used to identify G. lingzhi. However, experts with practical experience to identify, consuming a large number of samples, and taking a long time are required. The near-infrared spectroscopy (NIRS) has the benefits of being high efficiency, accuracy, simplicity of operation, non-destructive samples, and non-polluting environment (Pita-Calvo et al., 2017). It can be used to evaluate the quality of traditional foods in both qualitative and quantitative terms. There are many precedents for combining NIR techniques with chemometrics in food quality control, such as ginger powder (Yu et al., 2022), cotton (Mata et al., 2022), Crataegus pinnatifida Bunge (Ye et al., 2023), pistachio (Genis et al., 2021), coconut powder (Pandiselvam et al., 2022), saffron (Li et al., 2020), and Anoectochilus roxburghii (Li et al., 2018). In addition, the 2020 edition of the Pharmacopoeia of the People's Republic of China (I) stipulates that the dried subentities of G. sinensis and G. lucidum are the sources of G. lingzhi herbs, but preliminary literature research revealed that the chemical composition of these two species of G. lingzhi differed greatly. The chemical composition of G. lucidum is mainly highly oxidized lanolin triterpenoids, while in G. sinensis it is mainly aromatic terpenoids, ergocalciferol, and alkaloids (Luo et al., 2022; Boh et al., 2007). Thus, in this investigation, the species differentiation of G. lucidum and G. sinensis was also carried out using the NIRS technique in combination with chemometrics.

In this research, the qualitative identification model of G. lingzhi species and adulteration as well as a quantitative prediction model of adulteration level were established. In addition, preprocessing and feature variables selection methods were developed to optimize the model and screen the best model, which is conducive to safeguarding the quality and efficacy of G. lingzhi, improving the quality standard system of G. lingzhi, and safeguarding the legitimate rights and interests of consumers.

2.1. Sample Preparation

G. applanatum was collected from Zhejiang. A total of 19 batches of G. sinensis were collected from Anhui (4), Fujian (2), Hainan (4), Jilin (5) and Yunnan (4) provinces. And 50 batches of G. lucidum were collected from Anhui (8), Fujian (1), Guangdong (1), Hainan (4), Hubei (4), Jilin (17), Shandong (2), Yunnan (8) and Zhejiang (5) provinces. All samples were sliced, powdered, dried to constant weight and sieved through a 20-mesh sieve. A total of 414 samples were prepared according to the adulteration ratios of 0%, 10%, 20%, 40%, 60%, and 80%. The samples were dried to constant weight in a thermostat at 50°C and placed in a desiccator for storage.

2.2. Data collection

At least 2 g samples were packed in a circular sample cup with a diameter of 40 mm and a height of 25 mm to ensure that the sample covered the bottom evenly. The sample was analyzed in diffuse reflectance mode with an AntarisII Fourier transform NIR spectrometer (Thermo Fisher Scientific, America). The reference spectrum was the built-in background of the instrument. The sample was scanned 32 times with the nominal resolution set to 8 cm^− 1 and 1557 wave numbers from 4000–10000 cm^− 1 were obtained and repeated 3 times and the average value was taken for subsequent data analysis.

2.3. Chemometrics analysis

2.3.1. Spectra preprocessing

The acquisition of spectra is easily affected by external factors, which are often doped with unnecessary, irrelevant information and noise. It is required to preprocess the original spectrum because these factors make NIRS spectral instability and baseline drift more likely. The spectral data were first-order derivatives (D1), Savitzky-Golay (SG) smoothing, second-order derivatives (D2), standard normal variate transform (SNV), and multiplicative scattering correction (MSC) preprocessed to increase the stability of the model (Shi et al., 2022). It is worth noting that before preprocessing, the synthetic minority over-sampling technique (SMOTE) was adopted to maintain the numerical balance between classes. The problem is that a small minority, usually a data-deficient group, tends to get most of the attention. One way to deal with imbalance is to generate additional data from a few classes to make up for their data shortage. The SMOTE is one of the main approaches in the literature to achieve such additional sample generation for data balancing (Elreedy and Atiya, 2019). The dataset is then divided into training and prediction sets in a ratio of 4:1 using the joint x-y distance (SPXY) algorithm, which achieves dataset partitioning by comparing the spatial distances between the dependent variables to the spatial distances between the independent variables. The SPXY algorithm takes into account both the independent variables (x) and the dependent variables (y) in the partitioning of the samples, which can lead to a more efficient distribution of the calibration samples in a multidimensional space, by simultaneous calculation of spectral and compound distances between samples, the samples can be partitioned more rationally. (Li et al., 2021; Shao et al., 2022).

2.3.2. Screening of characteristic variables

Selecting variables before data processing can filter the spectral regions required for modeling, eliminate redundant variables in the spectrum, reduce modeling variables, and simplify the model. Thus, the feature variable selection method was adopted in our study (Jiang et al. 2022). Competitive adaptive reweighted sampling (CARS), successive projection algorithm (SPA), genetic algorithm (GA), and uninformative variable elimination (UVE) are four commonly used feature variable selection methods. CARS (Li et al., 2009) is based on Darwin's theory of evolution and reports the selection of variables depending on the survival of the fittest. The CARS algorithm is designed to obtain a subset of wave numbers using exponential decay functions and adaptive weighted sampling based on meritocracy and Monte Carlo sampling (Xu et al., 2023). SPA (Araújo et al., 2001) selects the feature variables in two main procedures. First, the projection of each wavelength dot on the remaining wavelengths is calculated and the wavelength dot with the largest projection value is chosen. Second, the projection process in the first process is duplicated until the desired number of wavelengths is selected and the selected wavelength is the smallest (Yuan et al., 2021). GA (Jouan-Rimbaud et al., 1996) uses probabilistic and non-local search procedures that are motivated by Darwinian natural selection theory. The binary code for each feature shows whether the feature was picked or not, and in GA, subsets of features are represented as binary strings called chromosomes. The GA enables a population of multiple individuals to evolve to a maximal adaptive state under a predetermined selection technique in order to establish the suitable wave number of features (Maraphum et al., 2022). UVE (Daszykowski et al., 2005) is a variable selection technique for stability analysis based on PLS regression coefficients. It is employed to stop reporting spectral variables or variables with redundant or pointless information. Moreover, in this research, the usage of UVE did not result in a significant reduction of the feature variables. Therefore, feature variable extraction on top of UVE was performed to extract fewer feature variables while the performance of the model remained unchanged or even improved (Huang et al., 2017).

2.3.3. Qualitative analysis

In this project, partial least squares discriminant analysis (PLS-DA), back propagation neural network (BPNN), support vector machine (SVM), and random forest (RF) were used for the qualitative identification of G. lingzhi species and whether they are adulterated or not. In order to ensure the accuracy and reliability of the data, all the methods were run five times, and the average accuracy of the training set and the prediction set was used as the evaluation index. Moreover, it can prevent the different wavelengths selected in one operation from correctly reflecting the overall situation and avoiding the error caused by chance on the results (Cai et al., 2023).

PLS-DA is a classification model based on a conjunction of linear discriminant analysis based on Bayes' theorem and the PLS algorithm. The method can predict quantitative variables of a sample using a small number of potential variables that represent the most informative variables explaining the covariance of predictors and responses. Spectral data usually have high dimensionality, so the method is useful for dealing with them (Quan et al., 2023).

RF is one of the most widely used classification algorithms. It consists of many decision trees (called weak learners) training in parallel to predict the classification results based on the majority voting scores. By using an ensemble of decision trees, RF overcomes the drawbacks of decision tree methods. Theoretically, the approach would allow RF to outperform individual decision trees in terms of classification accuracy and generalization ability (Karabadji et al., 2023).

SVM is a supervised machine learning approach based on statistical learning theory, which is usually applied to pattern recognition, classification, and regression analysis. SVM is widely used in the classification of high-dimensional data with limited training sets because their classification accuracy does not decrease with the number of wavelengths (Zhang et al., 2021).

BPNN is an excellent chemometric measure for processing nonlinear data. The optimal parameters of BPNN are selected in terms of the number of hidden nodes, learning rate, and the number of iterations with the criterion of minimum prediction error (He et al., 2023). Each methodology was preprocessed, and the feature variable was filtered to select the best combination of modeling methods for each task.

2.3.4. Quantitative analysis

In this experiment, the contents of G. applanatum in the samples were determined using a PLS regression model. Similar to the qualitative analysis process, PLS was also pre-treated and screened for characteristic variables, and then 10-fold cross-validation was introduced to validate the model. The metrics for evaluating the PLS model are chosen as the correlation coefficient (R²_C), root mean square error (RMSEP) of the cross-validation set, the correlation coefficient (R²_P), and root mean square error (RMSEP) of the prediction set. Generally, the closer the R² is to 1 as well as the closer the RMSE is to 0, the more effective the model is (Rossi and Lozano, 2020).

2.4. Software and algorithms

Unscrambler X 10.4 software was used to convert the spectral data format. The spectral data collected from each sample after scanning through the NIR instrument was imported into this software, then converted into the format and finally exported to an excel sheet. The rest of the operations were realized on MATLAB 2022a software. In this software, the required codes were opened and the names of the converted data tables were added to each code. Combining different modeling methods, for example, the data preprocessing is carried out first by using PLS-DA modeling method. After which the best preprocessing is selected, based on which the selection of feature variables for different algorithms is carried out, and finally the best algorithmic model is arrived at, and the others are followed in the same order. The same methodology was used for all three tasks, except that the models and algorithms used for the two qualitative tasks and one quantitative task differed.

3.1. Sample and spectral analysis

Dried samples of G. lucidum (Fig. 1A), G. sinense (Fig. 1B), and G. applanatum powder (Fig. 1C) were very similar in appearance. It was difficult for ordinary consumers to verify the authenticity by their appearance. Figure 2 showed that despite the different adulteration levels of G. lingzhi powder, the positions of the peaks and the absorbance curves follow a similar trend. In the bands 4000–5000, 7000–7500, and 8000–8500 cm^− 1, the absorption values of the peaks were significantly different, which may be an important factor in distinguishing G. lucidum, G. sinense, and adulterated G. lingzhi. The 7000–7500 cm^− 1 region is associated with the C-H first overtone stretching vibrational modes in the CH₃ and CH₂ groups, while the bands located among 4000–5000 cm^− 1 are from the combined C-H bands that are characteristic of proteins, amino acids, and carbohydrates. And the band 8000–8500 may be caused by C-H second overtone stretch vibrations in -CH₃ (Tahir et al., 2023; Chen et al., 2022). Meanwhile, the graph reveals that the absorption of curve A was higher than that of curve B. This may be due to that G. lucidum contains more triterpenoids and G. sinensis contains more aromatic terpenoids, triterpenoids had stronger absorption ability for NIRS than aromatic terpenoids, whereas the NIRS curves of G. lingzhi doped with G. applanatum are lower than the pure samples. Obviously, no significant differences can be observed by the naked eye from the original spectra. Therefore, it was necessary to use chemometric methods to identify doped samples and gauge the doping level.

3.2. Qualitative analysis

In this experiment, identifications of G. sinensis, G. lucidum, and adulteration were done. Different combinations of modeling methods, including PLA-DA, RF, SVM, and BPNN, as well as how various spectrum preprocessing techniques affected the models were compared. The final table was obtained by SG, D1, D2, SNV, and MSC preprocessing (Table 1).

Table 1

Classification results of PLS-DA, RF, SVM, and BPNN models combining different preprocessing, feature variable selection.
Mission	Combination of modeling methods	Number of characteristic variables	Training set correct rate	Prediction set correct rate
Identification of G. sinensis and G. lucidum species	PLS-DA original data	1557	95.00	90.00
	PLS-DA + SG	1557	97.50	95.00
	PLS-DA + SNV	1557	98.75	95.00
	PLS-DA + MSC	1557	98.75	95.00
	PLS-DA + D1	1557	100.00	100.00
	PLS-DA + D2	1557	100.00	95.00
	PLS-DA + D1 + CARS	54	100.00	100.00
	PLS-DA + D1 + SPA	99	100.00	90.00
	PLS-DA + D1 + GA	94	100.00	100.00
	PLS-DA + D1 + UVE	1087	100.00	100.00
	PLS-DA + D1 + UVE + CARS	62	100.00	100.00
	PLS-DA + D1 + UVE + SPA	99	100.00	90.00
	PLS-DA + D1 + UVE + GA	42	100.00	100.00
	RF original data	1557	100.00	95.00
	RF + SG	1557	100.00	85.00
	RF + SNV	1557	100.00	100.00
	RF + MSC	1557	100.00	95.00
	RF + D1	1557	100.00	85.00
	RF + D2	1557	100.00	95.00
	RF + SNV + CARS	16	100.00	75.00
	RF + SNV + SPA	99	100.00	90.00
	RF + SNV + GA	19	100.00	90.00
	RF + SNV + UVE	1010	100.00	95.00
	RF + SNV + UVE + CARS	17	100.00	85.00
	RF + SNV + UVE + SPA	99	100.00	90.00
	RF + SNV + UVE + GA	36	100.00	85.00
	SVM original data	1557	93.75	70.00
	SVM + SG	1557	90.00	85.00
	SVM + SNV	1557	93.75	90.00
	SVM + MSC	1557	96.25	80.00
	SVM + D1	1557	100.00	100.00
	SVM + D2	1557	100.00	90.00
	SVM + D1 + CARS	54	100.00	100.00
	SVM + D1 + SPA	99	100.00	100.00
	SVM + D1 + GA	94	100.00	100.00
	SVM + D1 + UVE	1087	100.00	100.00
	SVM + D1 + UVE + CARS	62	100.00	100.00
	SVM + D1 + UVE + SPA	99	100.00	95.00
	SVM + D1 + UVE + GA	42	97.50	95.00
	BPNN original data	1557	98.75	95.00
	BPNN + SG	1557	98.75	90.00
	BPNN + SNV	1557	95.00	85.00
	BPNN + MSC	1557	96.25	95.00
	BPNN + D1	1557	100.00	100.00
	BPNN + D2	1557	98.75	95.00
	BPNN + D1 + CARS	54	98.75	95.00
	BPNN + D1 + SPA	99	97.50	95.00
	BPNN + D1 + GA	94	100.00	95.00
	BPNN + D1 + UVE	1087	100.00	95.00
	BPNN + D1 + UVE + CARS	62	100.00	100.00
	BPNN + D1 + UVE + SPA	99	98.75	95.00
	BPNN + D1 + UVE + GA	42	95.00	95.00
Whether G. lucidum is adulterated	PLS-DA original data	1557	90.25	86.00
	PLS-DA + SG	1557	89.75	83.00
	PLS-DA + SNV	1557	88.00	80.00
	PLS-DA + MSC	1557	86.50	81.00
	PLS-DA + D1	1557	100.00	92.00
	PLS-DA + D2	1557	100.00	98.00
	PLS-DA + D2 + CARS	168	100.00	99.00
	PLS-DA + D2 + SPA	81	97.75	89.00
	PLS-DA + D2 + GA	94	95.75	91.00
	PLS-DA + D2 + UVE	235	100.00	93.00
	PLS-DA + D2 + UVE + CARS	58	96.50	95.00
	PLS-DA + D2 + UVE + SPA	84	99.00	93.00
	PLS-DA + D2 + UVE + GA	83	97.50	92.00
	RF original data	1557	100.00	84.00
	RF + SG	1557	99.75	85.00
	RF + SNV	1557	100.00	79.00
	RF + MSC	1557	100.00	86.00
	RF + D1	1557	100.00	85.00
	RF + D2	1557	100.00	97.00
	RF + D2 + CARS	168	100.00	98.00
	RF + D2 + SPA	81	100.00	96.00
	RF + D2 + GA	94	100.00	94.00
	RF + D2 + UVE	235	100.00	93.00
	RF + D2 + UVE + CARS	58	100.00	97.00
	RF + D2 + UVE + SPA	84	100.00	99.00
	RF + D2 + UVE + GA	83	100.00	97.00
	SVM original data	1557	82.00	78.00
	SVM + SG	1557	85.00	79.00
	SVM + SNV	1557	87.50	81.00
	SVM + MSC	1557	88.00	81.00
	SVM + D1	1557	100.00	93.00
	SVM + D2	1557	100.00	98.00
	SVM + D2 + CARS	168	100.00	100.00
	SVM + D2 + SPA	81	97.00	91.00
	SVM + D2 + GA	94	97.25	92.00
	SVM + D2 + UVE	235	100.00	99.00
	SVM + D2 + UVE + CARS	58	97.50	96.00
	SVM + D2 + UVE + SPA	84	99.50	95.00
	SVM + D2 + UVE + GA	83	98.25	97.00
	BPNN original data	1557	92.50	85.00
	BPNN + SG	1557	92.25	89.00
	BPNN + SNV	1557	98.00	96.00
	BPNN + MSC	1557	98.50	94.00
	BPNN + D1	1557	98.25	94.00
	BPNN + D2	1557	99.25	95.00
	BPNN + D2 + CARS	168	100.00	100.00
	BPNN + D2 + SPA	81	98.25	95.00
	BPNN + D2 + GA	94	94.50	91.00
	BPNN + D2 + UVE	235	95.50	86.00
	BPNN + D2 + UVE + CARS	58	97.50	93.00
	BPNN + D2 + UVE + SPA	84	98.25	93.00
	BPNN + D2 + UVE + GA	83	98.00	94.00
Whether G. sinensis is adulterated	PLS-DA original data	1557	85.06	81.58
	PLS-DA + SG	1557	88.96	73.68
	PLS-DA + SNV	1557	85.71	76.32
	PLS-DA + MSC	1557	85.06	81.58
	PLS-DA + D1	1557	100.00	100.00
	PLS-DA + D2	1557	100.00	98.68
	PLS-DA + D1 + CARS	113	100.00	100.00
	PLS-DA + D1 + SPA	114	100.00	84.21
	PLS-DA + D1 + GA	109	100.00	92.11
	PLS-DA + D1 + UVE	249	100.00	92.11
	PLS-DA + D1 + UVE + CARS	43	100.00	100.00
	PLS-DA + D1 + UVE + SPA	114	100.00	92.11
	PLS-DA + D1 + UVE + GA	40	98.05	97.37
	RF original data	1557	100.00	86.84
	RF + SG	1557	99.35	84.21
	RF + SNV	1557	100.00	68.42
	RF + MSC	1557	100.00	71.05
	RF + D1	1557	100.00	94.74
	RF + D2	1557	100.00	100.00
	RF + D2 + CARS	129	100.00	100.00
	RF + D2 + SPA	114	100.00	100.00
	RF + D2 + GA	86	100.00	100.00
	RF + D2 + UVE	83	100.00	97.37
	RF + D2 + UVE + CARS	34	100.00	97.37
	RF + D2 + UVE + SPA	29	100.00	96.52
	RF + D2 + UVE + GA	24	100.00	94.74
	SVM original data	1557	74.03	71.05
	SVM + SG	1557	72.73	68.42
	SVM + SNV	1557	84.42	81.58
	SVM + MSC	1557	84.42	78.95
	SVM + D1	1557	100.00	97.37
	SVM + D2	1557	100.00	100.00
	SVM + D2 + CARS	129	100.00	100.00
	SVM + D2 + SPA	114	100.00	89.47
	SVM + D2 + GA	86	100.00	100.00
	SVM + D2 + UVE	83	100.00	97.37
	SVM + D2 + UVE + CARS	34	100.00	94.74
	SVM + D2 + UVE + SPA	29	98.56	97.35
	SVM + D2 + UVE + GA	24	99.35	97.37
	BPNN original data	1557	92.21	86.84
	BPNN + SG	1557	97.40	86.84
	BPNN + SNV	1557	98.70	92.11
	BPNN + MSC	1557	91.56	81.58
	BPNN + D1	1557	94.81	78.95
	BPNN + D2	1557	100.00	100.00
	BPNN + D2 + CARS	129	100.00	100.00
	BPNN + D2 + SPA	114	94.81	89.47
	BPNN + D2 + GA	86	93.51	92.11
	BPNN + D2 + UVE	83	98.70	94.74
	BPNN + D2 + UVE + CARS	34	99.35	97.37
	BPNN + D2 + UVE + SPA	29	99.78	97.37
	BPNN + D2 + UVE + GA	24	100.00	97.37

For the species discrimination task of G. sinensis and G. lucidum, we found that the better models were PLS-DA + D1, RF + SNV, SVM + D1, and BPNN + D1, which had 100% correctness in both the training and validation sets. And on this basis, four algorithms, CARS, SPA, GA, and UVE were applied for characteristic variable selection. Owing to the unsatisfactory results of feature variable selection combined with the UVE algorithm (the number of variables screened out was too numerous), the associations with CARS, SPA, and GA algorithms on the basis of UVE were performed. It was found that the number of variables screened out by the association on the foundation of UVE would be reduced. The five preprocessing methods combined with RF were the least effective, probably because too much useful information was removed, resulting in too many feature variable selections and not as high a correct rate as the other three models. So, the best model in the variety identification task was PLS-DA + D1 + UVE + GA.

In order to determine whether G. lucidum was adulterated, the preprocessing correctness was ranked as D2 > D1 > MSC > SNV > SG among the four modeling methods. The best feature variable selection method was UVE coupled with the SPA algorithm, which used RF modeling and combined with D2 to select 84 feature variables (Fig. 3B(b)), which was 1473 less than the full wavelength. However, the effect of feature variables selection using a combination based on the UVE algorithm, despite a slight increase in the correct rate and the number of variables, was not better than the effect of the other algorithms alone. The effect of using CARS alone is better, so the best methods were selected as SVM + D2 + CARS and BPNN + D2 + CARS.

During the task of judging whether G. sinensis was adulterated, the preprocessing selection of D1 and D2 had obvious advantages and the accuracy was much higher than other preprocessing methods could be observed. In the PLS-DA model, the most suitable feature variable selection mean is UVE + CARS (Fig. 3C(a)). While in the RF and SVM models, the most appropriate feature variable selection approach was GA (Fig. 3C(b); Fig. 3C(c)), and in the BPNN model, the most effective feature variable selection choice was CARS (Fig. 3C(d)). It can also be noticed that other algorithms used alone such as SPA and UVE also had higher accuracy, comparing UVE jointly with the other three algorithms, the number of feature variables filtered out using alone was higher, but the correct rate was improved. Therefore, using one algorithm alone was better than jointly to determine whether G. sinensis was adulterated, which coincided with the discoveries of earlier research (Gao & Xu, 2022).

Overall, the UVE algorithm was more suitable for identifying species in conjunction with other feature variable selection algorithms, while identifying whether to adulterate was more suitable for combining a feature wavenumbers selection method alone, which was to be expected since different algorithms had different efficiency and adaptability in utilizing different tasks and information (Jiang et al., 2023).

3.3. Quantitative analysis

3.3.1. Quantitative analysis with PLS

The PLS regression method can accurately ascertain the amount of G. applanatum in adulterated samples. Different pretreatment methods had different effects on the PLS regression models. Various pretreatment methods were used to obtain the most suitable model for detecting adulterated content.

For the task of predicting G. lucidum powder adulteration levels (Table 2), MSC appeared to be the best pretreatment, as the model using MSC had the highest R²_P value (0.93) and the lowest RMSEP (0.11) among the five pretreatments, while R²_C and RMSEP were very close to the values of the other pretreatments, which indicated the strong predictive power and stability of the MSC model. The best combination of modeling methods was PLS + MSC + UVE + GA, 57 wavenumbers were screened (Fig. 4A). And the values of R²_P were preferentially compared when the RMSEP, R²_C, RMSEP, and R²_P of all modeling methods did not differ significantly, at which point R²_P (0.91) was the highest among all combinations. In general, the accuracy of the regression model is considered excellent when R² ≥ 0.90 (Wang et al., 2022).

Table 2

The quantitative results of PLS with different spectral pretreatment methods (*G. lucidum* mixed with *G. applanatum*) and (*G. sinense* mixed with *G. applanatum*)
Mission	Combination of modeling methods	Number of characteristic variables	Training set results		Prediction set results
Mission	Combination of modeling methods	Number of characteristic variables	RMSECV	R_C²	RMSEP	R_P²
Prediction of G. lucidum adulteration levels	PLS original data	1557	0.08	0.96	0.13	0.90
	PLS + SG	1557	0.08	0.96	0.12	0.91
	PLS + SNV	1557	0.08	0.96	0.11	0.89
	PLS + MSC	1557	0.09	0.95	0.11	0.93
	PLS + D1	1557	0.09	0.95	0.13	0.91
	PLS + D2	1557	0.09	0.95	0.13	0.88
	PLS + MSC + CARS	18	0.08	0.96	0.14	0.85
	PLS + MSC + SPA	76	0.08	0.95	0.13	0.90
	PLS + MSC + GA	51	0.10	0.93	0.15	0.85
	PLS + MSC + UVE	1319	0.11	0.92	0.12	0.88
	PLS + MSC + UVE + CARS	36	0.10	0.93	0.15	0.86
	PLS + MSC + UVE + SPA	76	0.11	0.93	0.13	0.88
	PLS + MSC + UVE + GA	57	0.11	0.92	0.13	0.91
Prediction of G. sinensis adulteration levels	PLS original data	1557	0.04	0.85	0.08	0.47
	PLS + SG	1557	5.32*10^− 5	1.00	0.14	0.86
	PLS + SNV	1557	2.95*10^− 5	1.00	0.14	0.87
	PLS + MSC	1557	4.54*10^− 5	1.00	0.14	0.84
	PLS + D1	1557	*4.0110**^− 5	1.00	0.14	0.92
	PLS + D2	1557	3.70*10^− 5	1.00	0.15	0.87
	PLS + D1 + CARS	44	*3.6310**^− 5	1.00	0.14	0.91
	PLS + D1 + SPA	115	4.89*10^− 5	1.00	0.13	0.89
	PLS + D1 + GA	78	2.94*10^− 5	1.00	0.16	0.86
	PLS + D1 + UVE	215	4.23*10^− 5	1.00	0.14	0.84
	PLS + D1 + UVE + CARS	32	3.52*10^− 5	1.00	0.18	0.70
	PLS + D1 + UVE + SPA	114	4.42*10^− 5	1.00	0.11	0.90
	PLS + D1 + UVE + GA	8	1.92*10^− 5	1.00	0.17	0.73

In the task of predicting the adulteration concentration of G. sinensis powder, the R²_C of models built by preprocessing such as D1, D2, SNV, MSC, and SG were similar and were all 1.00, indicating their high correlation. D1 had the best prediction accuracy with smaller RMSEP values and significantly higher R²_P values than the other preprocessing methods. In general, D1's preprocessing method showed excellent performance, and different preprocessing methods had different effects on different spectral datasets and the derivative preprocessing method worked best in the PLS model of G. sinensis. This result was consistent with previous studies (Li et al., 2020). PLS + D1 + CARS was the best modeling combination (Table 2, Fig. 4B), and other combinations could not guarantee prediction accuracy despite the smaller number of feature variables selected.

In this investigation, NIRS combined with the chemometric method was used for the qualitative and quantitative detection of G. lingzhi powder containing G. applanatum, and several data models were developed to find out the most suitable one. The results demonstrated that PLS-DA combined with D1 was the most effective method in correctly classifying the samples, with 100% correctness in both the training and prediction sets, and the wavenumbers obtained after feature variable selection was the fewest. The best model for the quantitative mission was PLS combined with D1, with R²_C = 1.00 and RMSEP close to 0, for the cross-validation set, with R²_P and RMSEP of 0.91 and 0.14, respectively, for the prediction set, which was the best choice for the determination of adulteration content in G. lingzhi powder. It is demonstrated that NIRS and chemometrics can be used to evaluate adulterated G. lingzhi powder. The model provides a fast, simple and accurate method for the detection of adulterated G. applanatum powder in G. lingzhi powder. In the traditional detection of G. lingzhi powder, the identification of adulterated powder with similar appearance is a complicated process, and there are corresponding requirements for the detection ability and analytical ability of the inspectors. Modern near infrared spectroscopy technology can solve the problem of difficult to identify whether G. lingzhi powder is adulterated or not. Detectors can use the constructed spectral model and then analyze the infrared spectrum of G. lingzhi samples to get its spectral information. By comparing and analyzing the two, it is possible to accurately determine whether the sample is adulterated with forgeries, and also to obtain information on the content of the forgeries. It can effectively improve the market supervision of G. lingzhi, accurately identify the adulterated G. lingzhi powder through the constructed model, carry out strict management, and crack down on adulteration. At the same time, it can safeguard the rights and interests of consumers, ensure the quality of products, prevent other people from buying adulterated G. lingzhi powder and mistakenly believing that the treatment with traditional Chinese medicine is ineffective, and maintain the reputation of traditional Chinese medicine.

Competing Interests:

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding

This work was supported by the Key Research and Development Projects of “Vanguard” and “Leading Goose” in Zhejiang Province (2024C04001), and the Student Research Training Project of Zhejiang Agricultural and Forestry University (2024kx0166).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Author Contributions

Ailian Zhang and Qingsong Shao designed the research. Lingjiao Zhong performed the experiments. Lingjiao Zhong, Yanhong Zhang, Hui Su, Chenye Wang, and Pan Wang analyzed the data. Lingjiao Zhong wrote the manuscript. Qingsong Shao and Ailian Zhang revised the manuscript. All authors read and approved the manuscript.

Araújo, M. C. U., Saldanha, T. C. B., Galvão, R. K. H., Yoneyama, T., Chame, H. C., Visani, V. (2001). The successive projections algorithm for variable selection in spectroscopic multicomponent analysis. Chemometrics and Intelligent Laboratory Systems, 57(2), 65–73. https://doi.org/10.1016/S0169-7439(01)00119-8
Boh, B., Berovic, M., Zhang, J., Lin, Z. (2007). Ganoderma lucidum and its pharmaceutically active compounds. Biotechnology Annual Review, 13, 265–301. https://doi.org/10.1016/S1387-2656(07)13010-6
Cai, Z., Huang, Z., He, M., Li, C., Qi, H., Peng, J., Zhou, F., Zhang, C. (2023). Identification of geographical origins of Radix Paeoniae Alba using hyperspectral imaging with deep learning-based fusion approaches. Food Chemistry, 422, 136169. https://doi.org/10.1016/j.foodchem.2023.136169
Chen, J., Liu, H., Li, J., Wang, Y. (2022). A rapid and effective method for species identification of edible boletes: FT-NIR spectroscopy combined with ResNet. Journal of Food Composition and Analysis, 112, 104698. https://doi.org/10.1016/j.jfca.2022.104698
Daszykowski, M., Stanimirova, I., Walczak, B., Daeyaert, F., de Jonge, M. R., Heeres, J., Koymans, L. M. H., Lewi, P. J., Vinkers, H. M., Janssen, P. A., Massart, D. L. (2005). Improving QSAR models for the biological activity of HIV Reverse Transcriptase inhibitors: Aspects of outlier detection and uninformative variable elimination. Talanta, 68(1), 54–60. https://doi.org/10.1016/j.talanta.2005.04.071
Elreedy, D., Atiya, A. F. (2019). A Comprehensive analysis of synthetic minority oversampling technique (SMOTE) for handling class imbalance. Information Sciences, 505, 32–64. https://doi.org/10.1016/j.ins.2019.07.070
Gao, S., Xu, J. (2022). Hyperspectral image information fusion-based detection of soluble solids content in red globe grapes. Computers and Electronics in Agriculture, 196, 106822. https://doi.org/10.1016/j.compag.2022.106822
Genis, H. E., Durna, S., Boyaci, I. H. (2021). Determination of green pea and spinach adulteration in pistachio nuts using NIR spectroscopy. LWT, 136, 110008. https://doi.org/10.1016/j.lwt.2020.110008
He, M., Chen, X., Zhang, J., Li, J., Zhao, D., Huang, Y., Huo, D., Luo, X., Hou, C. (2023). Identification of liquors from the same brand based on ultraviolet, near-infrared and fluorescence spectroscopy combined with chemometrics. Food Chemistry, 400, 134064. https://doi.org/10.1016/j.foodchem.2022.134064
He, X., Chen, Y., Li, Z., Fang, L., Chen, H., Liang, Z., Abozeid, A., Yang, Z., Yang, D. (2023). Germplasm resources and secondary metabolism regulation in Reishi mushroom (Ganoderma lucidum). Chinese Herbal Medicines, 15(3), 376–382. https://doi.org/10.1016/j.chmed.2023.01.005
Huang, L., Zhou, Y., Meng, L., Wu, D., He, Y. (2017). Comparison of different CCD detectors and chemometrics for predicting total anthocyanin content and antioxidant activity of mulberry fruit using visible and near infrared hyperspectral imaging technique. Food Chemistry, 224, 1–10. https://doi.org/10.1016/j.foodchem.2016.12.037
Jiang, X., Tian, J., Huang, H., Hu, X., Han, L., Huang, D., Luo, H. (2022). Nondestructive visualization and quantification of total acid and reducing sugar contents in fermented grains by combining spectral and color data through hyperspectral imaging. Food Chemistry, 386, 132779. https://doi.org/10.1016/j.foodchem.2022.132779
Jiang, Z., Jin, K., Zhong, L., Zheng, Y., Shao, Q., Zhang, A. (2023). Near-infrared spectroscopy combined with machine learning for rapid identification of Atractylodis rhizoma decoction pieces. Industrial Crops and Products, 197, 116579. https://doi.org/10.1016/j.indcrop.2023.116579
Jouan-Rimbaud, D., Massart, D. L., De Noord, O. E. (1996). Random correlation in variable selection for multivariate calibration with a genetic algorithm. Chemometrics and Intelligent Laboratory Systems, 35(2), 213–220. https://doi.org/10.1016/S0169-7439(96)00062-7
Karabadji, N. E. I., Amara Korba, A., Assi, A., Seridi, H., Aridhi, S., Dhifli, W. (2023). Accuracy and diversity-aware multi-objective approach for random forest construction. Expert Systems with Applications, 225, 120138. https://doi.org/10.1016/j.eswa.2023.120138
Kozarski, M., Klaus, A., Nikšić, M., Vrvić, M., Todorović, N., Jakovljević, D., Van Griensven, L. (2012). Antioxidative activities and chemical characterization of polysaccharide extracts from the widely used mushrooms Ganoderma applanatum, Ganoderma lucidum, Lentinus edodes and Trametes versicolor. Journal of Food Composition and Analysis, 26, 144–153. https://doi.org/10.1016/j.jfca.2012.02.004
Lai, L. (2018). Study on the identification and content determination of Ganoderma lucidum. Journal of Bengbu Medical College, 02, 63. https://kns-cnki-net-443.webvpn.zafu.edu.cn/KCMS/detail/detail.aspx?dbname=CMFD201901&filename=1019011071.nh
Li, G., Mu, L., Zhou, M., Zhao, J., Wu, S., Lin, L. (2021). New strategy of sample set division in spectroscopy analysis——SWNW. Infrared Physics & Technology, 117, 103824. https://doi.org/10.1016/j.infrared.2021.103824
Li, H., Liang, Y., Xu, Q., Cao, D. (2009). Key wavelengths screening using competitive adaptive reweighted sampling method for multivariate calibration. Analytica Chimica Acta, 648(1), 77–84. https://doi.org/10.1016/j.aca.2009.06.046
Li, S., Wang, Z., Shao, Q., Fang, H., Zhu, J., Wu, X., Zheng, B. (2018). Rapid detection of adulteration in Anoectochilus roxburghii by near-infrared spectroscopy coupled with chemometric methods. Journal of Food Science and Technology, 2018, 55, 3518-3525. https://doi.org/10.1007/s13197-018-3276-x
Li, S., Xing, B., Lin, D., Yi, H., Shao, Q. (2020). Rapid detection of saffron(Crocus sativus L) adulterated with lotus stamens and corn stigmas by near-infrared spectroscopy and chemometrics. Industrial Crops and Products, 152, 112539. https://doi.org/10.1016/j.indcrop.2020.112539
Luo, Q., Cao, W., Cheng, Y. (2022). Alkaloids, sesquiterpenoids and hybrids of terpenoid with p-hydroxycinnamic acid from Ganoderma sinensis and their biological evaluation. Phytochemistry, 203, 113379. https://doi.org/10.1016/j.phytochem.2022.113379
Maraphum, K., Saengprachatanarug, K., Wongpichet, S., Phuphuphud, A., Posom, J. (2022). Achieving robustness across different ages and cultivars for an NIRS-PLSR model of fresh cassava root starch and dry matter content. Computers and Electronics in Agriculture, 196, 106872. https://doi.org/10.1016/j.compag.2022.106872
Mata, M., Rocha, P., Farias, I., Silva, J., Medeiros, E., Silva, C., Simões, S. (2022). Distinguishing cotton seed genotypes by means of vibrational spectroscopic methods (NIR and Raman) and chemometrics. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 266, 120399. https://doi.org/10.1016/j.saa.2021.120399
Pandiselvam, R., Mahanti, N., Manikantan, M., Kothakota, A., Chakraborty, S., Ramesh, S., Beegum, P. (2022). Rapid detection of adulteration in desiccated coconut powder: vis-NIR spectroscopy and chemometric approach. Food Control, 133. https://doi.org/10.1016/J.FOODCONT.2021.108588
Pita-Calvo, C., Guerra-Rodríguez, M. E., Vázquez, M. (2017). Analytical methods used in the quality control of honey. Journal of Agricultural and Food Chemistry, 65(4), 690–703. https://doi.org/10.1021/acs.jafc.6b04776
Quan, N. M., Phung, H. M., Uyen, L., Dat, L. Q., Ngoc, L. G., Hoang, N. M., Tu, T. K. M., Dung, N. H., Ai, C. T. D., Trinh, D. N. T. (2023). Species and geographical origin authenticity of green coffee beans using UV–VIS spectroscopy and PLS–DA prediction model. Food Chemistry Advances, 100281. https://doi.org/10.1016/j.focha.2023.100281
Rossi, G. B., Lozano, V. A. (2020). Simultaneous determination of quality parameters inyerba mate(Ilex paraguariensis) samples by application of near-infrared (NIR) spectroscopy and partial least squares (PLS). LWT, 126, 109290. https://doi.org/10.1016/j.lwt.2020.109290
Shao, J., Wang, L., Liu, X., Yang, M., Chen, H., Wu, B., Liu, C. (2019). Identification and characterization of circular RNAs in Ganoderma lucidum.Scientific Reports, 9, 3852-3856. https://doi.org/10.1038/s41598-019-52932-w
Shi, X., Gan, X., Wang, X., Peng, J., Li, Z., Wu, X., Shao, Q., Zhang, A. (2022). Rapid detection of Ganoderma lucidum spore powder adulterated with dyed starch by NIR spectroscopy and chemometrics. LWT, 167, 113829. https://doi.org/10.1016/j.lwt.2022.113829
Shao, Y., Shi, Y., Qin, Y., Xuan, G., Li, J., Li, Q., Yang, F., Hu, Z. (2022). A new quantitative index for the assessment of tomato quality using Vis-NIR hyperspectral imaging. Food Chemistry, 386, 132864. https://doi.org/10.1016/j.foodchem.2022.132864
Tahir, H., Mariod, A., Hashim, S.B.H., Arslan, M., Mahunu, G., Huang, X., Li, Z., Abdalla, I., Zou, X. (2023). Classification of Black Mahlab seeds (Monechma ciliatum) using GC–MS and FT-NIR and simultaneous prediction of their major volatile compounds using chemometrics. Food Chemistry, 408, 134948. https://doi.org/10.1016/j.foodchem.2022.134948
Wang, Z., Wu, Q., Kamruzzaman, M. (2022). Portable NIR spectroscopy and PLS based variable selection for adulteration detection in quinoa flour. Food Control, 138, 108970. https://doi.org/10.1016/j.foodcont.2022.108970
Wu, Q., Li, Y., Peng, K., Wang, X., Ding, Z., Liu, L., Xu, P., Liu, G. (2019). Isolation and characterization of three antihypertension peptides from the mycelia of Ganoderma lucidum (Agaricomycetes). Journal of Agricultural and Food Chemistry, 67, 8149-8159. https://doi.org/10.1021/acs.jafc.9b02276
Xu, Y., Liu, J., Sun, Y., Chen, S., Miao, X. (2023). Fast detection of volatile fatty acids in biogas slurry using NIR spectroscopy combined with feature wavelength selection. Science of the Total Environment, 857, 159282. https://doi.org/10.1016/j.scitotenv.2022.159282
Yao, R., He, C., Xiao, P. (2023). ‘Food and medicine continuum’ in the East and West: Old tradition and current regulation. Chinese Herbal Medicines, 15(1), 6–14. https://doi.org/10.1016/j.chmed.2022.12.002
Ye, T., Zheng, Y., Guan, Y., Sun, Y., Chen, C. (2023). Rapid determination of chemical components and antioxidant activity of the fruit of Crataegus pinnatifida Bunge by NIRS and chemometrics. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 289, 122215. https://doi.org/10.1016/j.saa.2022.122215
Yuan, R., Liu, G., He, J., Wan, G., Fan, N., Li, Y., Sun, Y. (2021). Classification of Lingwu long jujube internal bruise over time based on visible near-infrared hyperspectral imaging combined with partial least squares-discriminant analysis. Computers and Electronics in Agriculture, 182, 106043. https://doi.org/10.1016/j.compag.2021.106043
Yu, D., Guo, S., Zhang, X., Yan, H., Zhang, Z., Chen, X., Chen, J., Jin, S., Yang, J., Duan, J. (2022). Rapid detection of adulteration in powder of ginger (Zingiber officinale Roscoe)by FT-NIR spectroscopy combined with chemometrics. Food Chemistry: X, 15, 100450. https://doi.org/10.1016/j.fochx.2022.100450
Zhang, X., Sun, J., Li, P., Zeng, F., Wang, H. (2021). Hyperspectral detection of salted sea cucumber adulteration using different spectral preprocessing techniques and SVM method. LWT, 152, 112295. https://doi.org/10.1016/j.lwt.2021.112295

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Based on near-infrared spectroscopy and chemometrics to rapidly evaluate the adulteration of Ganoderma lingzhi powder

Status:

Version 1

Abstract

Figures

1. Introduction

2. Material and methods

2.1. Sample Preparation

2.2. Data collection

2.3. Chemometrics analysis

2.3.1. Spectra preprocessing

2.3.2. Screening of characteristic variables

2.3.3. Qualitative analysis

2.3.4. Quantitative analysis

2.4. Software and algorithms

3. Results and discussion

3.1. Sample and spectral analysis

3.2. Qualitative analysis

3.3. Quantitative analysis

3.3.1. Quantitative analysis with PLS

4. Conclusion

Declarations

References

Additional Declarations

Status:

Version 1