Construction and application of Raman spectrum model of Dayuan drink preparation

Abstract

The invention belongs to the technical field of quality detection and analysis of traditional Chinese medicines, and particularly relates to construction of a Raman spectrum model of a Dayuan decoction preparation and application of the Raman spectrum model in whole-process quality control of the Dayuan decoction preparation. The method comprises the following steps: collecting Raman spectrum original maps of sample types under different detection conditions and different technological processes; carrying out pretreatment on the original atlas; the PLS-DA is used to classify data of different types of samples, and a relation model between omics data and the sample types is established to realize identification of the sample types. According to the method, a Raman spectrum technology is combined with chemometrics, and an analysis method suitable for rapid and accurate identification and classification of the Dayuan granules is established for samples of different process steps of the whole process flow of the Dayuan granules; when part of medicinal materials are lost in the extraction step, the model distinguishes a negative sample, identification of the whole preparation process can be achieved in practice, and compared with traditional thin-layer chromatography, the method is quicker, simpler, more convenient and more accurate.

Description

Technical Field

[0001] The present invention belongs to the technical field of quality detection and analysis of traditional Chinese medicines, and specifically relates to the construction of a Raman spectroscopy model of a Dayuanyin preparation based on Raman spectroscopy technology and chemometrics and its application in the whole process quality control of the Dayuanyin preparation. Background Art

[0002] The quality control method of Dayuanyin is generally to select index components or effective ingredients reported in the literature for quality evaluation, such as using HPLC or MS methods to qualitatively or quantitatively analyze the components of a certain medicinal material or multiple Chinese medicinal materials. All of them use some substances in the prescription as indicators to carry out quality evaluation. However, the composition of Chinese medicine is complex, and using some components as indicators to evaluate quality has certain limitations.

[0003] The Chinese patent medicine Dayuanyin granules prepare the Dayuanyin prescription into granules for easy consumption. The preparation process of granules involves extraction, concentration, granulation and other process flows. According to the pharmacopoeia, the current quality control only conducts qualitative identification of the granules after preparation, and uses thin layer chromatography to qualitatively identify the characteristic components in the prescription. The thin layer chromatography method requires 1-2 days for sample pretreatment and inspection. When fake and inferior drugs with the same or similar active ingredients are used as substitutes, it is impossible to distinguish and identify them. In addition, there is currently no inspection method for quality control of the preparation process of Dayuanyin granules, whether it is thin layer chromatography or HPLC. Because the entire process of inspecting a single sample is time-consuming, it affects the production progress and cannot achieve dynamic quality control of the entire process.

[0004] In recent years, spectral technology has been widely used in the quality control of traditional Chinese medicine. Raman spectroscopy has the advantages of being non-destructive, rapid, and not subject to signal interference from water. Zhou Qun et al. [Raman spectroscopy for rapid identification of Scutellaria baicalensis in Chinese medicine [J]. Journal of Light Scattering, 2002, (03): 166-168.] used Raman spectroscopy to distinguish and identify Scutellaria baicalensis; Yin Wenjun et al. [CN112964690B] used Raman spectroscopy to monitor the extraction process of liquorice formula granules in real time. The existing model distinguishes and identifies a single medicinal material, or performs qualitative or quantitative analysis on a small number of clear index components in Chinese medicine formula granules. Data preprocessing or modeling for the index components will lose other important spectral information that is not related to the index components. Therefore, the existing distinction matching model cannot be applied to the quality control of prescriptions composed of multiple medicinal materials and their Chinese patent medicine preparations. Compared with single medicinal materials or single-flavor formula granules, compound formula granules have complex ingredients and more interference information. It is often necessary to increase the index components to improve the data quality in order to ensure the distinction effect of the model. When the indicator components are unclear or numerous, the existing model needs to undergo multi-step preprocessing to filter out effective information. The model calculation is complex and the accuracy may not be high. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention adopts Raman spectroscopy combined with chemometrics to establish a rapid, efficient, dynamic whole-process quality control and real-time monitoring method for Dayuanyin preparations:

[0006] A method for constructing a Raman spectroscopy model for Chinese patent medicine preparations:

[0007] (1) Collecting Raman spectra: Collecting original Raman spectra of samples under different detection conditions and different process flows;

[0008] (2) Spectral preprocessing: preprocessing the original spectrum;

[0009] (3) Establishing a spectral classification model: Using PLS-DA to classify data of different categories, and by establishing a relationship model between omics data and sample categories, the sample categories can be identified; the PLS-DA is partial least squares discriminant analysis.

[0010] By adjusting the scanning time and laser intensity of the Raman spectrometer, each category of samples is tested under multiple detection conditions, and multiple Raman spectra under different detection conditions are collected as original spectra.

[0011] The preprocessing method is one or more of normalization, Savizkg-Golag smoothing, first-order differential, and standard normal variable.

[0012] The first-order differential uses the Savitsky-Golay convolution first-order derivative to calculate the first-order derivative (i.e., first-order differential) of the data. The formula for the Savitsky-Golay convolution first-order derivative is as follows:

[0013] dx[n]=y[n]-x[n]

[0014] x[n] is the sample data sequence, dx[n] is the derivative result, where n is the index. The convolution kernel matrix H is constructed as follows:

[0015]

[0016] Each element of the matrix is ​​the coefficient of the polynomial; applying the convolution matrix H to the data sequence x[n] yields a new sequence y[n]:

[0017] y[n]=H·x[n]

[0018] The normalized formula is as follows:

[0019]

[0020] Where X′ is the normalized data, X is the original data, min and max are the minimum and maximum values ​​in the original data respectively.

[0021] Preferably, in combination with the PLS-DA classification model, the preprocessing method is selected as the first-order differential according to the classification results from three aspects: accuracy, correlation coefficient, and determination coefficient.

[0022] Further,

[0023] Partial least squares discriminant analysis (PLS-DA) uses the partial least squares (PLS) algorithm to perform orthogonal decomposition on the sample spectrum matrix X and the category matrix Y respectively, and decomposes the spectrum matrix X to obtain the score matrix T:

[0024] X=TP T +E

[0025] The category matrix Y is also orthogonally decomposed to obtain the score matrix U:

[0026] Y=UQ U +F

[0027] Where T and U are the score matrices of X and Y, respectively, P and Q are the loading matrices of X and Y, P T and Q T represents the conversion matrix of P and Q, E and F are error matrices;

[0028] When predicting, first calculate the unknown sample spectrum matrix X according to the load matrix P text The score is then calculated to get the regression category information matrix Y text , according to Y text The degree of proximity to different categories determines which category the sample belongs to, thereby enabling the discrimination of samples at different stages of the preparation process.

[0029] The prediction accuracy of the model is evaluated through cross-validation; in the cross-validation process, the data set is divided into multiple parts, one part is extracted as the test set, and the rest is used to train the model. By repeating this process many times, a comprehensive Q 2 Value, Q 2 The higher the value, the stronger the model's predictive ability;

[0030] The multiple determination coefficient (r 2 ) measures the model's fitting ability, r 2 The higher the value, the better the model fit;

[0031] Q 2 Calculation formula:

[0032]

[0033] r2 Calculation formula:

[0034]

[0035] Among them, y i is the measured value of the ith sample, is the predicted value of the ith observation using the model constructed using all other observations except the ith observation, is the predicted value of the model for the i-th sample, y is the average of all measured values, and n is the total number of samples.

[0036] The Chinese patent medicine preparation is Dayuanyin granules.

[0037] An application of a Raman spectroscopy model constructed according to the above method in the whole process quality control of Dayuanyin preparation.

[0038] The Dayuanyin preparation is Dayuanyin granules.

[0039] Beneficial effects of the detection method of the present invention:

[0040] (1) Raman spectroscopy combined with chemometrics was used to establish an analytical method suitable for rapid and accurate identification and classification of Dayuanyin granules based on samples from different process steps of the entire production process.

[0041] (2) When some medicinal materials are missing during the extraction step, the model can distinguish negative samples and can realize the identification of the entire preparation process in actual quality control. Compared with traditional thin-layer chromatography, it is faster, simpler, and more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The Raman spectra are as follows: the left sub-figure is the Raman spectra of the two mixed extracts of the normal prescription, and the right sub-figure is the Raman spectra of the two mixed extracts of the negative prescription;

[0043] Figure 2 This is the result diagram of the data preprocessing method;

[0044] Figure 3 This is the result diagram of first-order differential-normalization-PLS-DA distinguishing negative samples (AD is the first extract, second extract, mixed extract, and granules of the normal process; G is the negative mixed extract). DETAILED DESCRIPTION

[0045] 1. Materials and Instruments

[0046] The seven traditional Chinese medicine pieces of Dayuanyin: Areca nut, Magnolia officinalis, Amomum villosum, Rhizoma Anemarrhenae, Radix Paeoniae Root, Radix Scutellariae, and Radix Glycyrrhizae. Portable Raman spectrometer (BWS015 American Bitatec Corporation) equipped with a 785nm laser light source (350mw).

[0047] 2. Experimental Methods

[0048] 1. Preparation of test samples

[0049] The process flow of Dayuanyin granules is divided into extraction, concentration, granulation and other processes. The established quality control method needs to be able to quickly monitor the sample categories in each link and determine whether it is inconsistent with the prescription.

[0050] Positive test solution: Take 18g of Areca nut, 9.5g of Magnolia officinalis, 4g of Amomum villosum, 9.4g of Rhizoma Anemarrhenae, 9.2g of Radix Paeoniae Alba, 9.3g of Radix Scutellariae, and 4.2g of Radix Glycyrrhizae, add 380ml of water, soak for 30 minutes, boil and simmer for 2 hours, filter as the first extract (category A). Add 285ml of water to the residue, boil and simmer for 1 hour, filter, and get the second extract (category B). Combine the two filtrates to get the final mixed extract (category C).

[0051] Finished product solution sample: Take multiple batches of Dayuanyin granules, take 10g of granules and dissolve them in 80 degrees Celsius hot water, let it stand and cool, and use it as the test solution for the finished product (Category D).

[0052] Negative test solution: Take 4g of Amomum villosum, 9.4g of Rhizoma Anemarrhenae, 9.2g of Radix Paeoniae Alba, 9.3g of Radix Scutellariae, and 4.2g of Radix Glycyrrhizae, add 380ml of water, soak for 30 minutes, boil and simmer for 2 hours, filter as the first extract. Add 285ml of water to the residue, boil and simmer for 1 hour, filter as the second extract. Combine the two filtrates to obtain the final mixed extract (category G).

[0053] 2. Spectral processing and model building

[0054] (1) Raman spectrum acquisition:

[0055] Raman spectroscopy detection conditions: By adjusting the scanning time and laser intensity of the Raman spectrometer, each type of sample is tested under multiple detection conditions, and multiple Raman spectra under different detection conditions are collected as the original spectrum. Raman shift: 0-3378cm -1 , data preprocessing and graphics: matalab (2020 version) and metaboanalys.

[0056] After preparing various test solutions, transfer the solutions into cuvettes in turn and place them in the detection pool. Using the manual acquisition mode, 120 spectra of samples of categories A, B, C, D, and G under different detection conditions are collected as original spectra. Some spectra are shown in ( Figure 1 ). Before changing to another type of solution, add purified water to the cuvette to test whether there is any residue in the cuvette. After ensuring that there is no residue, start the next type of solution test.

[0057] (2) Spectral preprocessing method

[0058] Because Raman spectroscopy is easily interfered by fluorescence and instrument noise, the interference signals in the Raman spectrum need to be processed. In this study, normalization, Savizkg-Golag smoothing, first-order differential, standard normal variable (SNV) and other methods were used to process the original spectrum. Combined with the PLS-DA classification model, the data preprocessing method was selected according to the classification results (see Figure 2 and Table 1). In the PLS-DA model, accuracy refers to the proportion of samples that are correctly classified by the classification model, that is, the number of correct classifications divided by the total number of samples. 2 ) represents the predictive ability of the model, that is, the accuracy of the model in predicting new data (data not involved in modeling). 2 The higher the value, the stronger the predictive ability of the model. 2 ) represents the explanatory rate of the model, that is, the proportion of the original data variation that the model can explain. 2 The closer the value is to 1, the stronger the explanatory power of the model. In PLS-DA, this indicator can reflect the overall effect of the model on sample classification and select the optimal data preprocessing method.

[0059] Among them, the first-order differential uses the Savitsky-Golay convolution first-order derivative to calculate the first-order derivative of the data. The formula for the Savitsky-Golay convolution first-order derivative is as follows:

[0060] dx[n]=y[n]-x[n]

[0061] x[n] is the sample data sequence, dx[n] is the derivative result, where n is the index. The convolution kernel matrix H is constructed as follows:

[0062]

[0063] Each element of the matrix is ​​the coefficient of the polynomial; applying the convolution matrix H to the data sequence x[n] yields a new sequence y[n]:

[0064] y[n]=H·x[n]

[0065] Parameter settings for the first-order differential in the Matlab software are: degree of the derivative: 1, window (the number of points in filter): 5;, order (the order of the polynomial): 2.

[0066] The normalized formula is as follows:

[0067]

[0068] Where X′ is the normalized data, X is the original data, min and max are the minimum and maximum values ​​in the original data respectively.

[0069] (3) Establishment of spectral classification model

[0070] The collected spectral data were divided into training set and test set according to 7:3, and partial least squares discriminant analysis (PLS-DA) was used for classification analysis. Partial least squares discriminant analysis (PLS-DA) predicts sample categories by establishing a relationship model between omics data and sample categories, which is a supervised modeling method. According to the PLS-DA results, the original spectral data was preprocessed first, and then PLS-DA was used to classify different categories of data as a quality control method for the preparation process of Dayuanyin granules.

[0071] Among them, PLS-DA uses the partial least squares (PLS) algorithm to decompose the sample spectrum matrix X and the category matrix Y respectively, and decomposes the spectrum matrix X to obtain the score matrix T:

[0072] X=TP T +E

[0073] The category matrix Y is also orthogonally decomposed to obtain the score matrix U:

[0074] Y=UQ U +F

[0075] Where T and U are the score matrices of X and Y, respectively, P and Q are the loading matrices of X and Y, P T and Q U The conversion matrix of P and Q is represented, and E and F are error matrices, so as to find the variables and directions of the multivariate space and identify the category of the correction set. When predicting, first calculate the unknown sample spectrum matrix X according to the load matrix P. text The score is then calculated to get the regression category information matrix Y text, according to Y text The degree of proximity to different categories determines which category the sample belongs to, thereby achieving the discrimination of samples at different stages of the seed. PLS-DA extracts the eigenvectors of the spectral matrix and the category matrix at the same time, strengthening the role of the sample category information during spectral decomposition, and can extract the spectral information most relevant to the sample extraction component, thereby maximizing the difference between the extracted spectra of different categories.

[0076] The prediction accuracy of the model was evaluated by cross validation. 2 The higher the value, the stronger the model's predictive power. During cross-validation, the dataset is divided into multiple parts, one part is taken out as the test set, and the rest is used to train the model. By repeating this process multiple times, a comprehensive Q 2 value.

[0077] The multiple determination coefficient (r 2 ) measures the model's fit. 2 The higher the value, the better the model fit.

[0078] Q 2 Calculation formula:

[0079]

[0080] r 2 Calculation formula:

[0081]

[0082] Among them, y i is the measured value of the ith sample, is the predicted value of the ith observation using the model constructed using all other observations except the ith observation, is the predicted value of the model for the i-th sample, y is the average of all measured values, and n is the total number of samples.

[0083] 3. Results

[0084] Depend on Figure 1 It can be seen that due to the complex composition and weak spectral characteristics of various samples, it is impossible to distinguish them by simple comparison. PLS-DA is used to classify and predict the original data and the data processed by different data preprocessing methods. The results are shown in Table 1. Table 1 shows the evaluation results of various data preprocessing methods combined with PLS-DA classification. According to the accuracy (Accuracy), correlation coefficient (R 2 ), determination coefficient (Q 2 ) results, the first-order differential was selected as the final data preprocessing method.

[0085] Figure 3As an example verification of negative sample classification prediction, ABCDG class samples are preprocessed by first-order differential and then discriminated and predicted by PLS-DA classification model. The model can well distinguish between G class and other class samples.

[0086] Table 1 PLS-DA classification results of different data preprocessing methods

[0087]

[0088] When the prescription ingredients change, the PLS-DA model combined with the first-order differential data preprocessing method can well identify and predict the Raman spectra of samples of questionable quality, and its prediction accuracy is improved from 0.77173 to 0.85645. Therefore, in each process link of the preparation process of Dayuanyin granules, there is no need to pre-treat the samples, and the Raman spectral data of the samples can be directly collected. Using the model established above, the samples in the preparation process of Dayuanyin granules can be quickly qualitatively identified. The present invention is the first to establish a method for identifying Dayuanyin granules based on Raman spectroscopy technology and chemometrics. It has been verified that this method is simple, efficient, and has good predictive effect.

Claims

1. A method for constructing a Raman spectrum model of a Chinese patent medicine preparation, characterized in that: (1) Collecting Raman spectra: Collecting original Raman spectra of sample types at different stages of the preparation process of Chinese patent medicines under different testing conditions; (2) Spectral preprocessing: preprocessing the original spectrum; (3) Establish a spectral classification model: Use PLS-DA to classify data of different sample categories, and identify sample categories by establishing a relationship model between omics data and sample categories.

2. The method according to claim 1, characterized in that: By adjusting the scanning time and laser intensity of the Raman spectrometer, each category of samples is tested under multiple detection conditions, and multiple Raman spectra under different detection conditions are collected as the original Raman spectra.

3. The method according to claim 1, characterized in that: The preprocessing method is one or more of normalization, Savizkg-Golag smoothing, first-order differential, and standard normal variable.

4. The method according to claim 3, characterized in that: The first-order derivative uses the Savitsky-Golay convolution first-order derivative to calculate the first-order derivative of the data. point; The formula for the first-order derivative of Savitsky-Golay convolution is as follows: dx[n]=y[n]-x[n], x[n] is the sample data sequence, dx[n] is the derivative result, where n is the index; the convolution kernel matrix H is constructed as follows: Each element of the matrix is ​​the coefficient of the polynomial; applying the convolution matrix H to the data sequence x[n] yields a new sequence y[n]: y[n]=H·x[n] Parameter settings in the first-order differential matalab software: degree of the derivative: 1; window: 5; order: 2; The normalized formula is: Where X′ is the normalized data, X is the original data, min and max are the minimum and maximum values ​​in the original data respectively.

5. The method according to claim 3, characterized in that: Combined with the PLS-DA classification model, the preprocessing method was selected as the first-order differential according to the classification results from three aspects: accuracy, correlation coefficient, and determination coefficient.

6. The method according to claim 1, characterized in that: PLS-DA uses the PLS algorithm to perform orthogonal decomposition on the sample spectral matrix X and the category matrix Y respectively, and decomposes the spectral matrix X to obtain the score matrix T: X=TP T +E The category matrix Y is also orthogonally decomposed to obtain the score matrix U: Y=UQ U +F Where T and U are the score matrices of X and Y, respectively, P and Q are the loading matrices of X and Y, P T and Q T represents the conversion matrix of P and Q, E and F are error matrices; When predicting, first calculate the unknown sample spectrum matrix X according to the load matrix P text The score is then calculated to get the regression category information matrix Y text , according to Y text The degree of proximity to multiple categories determines which category the sample belongs to, thereby enabling the discrimination of sample categories at different stages of the preparation process.

7. The method according to claim 6, characterized in that: The prediction accuracy of the model is evaluated through cross-validation. In the cross-validation process, a part of the data set is extracted as a test set, and the rest is used to train the model. By repeating this process many times, a comprehensive Q 2 Value, Q 2 The higher the value, the stronger the model's predictive ability; Using the multiple determination coefficient r 2 Measures the model's fit, r 2 The higher the value, the better the model fit; Q 2 Calculation formula: r 2 Calculation formula: Among them, y i is the measured value of the ith sample, is the predicted value of the ith observation using the model constructed using all other observations except the ith observation, is the model's predicted value for the i-th sample, is the average of all measured values, and n is the total number of samples.

8. The method according to claim 1, characterized in that: The Chinese patent medicine preparation is Dayuanyin granules.

9. Application of a Raman spectroscopy model constructed according to any one of claims 1 to 8 in the whole process quality control of Dayuanyin preparation.

10. The use according to claim 9, characterized in that: The Dayuanyin preparation is Dayuanyin granules.

Images