CN114577967B - Chromatographic analysis method of traditional Chinese medicine compound samples based on artificial neural network and differential spectrum - Google Patents

Abstract

The invention relates to a chromatographic fingerprint spectrum analysis method of a traditional Chinese medicine compound. The invention aims to provide a traditional Chinese medicine compound fingerprint analysis method based on a linear artificial neural network model and a chromatographic difference spectrum algorithm, aiming at the limitations of the existing traditional Chinese medicine chromatographic fingerprint processing method. The single medicine composition of the traditional Chinese medicine compound is analyzed by a method of performing difference spectrum analysis finally through a chromatographic peak matching, chromatogram correction and chromatogram reconstruction flow of a linear artificial neural network model, and the method has a good visualization effect. Lays an important foundation for the improvement of the quality control technology of the traditional Chinese medicine compound fingerprint.

Description

Chromatographic analysis method of traditional Chinese medicine compound samples based on artificial neural network and difference spectrum

Technical Field

The invention belongs to the field of Chinese medicine compound spectrum analysis methods, and specifically is a Chinese medicine compound chromatographic fingerprint analysis method based on an artificial neural network model LSM algorithm and a difference spectrum algorithm.

Background Art

With the continuous development of the Chinese medicine industry and Chinese medicine culture, the production scale of Chinese medicine has been expanding in recent years, and the number of Chinese medicine manufacturers has been increasing. The problem of compound medicinal materials counterfeiting has become more and more obvious, such as stealing and reducing medicinal materials, replacing decoction pieces, and expiration of medicinal materials. These have caused significant losses to the safety and reputation of Chinese medicine, and seriously endangered the sustainable development of the Chinese medicine industry. It is of great significance to effectively analyze various compound medicinal materials of Chinese medicine and effectively establish the quality standards of Chinese medicine. If a systematic, comprehensive, and visual Chinese medicine chromatographic analysis method can be established, it is possible to effectively analyze the components of Chinese medicine, effectively supervise the quality of Chinese medicine compound products, and ensure the healthy development of the market.

At present, the analysis of fingerprint spectra of traditional Chinese medicine compounds is still in its early stages in China, such as manual peak matching and characteristic peak attribution based on retention time. This method is time-consuming, has low accuracy, and has a low degree of visualization when processing compound samples on a large scale, which limits its application.

Summary of the invention

The purpose of the present invention is to provide a method for analyzing Chinese medicine compound prescriptions based on a linear artificial neural network model and a chromatographic difference spectrum algorithm in view of the limitations of existing Chinese medicine chromatographic fingerprint processing methods. The method finally obtains a chromatographic difference spectrum and performs analysis through chromatographic peak matching, chromatogram correction, and a chromatographic process reconstructed by a linear artificial neural network model, so as to accurately and automatically analyze the composition of single herbs in the compound prescription and present it intuitively.

The technical solution adopted by the present invention to achieve the above-mentioned purpose is: a chromatographic analysis method of a Chinese medicine compound sample based on an artificial neural network and a difference spectrum, comprising the following steps:

For different compound or single Chinese medicine samples, each sample is set to a fixed wavelength through a diode array detector to collect ultraviolet spectrum data, the chromatograms of different samples are extracted and integrated, and the peaks of the chromatogram integration results of any two samples are matched by spectral matching;

In the same time window T through spectrum matching, the peak with the largest similarity value and greater than the threshold is classified as a unified peak F _i as the chromatographic peak matching result, where F _i is the peak class number;

For each chromatographic peak matching result, the chromatographic peak apex is used as a reference, the chromatogram is resampled using the Fourier transform method, and the chromatographic time is corrected according to the time interval to obtain the corrected compound Chinese medicine spectrum and single Chinese medicine spectrum;

对于未知的复方中药样品，将校正后的复方中药谱图通过使用未知复方样品输入线性人工神经网络模型得到结果

对于已知的复方或单味中药样品，将校正后的复方中药谱图或单味中药谱图使用已知复方样品输入线性人工神经网络模型得到结果

与单味中药谱图进行运算得到解析谱图；According to the calibration chromatogram, the unknown compound sample input linear artificial neural network model and the known compound sample input linear artificial neural network model are fitted respectively. The outputs of both models are sample coefficient vectors.

For unknown compound Chinese medicine samples, the corrected compound Chinese medicine spectrum is input into the linear artificial neural network model by using the unknown compound sample to obtain the result.

For known compound or single Chinese medicine samples, the corrected compound Chinese medicine spectrum or single Chinese medicine spectrum is input into the linear artificial neural network model using the known compound sample to obtain the result.

Calculate with the spectrum of single Chinese medicine to obtain the analytical spectrum;

The analyzed spectrum is compared with the single-herb spectrum of known samples to parse out the missing medicinal materials in the compound sample.

The method of performing peak matching of chromatographic integration results of any two samples by using spectrum matching comprises the following steps:

The peaks in the time window T are used to extract the vertex spectra and stored as peak-spectrum vectors _Xij , where i is the peak number and j is the sample number of the compound or single-flavor herbal medicine piece to which it belongs, which are irrelevant positive integers.

The cosine similarity calculation is performed on any two of X ₁₁ , X ₂₂ ……X _ij , and the similarity result is

k为峰编号，m为所属复方或单味饮片样品编号，其相似度公式如下：

k is the peak number, m is the sample number of the compound or single herbal medicine, and the similarity formula is as follows:

By selecting the time window T value, the size represents the time range of the peaks involved in the matching, and the value range is between (0 and 10) minutes.

The chromatogram is resampled using the Fourier transform method, and the chromatogram time correction is performed according to the time interval, which is achieved by the following formula:

Where t is time, f(t) is the chromatographic signal function, p is the sampling rate, and k is a positive integer.

The fitting of the unknown compound sample input linear artificial neural network model or the standard compound sample input linear artificial neural network model comprises the following steps:

单味饮片谱图为

j为单味饮片样品编号，将

组合构成单味饮片谱图矩阵D，定义系数向量为

在迭代次数为n时系数向量为

学习率为α，残差为‖Δ‖则有：The calibrated chromatogram is the calibrated compound Chinese medicine spectrum or single Chinese medicine spectrum. The calibrated chromatogram is a vector

The spectrum of single herbal medicine pieces is

j is the sample number of the single-ingredient decoction piece.

The single-ingredient herbal medicine spectrum matrix D is composed, and the coefficient vector is defined as

When the number of iterations is n, the coefficient vector is

If the learning rate is α and the residual is ‖Δ‖, then:

Taking partial derivatives of Θ, we have:

At this time, there is an iterative formula:

Iterate until ‖Δ‖ is less than the set value and stop learning.

模型输出为样品系数向量

The input of the linear artificial neural network model for unknown compound samples and the linear artificial neural network model for standard compound samples is the vector

The model output is a sample coefficient vector

The analytical spectrum is obtained by the following formula:

为标准复方样品输入线性人工神经网络模型输出参数，

为未知复方样品输入线性人工神经网络模型输出参数。Where P is the difference spectrum result, D is the spectrum of a single herbal medicine piece,

Input the output parameters of the linear artificial neural network model for the standard compound sample,

Input the output parameters of the linear artificial neural network model for unknown compound samples.

The ω value indicates the degree of spectral similarity between different peaks of different samples, and its value range is between (0 and 1).

The diode array detector can be replaced by one of a high performance liquid chromatography-DAD detector, a high performance liquid chromatography-mass spectrometry detector, a gas chromatography-DAD detector, a gas chromatography-mass spectrometry detector, a high performance capillary electrophoresis chromatography-DAD detector or a high performance capillary electrophoresis chromatography-mass spectrometry detector.

The unknown compound sample input linear artificial neural network model or the standard compound sample input linear artificial neural network model is constructed, and the residual is minimized by one of LSM, gradient descent and least square method.

The chromatography is high performance liquid chromatography, gas chromatography or high performance capillary electrophoresis chromatography.

The present invention has the following beneficial effects and advantages:

The present invention considers the correlation between compound samples and single-flavor sample pieces, gives full play to the advantages of linear artificial neural network algorithm, and visualizes the complex spectrum analysis results by using the difference spectrum analysis method.

The method of the present invention is simple and easy to implement, and has wide algorithm applicability and high degree of automation, so it is suitable for large-scale detection. The time correction by resampling of Fourier transform improves the consistency of spectrum and reduces the requirements of operating conditions. The linear artificial neural network algorithm is used to establish the model, thereby greatly improving the accuracy of measurement. The difference spectrum method is used to analyze the compound spectrum, thereby improving the visualization of the results.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a flow chart of the method of the present invention;

Fig. 2 is an example of the present invention - a spectrum of a Qinggusan compound recipe sample;

Fig. 3 is an example of the present invention - a calibrated spectrum of Qinggusan;

Fig. 4 is a diagram of the linear neural network structure of the present invention;

FIG5 is an example of the present invention - a chromatogram of Qinggusan reconstructed based on a linear neural network;

FIG6 is an example of the present invention - missing Hu Huang Lian Qing Gu San, Hu Huang Lian and difference spectrum comparison diagram;

FIG. 7 is an example of the present invention - missing Gancao Qinggu San, Picrorhizoma Coptidis and difference spectrum comparison diagram;

FIG8 is an example of the present invention - missing Qinji Qinggu San, Picrorhizoma Coptidis and difference spectrum comparison diagram.

DETAILED DESCRIPTION

The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

The chromatographic analysis method of a Chinese medicine compound sample based on an artificial neural network model and a difference spectrum method is based on liquid chromatography technology, uses a spectrum matching method to perform peak matching, uses a resampling method to perform chromatographic time correction, uses an artificial neural network model to calculate the weight of a single flavor sample and reconstruct the spectrum, and uses a difference spectrum method to analyze the chromatographic fingerprint of the Chinese medicine compound, including the following steps:

(1) Collecting diode array ultraviolet spectrum (DAD) data according to the diode array detector of different samples, selecting a unified wavelength to extract and integrate the chromatogram according to the characteristics of the compound sample, and performing peak matching of the chromatogram integration results by spectrum matching;

其相似度公式如下：(2) The matching method used in step (1) uses the peaks in the time window T to extract the vertex spectra and stores them as peak-spectrum vectors _Xij , where i is the peak number and j is the sample number of the compound or single-flavor herbal medicine piece to which it belongs, which are irrelevant positive integers. The cosine similarity calculation is performed on any two of _Xi11 , _Xi22 ... _Xij , and the result is

The similarity formula is as follows:

(3) The peaks with the largest ω calculated within the same time window T and greater than the threshold are classified as unified peaks F _i to obtain matching results, where F _i is the peak class number.

(4) Based on the chromatographic peak matching result obtained in step (3), the chromatographic peak apex is used as a reference, the chromatogram is resampled using the Fourier transform method, and the chromatographic time correction is performed according to the time interval. The core of the algorithm is:

Where t is time, f(t) is the chromatographic signal function, p is the sampling rate, and k is a positive integer.

(5) According to the calibrated chromatogram obtained in step (4), a least mean square algorithm (LMS) is used to fit the linear artificial neural network model. The LMS learning algorithm and the linear artificial neural network formula are as follows:

单味饮片谱图为

(j为所属单味饮片样品编号)。将

组合构成单味饮片谱图矩阵D，定义系数向量为

在迭代次数为n时系数向量为

学习率为α，残差为‖Δ‖则有：Let the compound spectrum (standard compound sample spectrum or compound sample spectrum to be analyzed) be vector

The spectrum of single herbal medicine pieces is

(j is the sample number of the single herbal medicine piece).

The single-ingredient herbal medicine spectrum matrix D is composed, and the coefficient vector is defined as

When the number of iterations is n, the coefficient vector is

If the learning rate is α and the residual is ‖Δ‖, then:

Taking partial derivatives of Θ, we have:

At this time, there is an iterative formula:

Iterate until ‖Δ‖ is less than a certain value and stop learning, and get

模型输出为样品系数向量

Among them, the input of the specific compound sample model in each analysis is the compound spectrum vector

The model output is a sample coefficient vector

使用标准复方样品输入线性人工神经网络模型得到结果

与单味饮片谱图矩阵计算得到解析谱图，其公式为：(6) Use unknown compound samples to input the linear artificial neural network model to obtain the results

The results were obtained by inputting the standard compound samples into the linear artificial neural network model.

The analytical spectrum is calculated with the spectrum matrix of single-flavor decoction pieces, and the formula is:

为标准谱人工神经网络模型参数，

为未知样品谱人工神经网络模型参数。Where P is the difference spectrum result, D is the spectrum of a single herbal medicine piece,

are the standard spectral artificial neural network model parameters,

are the artificial neural network model parameters of the unknown sample spectrum.

(7) The difference spectrum result P can be compared with the corresponding single-ingredient herbal medicine spectrum for intuitive comparison and analysis. Based on the characteristics of the difference spectrum, the missing herbal medicine in the compound sample can be easily analyzed.

By selecting the time window T value, the size indicates the time range of the peaks involved in the matching, and the value range is between (0 and 10) minutes. The ω value indicates the spectral similarity between different peaks of different samples, and the value range is between (0 and 1). Within the same time window, select the value of ω in the range of (0.2 to ∞), and take the maximum value as the matching peak.

The chromatographic peak similarity calculation is performed through diode array three-dimensional data or three-dimensional mass spectrometry data, including the two-dimensional detection technology, including but not limited to the time-intensity-wavelength/charge-to-mass ratio of a high performance liquid chromatography-DAD detector, a high performance liquid chromatography-mass spectrometry detector, a gas chromatography-DAD detector, a gas chromatography-mass spectrometry detector, a high performance capillary electrophoresis chromatography-DAD detector or a high performance capillary electrophoresis chromatography-mass spectrometry detector.

Based on the matching peak retention times, the data between adjacent peak apexes were resampled to make the number of data points consistent, and the chromatograms were corrected using the same time interval.

The linear model was constructed using compound chromatograms and single herbal medicine chromatograms, and the algorithms included but were not limited to LSM, gradient descent, and least squares method to minimize the residual. The model and target residual ‖Δ‖ was used as the termination condition, and the termination was terminated when it met the value range, and the value range of ‖Δ‖ was (-∞～0.5).

The chromatogram after time correction or the chromatogram reconstructed by the data model is subtracted to obtain the chromatogram difference spectrum.

The material components are analyzed by using the chromatographic difference spectrum of traditional Chinese medicine, and its characteristic is that the chromatographic method includes high performance liquid chromatography, gas chromatography or high performance capillary electrophoresis chromatography.

As shown in Figure 1, the present invention collects DAD data graphs according to the diode array detector of different samples, extracts and integrates the chromatogram according to the wavelength selected by the medicinal material, uses spectrum matching to perform peak matching of the chromatogram integration result, uses the matching result, adopts a resampling method to perform chromatogram peak correction, and then uses a linear artificial neural network model to obtain a chromatogram model, reconstructs the chromatogram, and performs subtraction to obtain differential spectrum data.

控制相似度阈值为0.9，取最大值ω后得到匹配结果如表1所示。Taking the Qinggusan compound and single-flavor samples as examples, the Qinggusan sample is used, and its chromatogram is shown in Figure 2. The peak in the time window is used to extract the vertex spectrum, which is stored as a vector stored as a peak-spectrum vector _Xij , where i is the peak number, j is the sample number to which it belongs, and is an irrelevant positive integer. The time window is selected as 1 minute, and the cosine similarity calculation is performed within the time window T. The cosine similarity calculation is performed on any two of _X11 , _X22 ... _Xij , and the result is

The similarity threshold is controlled to 0.9, and the matching results are obtained after taking the maximum value ω as shown in Table 1.

The chromatogram of Qinggu Powder was calibrated based on the contents of Table 1 as the internal standard peak. First, the chromatogram was Fourier transformed to obtain F(w), and then the inverse Fourier transform was performed to resample it. The obtained calibrated chromatogram is shown in Figure 3.

The linear artificial neural network model shown in FIG4 was used to model Qinggu Powder, and the chromatogram of Qinggu Powder was reconstructed using the model. The obtained model is shown in FIG5 .

The standard spectrum of Qinggu San was subtracted from the spectrum of Qinggu San with different medicinal materials missing to obtain the difference spectrum, as shown in Figures 6 to 8, wherein Figure 6 is the standard spectrum of Qinggu San, the spectrum of Huhuanglian Qinggu San missing, the Huhuanglian chromatogram and the difference spectrum, Figure 7 is the standard spectrum of Qinggu San, the spectrum of Gancao Qinggu San missing, the chromatogram and the difference spectrum of Gancao, and Figure 8 is the standard spectrum of Qinggu San, the spectrum of Qinjiao Qinggu San missing, the chromatogram and the difference spectrum of Qinjiao.

By comparing with the chromatogram of single-ingredient medicinal materials, the missing single-ingredient medicinal materials in the compound sample can be easily analyzed.

Table 1. Peak matching results of Qinggusan and single-ingredient decoction pieces

The present invention first matches the sample peaks according to the compound sample chromatogram and its DAD spectrum, and selects the matching time window and similarity threshold according to the actual distribution of the sample peaks. Usually, the time window is controlled between 0.1 and 10 minutes, and the similarity threshold is between (0 and 1), such as using 8 minutes as the time window and 0.2 as the similarity to match the chromatographic peaks. Then, according to the peak matching result as the internal standard, the chromatographic data is resampled and time-corrected using Fourier transform. While fully considering the peak source in the compound sample, the correction range is expanded. Finally, a linear artificial neural network model is used to establish an analytical model for the compound sample and the single-flavor decoction piece sample, and a differential spectrum operation is performed after the chromatogram is reconstructed using the model for analysis.

Claims (9)

1. The chromatographic analysis method of the traditional Chinese medicine compound sample based on the artificial neural network and the difference spectrum is characterized by comprising the following steps of:

for different compound or single traditional Chinese medicine samples, setting fixed wavelength for each sample through a diode array detector to acquire ultraviolet spectrum data, extracting chromatograms and integrating different samples, and performing peak matching of chromatogram integration results on any two samples in a spectrum matching mode;

in the same time window T in the mode of spectrum matching, classifying the peak with the maximum similarity value omega and larger than the threshold value as a uniform peak F _i As a result of chromatographic peak matching, wherein F _i Numbering peak classes;

for each chromatographic peak matching result, taking the chromatographic peak top as a reference, resampling the chromatographic peak by using a Fourier transform method, and carrying out chromatographic time correction according to time intervals to obtain a corrected compound traditional Chinese medicine spectrogram and a corrected single traditional Chinese medicine spectrogram;

respectively fitting an unknown compound sample input linear artificial neural network model and a known compound sample input linear artificial neural network model according to the correction chromatogram, wherein the outputs of the two models are sample coefficient vectors

For unknown compound Chinese medicine sample, the corrected compound Chinese medicine spectrogram is input into linear artificial neural network model by using the unknown compound Chinese medicine sample to obtain the result

Inputting the corrected compound Chinese medicinal spectrogram or single Chinese medicinal spectrogram into linear artificial neural network model to obtain the result

Operating with the spectrogram of a single Chinese medicine to obtain an analytic spectrogram;

the fitting of the unknown compound sample input linear artificial neural network model or the standard compound sample input linear artificial neural network model comprises the following steps:

the corrected chromatogram is a compound Chinese medicinal spectrogram or a single Chinese medicinal spectrogramCorrecting the chromatogram to vector

The spectrogram of the single-ingredient decoction piece is

j is the sample number of the single-ingredient decoction pieces, and

combining to form single-ingredient decoction piece spectrogram matrix D with defined coefficient vector of

The coefficient vector is n when the number of iterations is n

The learning rate is α, the residual error is | Δ | then:

the partial derivatives for Θ are:

there is an iterative formula at this time:

iterating to stop learning when | is less than the set value, to obtain

Wherein, the input of the unknown compound sample input linear artificial neural network model and the standard compound sample input linear artificial neural network model is vector

The model output is the sample coefficient vector

And comparing the analysis spectrogram with the spectrogram of a single traditional Chinese medicine of a known sample to analyze the missing medicinal materials of the decoction pieces in the compound sample.

2. The chromatographic analysis method for the traditional Chinese medicine compound sample based on the artificial neural network and the difference spectrum as claimed in claim 1, wherein the peak matching of the chromatographic integration result is performed on any two samples in a spectral matching manner, comprising the following steps:

extracting a vertex spectrogram using peaks within a time window T, stored as a peak-to-spectrum vector X _ij Wherein i is a peak number, and j is a number of the compound or single decoction piece sample and is an irrelevant positive integer;

to X ₁₁ ，X ₂₂ ……X _ij Any two of the above two are subjected to cosine similarity calculation, and the obtained similarity result is correspondingly:

wherein k is a peak number, m is a sample number of the compound or single decoction piece, j is not equal to m, and the similarity formula is as follows:

wherein j ≠ m.

3. The chromatographic analysis method for the traditional Chinese medicine compound sample based on the artificial neural network and the difference spectrum as claimed in claim 1, characterized in that: the time window T value is selected, the size of the time window T value represents the time range of the peak participating in matching, and the value range is between 0 and 10 minutes.

4. The method for chromatographic analysis of compound Chinese medicine samples based on artificial neural networks and difference spectra as claimed in claim 1, wherein the re-sampling of the chromatogram using fourier transform method and the time correction of the chromatogram according to the time interval are achieved by the following formula:

where t is time, f (t) is the chromatographic signal function, p is the sampling rate, and k is a positive integer.

5. The method for chromatographic analysis of Chinese herbal compound sample based on artificial neural network and difference spectrum as claimed in claim 1, wherein said analytical spectrogram is obtained by the following formula:

wherein P is the difference spectrum result, D is the single-flavor decoction piece spectrogram,

inputting output parameters of a linear artificial neural network model for a standard compound sample,

and inputting output parameters of the linear artificial neural network model for the unknown compound sample.

6. The chromatographic analysis method for the traditional Chinese medicine compound sample based on the artificial neural network and the difference spectrum as claimed in claim 1, wherein the similarity value ω represents the spectral similarity degree of different peaks among samples of different samples, and the value range is between 0 and 1.

7. The chromatographic analysis method for compound traditional Chinese medicine samples based on the artificial neural network and the difference spectrum as claimed in claim 1, wherein the diode array detector can be replaced by one of a high performance liquid chromatography-DAD detector, a high performance liquid chromatography-mass spectrometry detector, a gas chromatography-DAD detector, a gas chromatography-mass spectrometry detector, a high performance capillary electrophoresis chromatography-DAD detector or a high performance capillary electrophoresis-mass spectrometry detector.

8. The method for chromatographic analysis of traditional Chinese medicine compound samples based on artificial neural networks and difference spectra as claimed in claim 1, wherein the construction of the unknown compound sample input linear artificial neural network model or the standard compound sample input linear artificial neural network model minimizes the residual error by one of LSM, gradient descent and least squares.

9. The method for chromatographic analysis of Chinese herbal compound sample based on artificial neural network and difference spectrum as claimed in claim 1, wherein the chromatogram is high performance liquid chromatography, gas chromatography or high performance capillary electrophoresis chromatography.

Images