Chinese mugwort extract test data processing method and system

Abstract

The invention relates to the technical field of electronic data processing, and discloses a Chinese mugwort extract test data processing method and system, and the method comprises the steps: obtaining a data matrix of a Chinese mugwort extract; identifying potential peaks in the data matrix, calculating a weighted fusion value of the potential peaks, and when the weighted fusion value is greater than a first threshold value, identifying the potential peaks as candidate characteristic peaks; calculating a cross validation decision coefficient, and determining a peak group division threshold value; calculating mutual information entropy, when the mutual information entropy is greater than a peak group division threshold, classifying the two candidate characteristic peaks into the same peak group, and selecting a representative characteristic peak; calculating a VIP value sequence of each representative characteristic peak; calculating a comprehensive importance score; and calculating a screening threshold according to the kurtosis and skewness of the comprehensive importance score, and determining the representative characteristic peaks of which the comprehensive importance scores are greater than the screening threshold as quality control characteristic variables. According to the method, the characteristic peak recognition precision, the variable importance value stability and the screening threshold accuracy are improved, and the Chinese mugwort quality evaluation accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electronic data processing, in particular to a test data processing method and system for badiannert extract. BACKGROUND

[0002] As an important traditional Chinese medicine, the quality control of badiannert is a key link to ensure the clinical efficacy. In order to evaluate the quality, modern analytical techniques such as high performance liquid chromatography and gas chromatography are often used to obtain the chemical fingerprint of the extract. These fingerprint data usually have complex characteristics such as high dimension, strong collinearity between multiple variables and low signal-to-noise ratio.

[0003] In the data preprocessing stage, the traditional feature peak identification method mostly depends on the simple signal intensity or signal-to-noise ratio threshold. This way is not only easy to misjudge the baseline noise, solvent peak and other pseudo peaks as effective chemical components, but also may miss some real component peaks with low intensity but significant biological activity, thereby affecting the accuracy of the initial variable set. In addition, there are a large number of highly correlated variables in the fingerprint data caused by chemical structure similar homologues or isomers, and these redundant information will interfere with the stability and accuracy of the subsequent model.

[0004] In order to screen out the markers directly related to efficacy from complex chemical fingerprints, chemometric methods, especially partial least squares (PLS), are widely used to establish the correlation model between chemical components and efficacy indicators. After the model is established, variable projection importance (VIP) and other indicators are usually used to evaluate the contribution of each variable. However, the VIP value calculated by single modeling is very sensitive to the slight disturbance of sample set, and the stability is poor, which makes it difficult to ensure the reliability of the results. More importantly, when screening key variables, a fixed empirical threshold is generally used for judgment. This "one-size-fits-all" method ignores the distribution characteristics of the data itself, and has strong subjectivity and blindness.

[0005] Therefore, how to establish a systematic data processing method to overcome the shortcomings of traditional methods, realize the accurate identification of feature peaks, effective elimination of redundant variables, stable evaluation of variable importance and adaptive determination of screening threshold, is a technical problem to be solved in the field of traditional Chinese medicine quality control. SUMMARY

[0006] The present application provides a test data processing method and system for badiannert extract to solve the problems of low feature peak recognition accuracy, poor stability of variable importance value and inaccurate screening threshold in the prior art, which leads to low accuracy of badiannert quality evaluation.

[0007] In the first aspect, the test data processing method for badiannert extract of the present application comprises the following steps: The chromatographic or spectral test data matrix of the Wobai extract is obtained; potential peaks in the data matrix are identified, the peak shape symmetry of the potential peaks and the local signal-to-noise ratio in a preset window centered on the vertex of the potential peaks are calculated, a weighted fusion value is obtained, and when the weighted fusion value is greater than a first threshold value, the potential peak is identified as a candidate characteristic peak; an initial partial least squares model is established by using all candidate characteristic peaks and efficacy index data of the Wobai extract, and a cross-validation determination coefficient of the initial partial least squares model is calculated; according to the value of the cross-validation determination coefficient, a peak group division threshold value is determined; mutual information entropy between response intensity vectors of any two candidate characteristic peaks in all sample data is calculated, and when the mutual information entropy is greater than the peak group division threshold value, the two candidate characteristic peaks are classified into the same peak group, and the peak with the highest weighted fusion value in each peak group is selected as the representative characteristic peak of the peak group; a final partial least squares model is established by using the representative characteristic peaks and the efficacy index data, a variable projection importance VIP value of each representative characteristic peak is calculated in a sub-model established in each sampling by Monte Carlo resampling for multiple random sampling of the sample, and a VIP value sequence of each representative characteristic peak is obtained; the mean of the VIP value sequence and the reciprocal of the coefficient of variation of the VIP value sequence are calculated, and the product of the two is taken as the comprehensive importance score of the representative characteristic peak; a screening threshold value is calculated according to the kurtosis and skewness of the comprehensive importance scores of all representative characteristic peaks, and the representative characteristic peak with a comprehensive importance score greater than the screening threshold value is determined as a quality control characteristic variable.

[0008] Preferably, the peak shape symmetry of the potential peak comprises: For each potential peak, 10 data points on the left and right of the peak vertex are taken, the area A on the left of the peak vertex and the area B on the right of the peak vertex are calculated respectively, and the peak shape symmetry is calculated by the following formula .

[0009] Preferably, the local signal-to-noise ratio in the preset window centered on the vertex of the potential peak is calculated by the following method: The peak height of the potential peak is taken as the signal value, and the standard deviation of the first 100 data points of the data matrix is taken as the noise value, and the local signal-to-noise ratio SNR is calculated.

[0010] Preferably, the weighted fusion value is calculated by the following formula: ; wherein, is the weighted fusion value, is the peak shape symmetry, SNR is the local signal-to-noise ratio, is a preset reference signal-to-noise ratio.

[0011] Preferably, the peak group division threshold value is calculated by the following formula: ;​ wherein, is a peak group division threshold, is a cross-validation determination coefficient, The value range of is [0, 1].

[0012] Preferably, the VIP value sequence coefficient of variation is equal to the ratio of the standard deviation to the mean of the VIP value sequence of the representative characteristic peak.

[0013] Preferably, the comprehensive importance score of the representative characteristic peak is calculated by the following formula: ; wherein, is the comprehensive importance score of the i-th representative characteristic peak, is the mean of the VIP value sequence of the i-th representative characteristic peak, is the standard deviation of the VIP value sequence of the i-th representative characteristic peak, is a preset tiny normal number to prevent the denominator from being zero.

[0014] Preferably, the screening threshold is calculated according to the kurtosis and skewness of the comprehensive importance scores of all representative characteristic peaks, comprising: calculating the mean , standard deviation , skewness S and kurtosis K of the comprehensive importance scores of all representative characteristic peaks; calculating the screening threshold by the following formula: .

[0015] Preferably, the chromatographic or spectral test data matrix of the eupolyphaga extract is obtained, comprising: The chromatographic or spectral test data matrix of the eupolyphaga extract is obtained by using high performance liquid chromatography-mass spectrometry technology.

[0016] In the second aspect, the eupolyphaga extract test data processing system of the present application comprises a memory and a processor, and the memory stores computer instructions. When the processor executes the computer instructions, the above-mentioned eupolyphaga extract test data processing method is realized.

[0017] The beneficial effects of the present application are: the present application can identify real characteristic peaks from complex chromatographic or spectral data by calculating the weighted fusion value of peak symmetry and local signal-to-noise ratio, eliminate the interference of baseline noise and pseudo peaks, and improve the quality of the initial variable set. The mutual information entropy is used to group the candidate characteristic peaks, and the peak with the highest quality in the group is selected as the representative, which solves the problems of collinearity and redundancy of data and simplifies the model. In the key variable screening stage, the present application overcomes the defect that the importance value of the variable is unstable in single modeling calculation by using the Monte Carlo resampling technique, and constructs a comprehensive score that measures the importance size and stability at the same time. The screening threshold is determined based on the overall distribution characteristics of the comprehensive importance score of all variables, so that the characteristic variables truly related to the efficacy of Radix Wobain can be screened out. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A flowchart of the method for processing test data of Radix Wobain extract provided by the embodiments of the present application is shown. DETAILED DESCRIPTION

[0019] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0020] As shown in Figure 1 The embodiments of the method for processing test data of Radix Wobain extract provided by the present application include the following steps: S1, obtaining a chromatographic or spectral test data matrix of Radix Wobain extract; identifying potential peaks in the data matrix, calculating the peak shape symmetry of the potential peaks and the local signal-to-noise ratio in a preset window centered on the vertex of the potential peaks, and obtaining a weighted fusion value; when the weighted fusion value is greater than a first threshold value, the potential peak is identified as a candidate characteristic peak.

[0021] The chromatographic or spectral test data matrix of Radix Wobain extract is obtained by using high performance liquid chromatography-mass spectrometry technology. Specifically, different batches of Radix Wobain extract samples are detected by using high performance liquid chromatography-mass spectrometry technology, and one total ion chromatogram is obtained for each sample. The chromatograms of all samples are aligned on a unified time axis, with the retention time as the horizontal coordinate and the response intensity as the vertical coordinate, to construct a data matrix X. In the matrix, each row represents a Radix Wobain extract sample, and each column represents the response intensity at a retention time point.

[0022] The data matrix is subjected to baseline correction and smoothing processing, and a first derivative method is used to preliminarily identify all potential peak vertices. In order to filter out pseudo peaks or noise peaks by comprehensively evaluating the quality of the peak shape, in an optional embodiment, the calculation of the peak shape symmetry of the potential peaks includes: For each potential peak, take 10 data points on each side of the peak's vertex, and calculate the area A on the left side of the peak's vertex and the area B on the right side of the peak's vertex. Then, calculate the peak's symmetry using the following formula. : .

[0023] The local signal-to-noise ratio within a preset window centered on the potential peak vertex is calculated in the following way: The local signal-to-noise ratio (SNR) is calculated using the peak height of the potential peak as the signal value and the standard deviation of the first 100 data points of the data matrix as the noise value.

[0024] The weighted fusion value is calculated using the following formula: ; in, For weighted fusion values, The peak shape symmetry is represented by SNR, which is the local signal-to-noise ratio. This is the preset reference signal-to-noise ratio.

[0025] For example, consider a potential peak: Through integration, the area of ​​the 10 data points to the left of the peak's apex is calculated to be A = 12000, and the area of ​​the 10 data points to the right is B = 11400. From this, the peak's symmetry is calculated to be S = 0.95. The peak height is 80000, and the baseline noise in the initial part of the data matrix (i.e., the standard deviation of the first 100 data points) is 400. Therefore, the local signal-to-noise ratio (SNR) of this peak is 200.

[0026] Assuming a preset reference signal-to-noise ratio The symmetry of 0.95 and the signal-to-noise ratio of 200 were substituted into the weighted fusion value formula for calculation. The resulting weighted fusion value (WIV) was approximately 0.96. This value is higher than the preset first threshold of 0.75, indicating that the potential peak has good symmetry and a high signal-to-noise ratio, and can be identified as a high-quality true chromatographic peak, thus retaining it for subsequent analysis. Conversely, if a peak exhibits severe tailing or an extremely low signal-to-noise ratio, its calculated weighted fusion value will be lower than 0.75, and it should be identified and discarded.

[0027] S2. An initial partial least squares model is established using all candidate characteristic peaks and efficacy index data of Artemisia argyi extract. The cross-validation determination coefficient of the initial partial least squares model is calculated. Based on the value of the cross-validation determination coefficient, the peak group division threshold is determined.

[0028] The peak area or peak height data of all candidate feature peaks in each sample are used to construct a new independent variable matrix. Simultaneously, efficacy index data (e.g., in vitro anti-inflammatory activity assay values) corresponding to each Artemisia argyi extract sample are obtained to form the dependent variable vector y. Partial least squares regression modeling is performed using the independent variable matrix as the independent variable and y as the dependent variable. Leave-one-out cross-validation is preferred during the modeling process, i.e., one sample is selected from n samples each time as the validation set, and the remaining n-1 samples are used as the training set, repeated n times. The sum of squared prediction errors (PRESS) and the sum of squared dependent variables are calculated. And according to the formula The cross-validation determination coefficient of the model is obtained. .

[0029] To allow the stringency of peak group partitioning to be adjusted based on the predictive power of the model within the dataset, in an optional embodiment, the peak group partitioning threshold is calculated using the following formula: ; in, Thresholds are assigned to peak groups. The coefficient of determination for cross-validation. The value range is [0, 1].

[0030] For example, when using a partial least squares discriminant analysis model to process data, the model calculated through seven-fold cross-validation... This indicates that the model has strong predictive power and stability. At this point, the peak group segmentation threshold... The threshold was set at 0.78. A higher threshold means that only highly correlated peaks will be grouped into the same peak group, which helps to identify functionally relevant metabolites in high-quality data.

[0031] If the dataset is of poor quality or the differences between groups are not significant, the model will suffer. At this point, the peak group segmentation threshold... The threshold is correspondingly reduced to 0.34. This relatively low threshold relaxes the criteria for peak grouping, allowing peaks with weaker correlations to cluster together, thus avoiding the loss of potentially important information in data with low signal-to-noise ratios.

[0032] S3. Calculate the mutual information entropy between the response intensity vectors of any two candidate feature peaks in all sample data. When the mutual information entropy is greater than the peak group division threshold, the two candidate feature peaks are assigned to the same peak group. Within each peak group, the peak with the highest weighted fusion value is selected as the representative feature peak of that peak group. The final partial least squares model is established using the representative feature peaks and efficacy index data. The samples are randomly sampled multiple times through Monte Carlo resampling. In the sub-model established in each sampling, the variable projection importance (VIP) value of each representative feature peak is calculated to obtain the VIP value sequence of each representative feature peak.

[0033] Specifically, for any two candidate feature peaks i and j, extract their response intensity vectors across all n samples. and Discretize the data from these two vectors (e.g., divide them into 10 intervals), and calculate the marginal entropy for each interval. , and their joint entropy The formula for calculating mutual information entropy (MI) is as follows: Traverse all peak pairs to construct a mutual information entropy matrix. Use clustering to group two feature peaks with an MI value greater than the peak grouping threshold into one class, iterating until all peaks are grouped. Within each divided peak group, compare the weighted fusion values ​​of all candidate feature peaks calculated in the previous steps, and select the peak with the highest weighted fusion value as the representative feature peak of that peak group.

[0034] Construct an independent variable matrix from all representative characteristic peaks A final partial least squares (PLS) model is established using the efficacy index vector y. The number of resampling iterations is set to M=500. In the i-th iteration, n samples are randomly drawn with replacement from a total of n samples to form a bootstrap resampling sample set. A PLS sub-model is built using the data in this sample set, and the VIP value of each representative feature peak is calculated based on this sub-model. The VIP value is calculated based on its cumulative contribution to the independent and dependent variables. This process is repeated M times, resulting in a sequence of M VIP values ​​for each representative feature peak.

[0035] S4. Calculate the mean of the VIP value sequence and the reciprocal of the coefficient of variation of the VIP value sequence, and use the product of the two as the comprehensive importance score of the representative feature peak. Calculate the screening threshold based on the kurtosis and skewness of the comprehensive importance scores of all representative feature peaks, and determine the representative feature peaks with comprehensive importance scores greater than the screening threshold as quality control feature variables.

[0036] In order to screen out characteristic peaks that are both highly important and stable, in an optional embodiment, the coefficient of variation of the VIP value sequence is equal to the ratio of the standard deviation to the mean of the VIP value sequence of the representative characteristic peak.

[0037] The comprehensive importance score of the representative characteristic peaks is calculated using the following formula: ; in, The overall importance score for the i-th representative characteristic peak is given. Let be the mean of the VIP value sequence of the i-th representative characteristic peak. Let be the standard deviation of the VIP value sequence of the i-th representative characteristic peak. To prevent the use of pre-defined tiny positive numbers with a denominator of zero.

[0038] For example, suppose that through multiple cross-validation or bootstrap sampling, representative feature peak A yields a sequence of projected importance (VIP) values ​​for ten variables, as shown below: {2.5, 2.6, 2.4, 2.5, 2.7, 2.5, 2.4, 2.6, 2.5, 2.3}. The calculated mean of this sequence is 2.5, and the standard deviation is 0.11. Therefore, its composite importance score (CIS) is approximately 2.52 divided by 0.11, resulting in approximately 56.8. In contrast, another representative feature peak, B, has a VIP value sequence of {1.5, 3.5, 2.8, 1.2, 3.1, 2.5, 1.8, 3.2, 1.4, 3.0}. Although the mean of this sequence is also approximately 2.4, similar to peak A, its standard deviation is 0.82, resulting in a composite importance score (CIS) of only 7.0. Thus, although the average importance of the two feature peaks is similar, peak A has a significantly higher importance score than peak B. This is mainly because peak A consistently exhibits high importance across different subsets of data, indicating greater stability. By penalizing the unstable nature of importance, this method can prioritize the selection of more reliable biomarkers.

[0039] In an optional embodiment, calculating the screening threshold based on the kurtosis and skewness of the comprehensive importance score of all representative characteristic peaks includes: Calculate the mean of the overall importance scores for all representative characteristic peaks. Standard deviation Skewness S and kurtosis K; Calculate the screening threshold using the following formula. : .

[0040] For example, instead of using a fixed empirical threshold, the overall shape of the data distribution is used to set the screening criteria. Suppose that after analyzing a batch of sample data, a set of Comprehensive Importance Score (CIS) values ​​for all representative characteristic peaks is obtained. Statistical analysis of this set yields a mean of 20, a standard deviation of 15, a skewness of 3.0, and a kurtosis of 5.0. High skewness and high kurtosis indicate that the score distribution exhibits a clear right-skewed, peak-like shape, meaning that there are a few characteristic peaks with extremely high scores. In this case, the multiplier term in the threshold calculation formula is 1.1, which is greater than the preset minimum value of 0.5; therefore, 1.1 is chosen as the final multiplier. The final screening threshold is thus determined. The calculated value is 36.5. A higher threshold can effectively monitor important features in the long tail of the distribution. If the rating distribution of another dataset is close to normal, with a skewness of 0.1 and a kurtosis of 0.2, the multiplier term is calculated to be 0.04. This value is less than 0.5, so the default value of 0.5 is used as the multiplier. In this case, the threshold will be 27.5, which is a relatively lenient standard, thus adapting to the screening needs of different data features. By traversing all representative feature peaks, the comprehensive importance score is compared with the screening threshold. All representative feature peaks with a comprehensive importance score greater than the screening threshold are identified as quality control feature variables related to efficacy.

[0041] The implementation principle of the data processing method for Artemisia argyi extract in this invention is as follows: By calculating the weighted fusion value of peak symmetry and local signal-to-noise ratio, this invention can accurately identify true characteristic peaks from complex chromatographic or spectroscopic data, effectively filtering out baseline noise and spurious peak interference, thereby significantly improving the quality of the initial variable set. Secondly, to address the issues of data collinearity and redundancy, this invention uses mutual information entropy to group candidate characteristic peaks and selects the highest-quality peak within each group as a representative to simplify the model. In the key variable screening stage, this invention introduces Monte Carlo resampling technology to overcome the instability of variable importance obtained from a single modeling iteration and constructs a comprehensive scoring system that can simultaneously assess importance and stability. Finally, this invention no longer relies on fixed empirical thresholds but dynamically determines the screening criteria based on the overall distribution characteristics of the comprehensive scores of all variables, thereby achieving accurate identification of characteristic variables truly related to the efficacy of Artemisia argyi.

[0042] An embodiment of the Artemisia argyi extract test data processing system provided by the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the Artemisia argyi extract test data processing method in the above embodiment.

[0043] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for processing experimental data of Artemisia argyi extract, characterized in that, include: Acquire the chromatographic or spectroscopic test data matrix of Artemisia argyi extract; identify potential peaks in the data matrix, calculate the peak shape symmetry of potential peaks and the local signal-to-noise ratio within a preset window centered on the peak apex, and obtain a weighted fusion value. When the weighted fusion value is greater than a first threshold, the potential peak is identified as a candidate feature peak; establish an initial partial least squares model using all candidate feature peaks and efficacy index data of Artemisia argyi extract, and calculate the cross-validation determination coefficient of the initial partial least squares model; determine the peak group division threshold based on the value of the cross-validation determination coefficient; calculate the mutual information entropy between the response intensity vectors of any two candidate feature peaks in all sample data. When the mutual information entropy is greater than the peak group division threshold, the two candidate feature peaks are assigned to the same peak group, and the peak with the highest weighted fusion value in each peak group is selected as the representative feature peak of that peak group; A final partial least squares model is established using representative characteristic peaks and efficacy index data. Multiple random samplings are performed on the sample using Monte Carlo resampling. In each sub-model established by the sampling, the variable projection importance (VIP) value of each representative characteristic peak is calculated, resulting in a VIP value sequence for each representative characteristic peak. The mean of the VIP value sequence and the reciprocal of the coefficient of variation of the VIP value sequence are calculated, and their product is used as the comprehensive importance score of the representative characteristic peak. A screening threshold is calculated based on the kurtosis and skewness of the comprehensive importance scores of all representative characteristic peaks. Representative characteristic peaks with comprehensive importance scores greater than the screening threshold are identified as quality control characteristic variables.

2. The method for processing experimental data of Artemisia argyi extract according to claim 1, characterized in that, The calculation of the potential peak shape symmetry includes: For each potential peak, take 10 data points on each side of the peak's vertex, and calculate the area A on the left side of the peak's vertex and the area B on the right side of the peak's vertex. Then, calculate the peak's symmetry using the following formula. : 。 3. The method for processing experimental data of Artemisia argyi extract according to claim 2, characterized in that, The local signal-to-noise ratio within a preset window centered on the potential peak vertex is calculated in the following way: The local signal-to-noise ratio (SNR) is calculated using the peak height of the potential peak as the signal value and the standard deviation of the first 100 data points of the data matrix as the noise value.

4. The method for processing experimental data of Artemisia argyi extract according to claim 3, characterized in that, The weighted fusion value is calculated using the following formula: ； in, For weighted fusion values, The peak shape symmetry is represented by SNR, which is the local signal-to-noise ratio. This is the preset reference signal-to-noise ratio.

5. The method for processing experimental data of Artemisia argyi extract according to claim 1, characterized in that, The peak group division threshold is calculated using the following formula: ； in, Thresholds are assigned to peak groups. The coefficient of determination for cross-validation. The value range is [0, 1].

6. The method for processing experimental data of Artemisia argyi extract according to claim 1, characterized in that, The coefficient of variation of the VIP value sequence is equal to the ratio of the standard deviation to the mean of the VIP value sequence of the representative characteristic peak.

7. The method for processing experimental data of Artemisia argyi extract according to claim 6, characterized in that, The comprehensive importance score of the representative characteristic peaks is calculated using the following formula: ； in, The overall importance score for the i-th representative characteristic peak is given. Let be the mean of the VIP value sequence of the i-th representative characteristic peak. Let be the standard deviation of the VIP value sequence of the i-th representative characteristic peak. To prevent the use of pre-defined tiny positive numbers with a denominator of zero.

8. The method for processing experimental data of Artemisia argyi extract according to claim 1, characterized in that, The calculation of the screening threshold based on the kurtosis and skewness of the comprehensive importance score of all representative characteristic peaks includes: Calculate the mean of the overall importance scores for all representative characteristic peaks. Standard deviation Skewness S and kurtosis K; Calculate the screening threshold using the following formula. : 。 9. The method for processing experimental data of Artemisia argyi extract according to claim 1, characterized in that, The method for obtaining the chromatographic or spectroscopic test data matrix of Artemisia argyi extract includes: Chromatographic or spectroscopic experimental data matrix of Artemisia argyi extract was obtained by high performance liquid chromatography-mass spectrometry.

10. A data processing system for Artemisia argyi extract experiments, characterized in that, It includes a memory and a processor. The memory stores computer instructions. When the processor executes the computer instructions, it implements the test data processing method for Artemisia argyi extract as described in any one of claims 1-9.

Images