A traditional Chinese medicine fingerprint detection method for traditional Chinese medicine preparations

Abstract

The present invention relates to the technical field of chemical component analysis, and in particular to a method for detecting a Chinese medicine fingerprint of a Chinese medicine preparation. The method comprises the following steps: obtaining high-dimensional spectrum data of Niushe Kaiqiao Powder, and performing qualitative characteristic analysis of the chemical components of Niushe Kaiqiao Powder to obtain a chemical component characteristic matrix of Niushe Kaiqiao Powder; obtaining multimodal production data of Niushe Kaiqiao Powder, and performing high-dimensional feature fusion on the chemical component characteristic matrix of Niushe Kaiqiao Powder and the multimodal production data of Niushe Kaiqiao Powder to obtain a sample feature embedding vector; constructing a chemical component dynamic network according to the sample feature embedding vector, thereby obtaining a Niushe Kaiqiao Powder component collaborative network; generating a global fingerprint spectrum based on the Niushe Kaiqiao Powder component collaborative network, and performing potential abnormal point spectrum peak correction to obtain an optimized fingerprint spectrum; constructing a feature model according to the optimized fingerprint spectrum to obtain a Niushe Kaiqiao Powder quality evaluation model. The present invention can improve the accuracy of quality control of Chinese medicine preparations.

Description

Technical Field

[0001] The invention relates to the technical field of chemical component analysis, and in particular to a method for detecting a traditional Chinese medicine fingerprint spectrum of a traditional Chinese medicine preparation. Background Art

[0002] In recent years, Chinese medicine preparations, as an important part of traditional Chinese medicine, have attracted widespread attention due to their precise efficacy and low toxicity and side effects. However, Chinese medicine preparations are usually made from a variety of Chinese medicinal materials through complex processes, and their chemical composition is complex, involving the synergistic effects of multiple active ingredients and auxiliary ingredients. Traditional Chinese medicine quality control methods are mainly based on the content detection of a single chemical component or several specific components, such as the determination of the content of certain index components by high performance liquid chromatography (HPLC), ultraviolet spectrophotometry (UV), thin layer chromatography (TLC), etc. Although this method can quantitatively analyze certain components, since the efficacy of Chinese medicine preparations usually depends on the synergistic effects of multiple components, the detection of a single component cannot fully reflect the overall characteristics of Chinese medicine, which may lead to deviations in quality evaluation results. Summary of the invention

[0003] Based on this, it is necessary for the present invention to provide a method for detecting a Chinese medicine fingerprint of a Chinese medicine preparation to solve at least one of the above technical problems.

[0004] To achieve the above object, a method for detecting a Chinese medicine fingerprint of a Chinese medicine preparation comprises the following steps:

[0005] Step S1: obtaining high-dimensional spectrum data of Niu Musk Kaiqiao Powder, and performing chemical component spectrum decomposition on the high-dimensional spectrum data of Niu Musk Kaiqiao Powder, so as to obtain a chemical component spectrum of Niu Musk Kaiqiao Powder; performing qualitative characteristic analysis of the chemical components of Niu Musk Kaiqiao Powder according to the chemical component spectrum of Niu Musk Kaiqiao Powder, so as to obtain a chemical component characteristic matrix of Niu Musk Kaiqiao Powder;

[0006] Step S2: obtaining the multimodal production data of Niu Musk Kaiqiao Powder, and unifying the Niu Musk Kaiqiao Powder sample source feature space of the multimodal production data of Niu Musk Kaiqiao Powder, thereby obtaining the Niu Musk Kaiqiao Powder sample source feature space; performing high-dimensional feature fusion on the Niu Musk Kaiqiao Powder chemical component feature matrix and the Niu Musk Kaiqiao Powder sample source feature space, thereby obtaining a sample feature embedding vector;

[0007] Step S3: performing sample relationship structure analysis according to the sample feature embedding vector to obtain the relationship structure data of the ingredients of Niushe Kaiqiao Powder, and constructing a chemical component dynamic network based on the relationship structure data of the ingredients of Niushe Kaiqiao Powder to obtain the synergistic network of the ingredients of Niushe Kaiqiao Powder;

[0008] Step S4: generating a global fingerprint spectrum based on the collaborative network of Niushe Kaiqiao Powder components, and screening the global fingerprint spectrum for potential abnormal points, thereby obtaining abnormal peak data of the global fingerprint spectrum; performing peak correction on the abnormal peak data of the global fingerprint spectrum, thereby obtaining an optimized fingerprint spectrum;

[0009] Step S5: construct and verify the characteristic model according to the optimized fingerprint spectrum, so as to obtain the quality evaluation model of Niushe Kaiqiao Powder.

[0010] The present invention utilizes high-dimensional spectrogram data decomposition technology to deeply mine the complex chemical component information of Niushe Kaiqiao Powder, accurately analyze the component characteristics, and provide a data basis for the construction of the chemical component feature matrix. This full-spectrum analysis method avoids the limitations of traditional single component analysis and effectively improves the comprehensive analysis capability of complex chemical components. In multimodal production data processing, the data offset problem caused by source differences is overcome through unified operation of the sample source feature space, ensuring the consistency and reliability of the data. Combined with the high-dimensional fusion of the chemical component feature matrix and the source feature space, the interactive information between multidimensional features is retained, and the accurate expression of sample features is achieved, laying the foundation for subsequent relationship analysis and quality evaluation. Based on the sample feature embedding vector, the sample relationship structure analysis can deeply mine the synergy between chemical components and fully reveal the dynamic relationship between the components in the complex system of Niushe Kaiqiao Powder. By constructing a chemical component synergy network, not only the global characteristics of the synergy are visualized, but also the dynamic cooperation law between the components is reflected, providing a new perspective for a comprehensive understanding of the mechanism of action of the complex system of traditional Chinese medicine. In the process of generating global fingerprint spectra, the introduction of potential outlier screening and peak correction technology effectively solves the error problem caused by noise or outliers in traditional spectrum analysis. The optimized fingerprint spectrum can more truly and accurately reflect the overall chemical properties and component distribution of Niushe Kaiqiao Powder, providing comprehensive and reliable quality data support. By optimizing the fingerprint spectrum to construct a quality evaluation model, a scientific evaluation of the quality of Niushe Kaiqiao Powder can be achieved as a whole. The model can accurately locate the key ingredients and their action characteristics, and fully reflect the contribution of the synergy between ingredients to the overall quality. This evaluation method breaks through the bottleneck of traditional single ingredient detection, provides a comprehensive and systematic quality control method, and provides important technical support for the standardization and normalization of Niushe Kaiqiao Powder. At the same time, this method shows the direction of modernization and intelligent development of quality control of traditional Chinese medicine, and brings a new solution to the technological progress of the traditional Chinese medicine preparation industry.

[0011] Optionally, step S1 specifically includes:

[0012] Step S11: obtaining high-dimensional spectrum data of Niushe Kaiqiao Powder, and performing spectrum data preprocessing on the high-dimensional spectrum data of Niushe Kaiqiao Powder, so as to obtain high-dimensional spectrum data of Niushe Kaiqiao Powder to be analyzed;

[0013] Step S12: obtaining chemical component characteristic data based on a preset chemical library, and integrating the chemical component qualitative characteristics according to the chemical component characteristic data, thereby obtaining a chemical component characteristic matrix;

[0014] Step S13: performing chemical component spectrum decomposition on the high-dimensional spectrum data of Niushe Kaiqiao Powder to be analyzed, thereby obtaining a chemical component spectrum, and performing chemical component relative intensity calculation on the chemical component spectrum, thereby obtaining chemical component intensity data;

[0015] Step S14: performing chemical component-specific spectrum decomposition on the chemical component spectrum according to the chemical component characteristic matrix and the chemical component intensity data, thereby obtaining the chemical component spectrum of Niushe Kaiqiao Powder;

[0016] Step S15: Perform qualitative characteristic analysis of the chemical components of Niushe Kaiqiao Powder according to the chemical component spectrum of Niushe Kaiqiao Powder, so as to obtain the chemical component characteristic matrix of Niushe Kaiqiao Powder.

[0017] The present invention effectively eliminates noise and data deviation by preprocessing the high-dimensional spectrogram data of Niushe Kaiqiao Powder, improves the accuracy of subsequent analysis, and provides high-quality input data support for the analysis of complex chemical components. The chemical component characteristic data in the chemical library are integrated to ensure the comprehensive coverage and information accuracy of the component characteristics, and lay a solid foundation for the construction of the chemical component characteristic matrix. Through the chemical component spectrum decomposition of the high-dimensional spectrum, a clear chemical component spectrum is successfully extracted from the complex raw data, effectively solving the problems of multi-component overlap and background interference. Combined with the relative intensity calculation of the chemical components, the intensity of each component can be accurately quantified, providing key data support for further analysis of the component characteristics and their role in the overall compatibility. Subsequently, based on the operation of specific spectrum decomposition, the refined analysis of the component spectrum is further realized, so that the difference and specificity of the complex component system can be fully displayed, providing a comprehensive basis for a deeper quality evaluation. Through the qualitative characteristic analysis of chemical components, the core chemical components of Niushe Kaiqiao Powder can be deeply analyzed, and the types, structures and potential relationships of the components with efficacy can be fully revealed. The chemical composition characteristic matrix finally constructed presents the chemical properties of the samples in a comprehensive and high-precision form, overcoming the limitations of single component or simple content detection in the quality evaluation of traditional Chinese medicine, and embodies the advantages of systematic, comprehensive and refined analysis.

[0018] Optionally, step S14 is specifically:

[0019] Step S141: performing Pearson correlation coefficient feature matching on the chemical component feature matrix and the chemical component intensity data, thereby obtaining a preliminary matching matrix;

[0020] Step S142: performing sparse coding specific factor separation based on the preliminary matching matrix, thereby obtaining a specific factor matrix;

[0021] Step S143: extracting the spectrogram features of the specific factor matrix to obtain specific factor spectrogram feature data, and performing intensity normalization on the specific factor spectrogram feature data and the chemical component intensity data to obtain intensity normalized spectrogram feature data;

[0022] Step S144: Calculating the relative contribution of components according to the intensity normalized spectrum feature data, thereby obtaining relative contribution data of components, and constructing a relative contribution matrix based on the relative contribution data of components;

[0023] Step S145: reconstructing the specific spectrum based on the relative contribution matrix and the specific factor matrix, thereby obtaining the specific spectrum of the chemical components of Niushe Kaiqiao Powder;

[0024] Step S146: According to the chemical component specific spectrum of Niushe Kaiqiao Powder, the original chemical component spectrum is specifically tested and optimized to obtain the chemical component spectrum of Niushe Kaiqiao Powder.

[0025] The present invention can effectively mine the correlation between the chemical component feature matrix and the intensity data through Pearson correlation coefficient feature matching, accurately identify the potential association between the components, and provide accurate preliminary data support for subsequent specific factor separation. The introduction of sparse coding specific factor separation further extracts specific factors with independent characteristics from high-dimensional data, which helps to focus on key components, reduce the interference of redundant data, and significantly improve the efficiency and pertinence of spectrogram analysis. In the specific factor spectrum feature extraction link, the comparative analysis of different component characteristics on a unified scale is realized through the intensity normalization operation, which overcomes the deviation caused by scale differences in the original intensity data and ensures the accuracy and consistency of the analysis results. The calculation of the relative contribution of the components and the construction of the relative contribution matrix provide a clear quantitative basis for the role of the chemical components in the overall spectrum, making the relationship of the synergistic effect of the components more intuitive and quantifiable. Based on this, the specific spectrum reconstruction further reorganizes the complex original data into a chemical component specific spectrum dominated by key features, which greatly improves the resolution and expression ability of the spectrum data. The original chemical component spectra were specifically tested and optimized through chemical component specific spectra, which fully verified the reliability and accuracy of the component spectra, and improved the overall expression effect of the chemical component spectra through optimization. This process realizes the full chain analysis from preliminary data matching to spectrum optimization, greatly overcoming the limitations of traditional methods in spectrum specificity and contribution analysis, and providing a scientific and efficient solution for the chemical component analysis and quality evaluation of complex Chinese medicine systems.

[0026] Optionally, step S15 is specifically:

[0027] Step S151: extracting peak features from the chemical component spectrum of Niushe Kaiqiao Powder to obtain a peak feature matrix;

[0028] Step S152: performing correlation matching according to the chemical component feature matrix and the peak feature matrix, thereby obtaining a chemical component annotation matrix;

[0029] Step S153: performing spectrum characteristic pattern recognition on the chemical component annotation matrix to obtain spectrum characteristic pattern data; performing characteristic pattern component distribution analysis based on the spectrum characteristic pattern data to obtain spectrum characteristic pattern component distribution data;

[0030] Step S154: integrating the chemical component qualitative characteristics based on the spectrum characteristic pattern data and the spectrum characteristic pattern component distribution data, thereby obtaining the chemical component qualitative characteristic data;

[0031] Step S155: construct the chemical component characteristic matrix of Niushe Kaiqiao Powder based on the peak characteristic matrix, the chemical component annotation matrix and the chemical component qualitative characteristic data.

[0032] The present invention obtains a peak feature matrix that can accurately characterize the spectrogram structure by extracting the spectrogram peak features of the chemical component spectrogram, providing rich and detailed basic data for subsequent analysis. In the correlation matching link, the matching analysis of the chemical component feature matrix and the peak feature matrix realizes the association between the chemical components and the spectrogram features, making the annotation of the chemical components more scientific and reliable, thereby significantly improving the accuracy of the spectrogram analysis. The introduction of spectral feature pattern recognition breaks through the limitations of the traditional method on the ability to analyze the spectrogram data pattern. By identifying the spectral feature pattern and performing component distribution analysis, the distribution law of the chemical components in the spectrogram is effectively revealed, and the problem of complex spectral information dispersion and difficulty in classification is solved, which lays the foundation for the normalization and systematization of spectral information. Based on the integration of spectral feature pattern data and component distribution data, the qualitative characteristics of the chemical components are fully extracted and optimized, thereby enhancing the accuracy of the qualitative analysis of the complex components in Niushe Kaiqiao Powder. Through the joint modeling of the peak feature matrix, the chemical component annotation matrix and the qualitative feature data of the chemical components, the chemical component feature matrix of Niushe Kaiqiao Powder is constructed. This matrix can not only comprehensively characterize the characteristics of chemical components, but also realize multi-dimensional correlation analysis at the molecular level and spectral level, significantly improving the efficiency of systematic analysis of complex components of traditional Chinese medicine. The overall method overcomes the shortcomings of traditional technology in systematic characterization of the relationship and characteristics of synergistic components of traditional Chinese medicine through multi-level data mining and integrated analysis, and provides scientific and precise technical support for the quality control and in-depth analysis of Niushe Kaiqiao Powder.

[0033] Optionally, step S2 specifically includes:

[0034] Step S21: acquiring the multimodal production data of Niushe Kaiqiao Powder, and performing data preprocessing on the multimodal production data of Niushe Kaiqiao Powder, so as to obtain the multimodal production data of Niushe Kaiqiao Powder to be analyzed;

[0035] Step S22: performing a principal component analysis of global sample source characteristics on the multimodal production data of Niushe Kaiqiao Powder to be analyzed, thereby obtaining principal component multimodal production data, and performing nonlinear distribution feature dimensionality reduction on the principal component multimodal production data, thereby obtaining Niushe Kaiqiao Powder sample source data;

[0036] Step S23: performing feature alignment on the Niushe Kaiqiao Powder sample source data to obtain aligned sample source data, and performing unified feature space mapping based on the aligned sample source data to obtain the Niushe Kaiqiao Powder sample source feature space;

[0037] Step S24: aligning the dimension and scale of the chemical component feature matrix of Niushe Kaiqiao Powder and the feature space of the source of Niushe Kaiqiao Powder samples, so as to obtain a joint feature matrix;

[0038] Step S25: Perform high-dimensional feature fusion according to the joint feature matrix to obtain a sample feature embedding vector.

[0039] The present invention significantly improves the data quality by acquiring and preprocessing the multimodal production data of Niushe Kaiqiao Powder, laying a reliable foundation for subsequent analysis. In the principal component analysis of global sample source characteristics, by extracting the principal component data, data redundancy is effectively reduced, while retaining the key information of the sample source characteristics, and combining the dimension reduction of nonlinear distribution characteristics, the analysis efficiency and accuracy are further improved, ensuring that the sample source data more accurately reflects the multimodal production characteristics. The feature alignment step of the sample source data ensures the uniformity and consistency of the data by eliminating the heterogeneity between sources, and the unified feature space constructed on this basis enables the sample source characteristics to be systematically integrated. The chemical component feature matrix and the sample source feature space are aligned in dimension and scale, achieving seamless docking and information coupling between data, and effectively breaking through the bottleneck of multimodal data being difficult to merge in traditional methods. By constructing a joint feature matrix and performing high-dimensional feature fusion, the generated sample feature embedding vector not only completely retains the key information of multimodal production data and chemical component characteristics, but also realizes the deep integration of multidimensional features, significantly enhancing the expression ability and analysis potential of data in high-dimensional space, and providing precise support for the comprehensive quality evaluation and complex feature analysis of Niushe Kaiqiao Powder.

[0040] Optionally, step S25 is specifically:

[0041] Step S251: performing feature importance evaluation on the joint feature matrix to obtain feature importance data;

[0042] Step S252: performing feature weighting processing on the joint feature matrix according to the feature importance data, thereby obtaining a weighted joint feature matrix;

[0043] Step S253: performing feature concatenation on the weighted joint feature matrix to obtain an initial high-dimensional feature vector;

[0044] Step S254: performing nonlinear feature combination processing on the initial high-dimensional feature vector to obtain a preliminary sample feature embedding vector;

[0045] Step S255: Perform embedding vector regularization overfitting optimization on the preliminary sample feature embedding vector to obtain a sample feature embedding vector.

[0046] The present invention can accurately identify the key features that have a significant impact on Niushe Kaiqiao Powder in quality evaluation by evaluating the feature importance of the joint feature matrix, and provide a clear direction for subsequent data processing. Based on the feature importance data, the joint feature matrix is ​​subjected to feature weighting processing, which effectively amplifies the weight of important features, suppresses the interference of minor features on the overall analysis, and improves the distinguishing ability of feature expression. The feature splicing performed on this basis realizes the efficient integration between multimodal features, ensuring that the initial high-dimensional feature vector comprehensively covers the multi-level characteristics of the sample. Through nonlinear feature combination processing, the complex nonlinear relationship between features is fully excavated, and the depth characterization ability of the initial sample feature embedding vector for sample characteristics is improved. Finally, the embedding vector regularization overfitting optimization step effectively reduces the overfitting risk that may occur in the model when processing high-dimensional data, and enhances the robustness and generalization ability of the embedding vector. The overall processing flow makes full use of the synergistic effect of chemical composition characteristics and multimodal production data, and provides strong technical support for the accurate characteristic description and efficient quality control of Niushe Kaiqiao Powder samples.

[0047] Optionally, step S251 is specifically:

[0048] Perform feature screening based on the joint feature matrix to obtain the joint feature matrix to be analyzed;

[0049] The chemical component proportion of Niushe Kaiqiao Powder is calculated based on the component distribution data of the spectrum characteristic pattern, so as to obtain the chemical component proportion data of Niushe Kaiqiao Powder;

[0050] According to the chemical component proportion data of Niushe Kaiqiao Powder, the chemical components of the joint feature matrix to be analyzed are mapped to obtain the feature weight factor;

[0051] Calculate the standard importance weight of the joint feature matrix to be analyzed, so as to obtain the feature standard importance weight;

[0052] Based on the feature weight factor and the feature standard importance weight, weighted fusion is performed to obtain the modified comprehensive feature importance value;

[0053] The feature importance is assigned to the joint feature matrix according to the modified comprehensive feature importance value to obtain the feature importance data.

[0054] The present invention can effectively eliminate redundant features, optimize the scale of feature sets, and ensure that the joint feature matrix to be analyzed is more focused on key information related to the quality of Niushe Kaiqiao Powder, thereby improving the efficiency and accuracy of subsequent analysis. The chemical component proportion calculation of the spectrum feature pattern component distribution data can clarify the relative proportion of each chemical component in Niushe Kaiqiao Powder, and provide a scientific basis for constructing the mapping relationship between chemical components and feature weight factors. The chemical component proportion data is used to map the chemical components of the joint feature matrix to be analyzed, and the weight consideration of chemical components in feature analysis is successfully introduced, further strengthening the influence of chemical components in feature expression. In the process of calculating the standard importance weight of the joint feature matrix, the basic weight of the feature in the entire analysis process is comprehensively considered to ensure the uniformity and standardization of feature evaluation. Subsequently, based on the weighted fusion of feature weight factors and standard importance weights, multi-dimensional importance information is effectively integrated, and a more accurate and comprehensive modified feature importance value is obtained. This process significantly improves the reliability of importance calculation and lays a solid foundation for the importance assignment of features. -Based on the modified comprehensive feature importance value, the joint feature matrix is ​​assigned a value, which further clarifies the contribution of each feature in the overall analysis, and provides key data support for the accuracy and robustness of the subsequent optimization model. Through this series of steps, not only the importance assessment of the features is deepened, but also the comprehensiveness and scientificity of the analysis of the characteristics of Niushe Kaiqiao Powder are improved.

[0055] Optionally, step S3 specifically includes:

[0056] Step S31: performing sample similarity calculation according to the sample feature embedding vector to obtain a sample similarity matrix, and constructing a sample relationship network based on the sample similarity matrix to obtain a sample relationship graph;

[0057] Step S32: performing sample component co-occurrence frequency analysis on the sample relationship graph to obtain sample component co-occurrence data, and constructing a component co-occurrence matrix based on the sample component contribution data;

[0058] Step S33: Calculating the component Pearson correlation coefficient of the component co-occurrence matrix to obtain co-occurrence component correlation data, and constructing a component relationship structure based on the co-occurrence component correlation data to obtain the component relationship structure data of Niushe Kaiqiao Powder;

[0059] Step S34: performing dynamic network modeling based on the component relationship structure data of Niushe Kaiqiao Powder to obtain a chemical component dynamic network, and performing chemical component synergy evaluation on the chemical component dynamic network to obtain chemical component synergy data;

[0060] Step S35: Calculating the strength of the relationship between components based on the relationship structure data of the Niushe Kaiqiao Powder components, thereby obtaining the strength data of the relationship between chemical components;

[0061] Step S36: construct a synergistic network of the ingredients of the musk deer awakening powder according to the chemical ingredient synergistic effect data and the chemical ingredient relationship strength data.

[0062] The present invention can intuitively reveal the intrinsic correlation between different samples through the similarity calculation of sample feature embedding vectors and the construction of sample similarity matrix, lay the foundation for the establishment of sample relationship network, and thus present the complex connection between samples in a visual way. The generation of sample relationship graph provides an intuitive graphical expression for further analyzing the interactive relationship between sample components. By analyzing the co-occurrence frequency of sample components, the co-occurrence law of different components between samples can be identified, and the potential synergy and distribution characteristics between components can be clarified. The construction of component co-occurrence matrix helps to quantify the co-occurrence pattern between components and provides data support for subsequent correlation analysis. By calculating the Pearson correlation coefficient of the components, the linear relationship between the components can be accurately evaluated, thereby providing a scientific basis for the construction of the component relationship structure. The extraction of the component relationship structure data of Niushe Kaiqiao Powder helps to intuitively present the correlation strength and network relationship between chemical components, and lays the foundation for dynamic network modeling. Based on dynamic network modeling, the network change characteristics of chemical components under different conditions can be dynamically analyzed, and the expression form and degree of synergistic effect between components can be revealed through the evaluation of chemical component synergy, and the understanding of chemical component synergy can be further deepened. The calculation of the strength of the component relationship can quantify the force of each chemical component in the relationship network, providing key data support for the construction of the synergistic network of the components of Niushe Kaiqiao Powder. Combining the synergistic effect data with the relationship strength data, the generated component synergistic network comprehensively displays the complex synergistic relationship and action mode of the chemical components in Niushe Kaiqiao Powder, and deepens the understanding of its compatibility mechanism. This process provides scientific guidance for the overall optimization and functional analysis of the chemical components of Niushe Kaiqiao Powder.

[0063] Optionally, step S4 specifically includes:

[0064] Based on the synergistic network of Niushe Kaiqiao Powder ingredients, the relative importance of ingredients and the synergistic effect characteristics of ingredients were extracted, so as to obtain the synergistic effect data of Niushe Kaiqiao Powder ingredients and the relative importance data of Niushe Kaiqiao Powder ingredients;

[0065] Peak intensity mapping is performed based on the synergistic effect data of the ingredients of Niushe Kaiqiao Powder and the relative importance data of the ingredients of Niushe Kaiqiao Powder, thereby obtaining the vertical axis of the fingerprint spectrum;

[0066] Sort the chemical component characteristics according to the chemical component annotation matrix to obtain the molecular characteristic axis;

[0067] Construct a global fingerprint spectrum based on the vertical axis of the fingerprint spectrum and the molecular feature axis;

[0068] Screening the global fingerprint spectrum for potential abnormal points, thereby obtaining abnormal peak data of the global fingerprint spectrum;

[0069] The abnormal peak data of the global fingerprint spectrum is corrected to obtain an optimized fingerprint spectrum.

[0070] The present invention can effectively identify the key role and synergistic performance of chemical components in the overall compatibility by extracting the relative importance and synergistic features of components based on the synergistic network of Niushe Kaiqiao Powder, help reveal the complex interaction mechanism between the chemical components of Niushe Kaiqiao Powder, and provide a scientific basis for subsequent analysis. The acquisition of component synergistic data and relative importance data provides accurate vertical axis information for spectrum construction, and the vertical axis of the fingerprint spectrum generated by peak intensity mapping can fully reflect the relative contribution and variation law of chemical components in different samples. The characteristic sorting of the chemical component annotation matrix lays the foundation for the establishment of the molecular characteristic axis, making the presentation of molecular characteristics in the spectrum more systematic and intuitive, and providing high-resolution characteristic data support for the accurate construction of the fingerprint spectrum. Through the fusion of the vertical axis and the molecular characteristic axis, the global fingerprint spectrum constructed can systematically reflect the chemical component distribution and characteristic law of Niushe Kaiqiao Powder, and provide a visualization tool for product quality control and the overall characteristic research of its chemical components. The screening of potential abnormal points in the global fingerprint spectrum helps to quickly locate abnormal chemical components or molecular characteristics and ensure the comprehensiveness and accuracy of the data. By correcting the abnormal peak data, the overall quality and accuracy of the spectrum are further optimized, possible errors or noise interference are eliminated, and the optimized fingerprint spectrum is more stable and reliable. This process not only improves the scientificity and accuracy of the fingerprint spectrum of Niushe Kaiqiao Powder, but also provides more systematic technical support for its quality control, component analysis and efficacy evaluation.

[0071] Optionally, the potential abnormal point screening in step S4 is specifically as follows:

[0072] Performing baseline correction on the global fingerprint spectrum to obtain the global fingerprint spectrum to be analyzed;

[0073] According to the preset division ratio, the global fingerprint spectrum to be analyzed is divided into regions to obtain a local fingerprint spectrum;

[0074] Performing local peak intensity distribution statistics on the local fingerprint spectrum to obtain local peak intensity distribution data, and performing significant deviation peak point identification on the local peak intensity distribution data to obtain abnormal local significant deviation peak point data;

[0075] The intensity value variation of the points adjacent to the peak point is calculated for the local fingerprint spectrum according to the preset window range, so as to obtain the intensity value variation data of the points adjacent to the peak point, and the intensity mutation variation data of the intensity value variation data of the points adjacent to the peak point is statistically analyzed, so as to obtain the abnormal intensity mutation peak point data;

[0076] The peak point intersection operation is performed on the abnormal local significant deviation peak point data and the abnormal intensity mutation peak point data to obtain the global fingerprint spectrum abnormal peak data.

[0077] The present invention can effectively remove the baseline drift or noise interference in the data by performing baseline correction on the global fingerprint spectrum, ensure the accuracy and stability of the spectrum, and provide clearer spectrum data for subsequent analysis. The global fingerprint spectrum to be analyzed is divided into regions according to a preset ratio, and different spectrum regions can be focused on for more detailed analysis, so as to obtain local fingerprint spectra, thereby improving the local resolution of data analysis and revealing the potential component differences between regions. The local peak intensity distribution statistics of the local fingerprint spectrum can help identify the key characteristic peaks in the spectrum and provide a basis for further analysis. The significant deviation peak point identification enables abnormal data (such as noise, error peaks, etc. in the spectrum) to be efficiently identified, avoiding the interference of erroneous information on the overall analysis results and ensuring the high quality of the data. Next, by calculating the change in the intensity value of the peak point neighboring points, the intensity fluctuation near the peak in the spectrum can be deeply analyzed, and the intensity mutation peak point can be further identified, which provides an effective means for identifying potential abnormal peaks or measurement errors. The intersection operation of these abnormal peak points, combined with the significant deviation peak point data and the intensity mutation peak point data, can further accurately locate the abnormal peak data in the global fingerprint spectrum. This process can effectively improve the accuracy and reliability of the global fingerprint spectrum, and provide strong data support for subsequent optimization, correction and precise analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0079] Figure 1 It is a schematic diagram of the steps of the method for detecting the fingerprint of the Chinese medicine preparation of the present invention;

[0080] Figure 2 Detailed step flow diagram of step S1 in the present invention;

[0081] Figure 3 Detailed step flow diagram of step S2 in the present invention;

[0082] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0083] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0084] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0085] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0086] To achieve this, please refer to Figures 1 to 3 The present invention provides a method for detecting a Chinese medicine fingerprint of a Chinese medicine preparation, the method comprising the following steps:

[0087] Step S1: obtaining high-dimensional spectrum data of Niu Musk Kaiqiao Powder, and performing chemical component spectrum decomposition on the high-dimensional spectrum data of Niu Musk Kaiqiao Powder, so as to obtain a chemical component spectrum of Niu Musk Kaiqiao Powder; performing qualitative characteristic analysis of the chemical components of Niu Musk Kaiqiao Powder according to the chemical component spectrum of Niu Musk Kaiqiao Powder, so as to obtain a chemical component characteristic matrix of Niu Musk Kaiqiao Powder;

[0088] In this embodiment, the high-dimensional spectrum data of Niushe Kaiqiao Powder is first obtained, and liquid chromatography-mass spectrometry (LC-MS) or gas chromatography-mass spectrometry (GC-MS) technology is used for analysis to generate original high-dimensional spectrum data. Then, the chemical component spectrum is decomposed by methods such as principal component analysis (PCA) or non-negative matrix decomposition (NMF) to extract the main component spectra of Niushe Kaiqiao Powder. Through spectrum decomposition, the chemical component spectrum of Niushe Kaiqiao Powder containing information of each chemical component is obtained. Then, based on a known chemical component database, a qualitative analysis of the components is performed to determine the chemical substances corresponding to each peak in the spectrum. Finally, through a feature extraction algorithm, such as a feature extraction method based on wavelet transform or Fourier transform, the chemical component feature matrix of Niushe Kaiqiao Powder is obtained, which contains the qualitative characteristics of each component presented in different spectra, and can effectively support subsequent analysis.

[0089] Step S2: obtaining the multimodal production data of Niu Musk Kaiqiao Powder, and unifying the Niu Musk Kaiqiao Powder sample source feature space of the multimodal production data of Niu Musk Kaiqiao Powder, thereby obtaining the Niu Musk Kaiqiao Powder sample source feature space; performing high-dimensional feature fusion on the Niu Musk Kaiqiao Powder chemical component feature matrix and the Niu Musk Kaiqiao Powder sample source feature space, thereby obtaining a sample feature embedding vector;

[0090] In this embodiment, multimodal data (such as sample source information, production condition data, component concentration, etc.) from different production batches are collected, and the data is preprocessed using a standardization method (such as Z-score standardization or minimum-maximum standardization). By using dimensionality reduction methods such as principal component analysis (PCA) or t-SNE, sample data from different sources are mapped to a unified feature space for subsequent analysis. In a specific implementation, factors such as sample source information, production batch number, and production equipment of different production batches can be converted into feature variables and unified into the same space. Afterwards, the chemical composition feature matrix is ​​fused with the sample source feature space for high-dimensional features, and the two feature sets are fused using an autoencoder (Autoencoder) or multi-layer perceptron (MLP) method based on deep learning to generate a sample feature embedding vector. This vector is a low-dimensional representation that can effectively integrate the composition information and source information of the sample, and provide data support for subsequent sample relationship analysis.

[0091] Step S3: performing sample relationship structure analysis according to the sample feature embedding vector to obtain the relationship structure data of the ingredients of Niushe Kaiqiao Powder, and constructing a chemical component dynamic network based on the relationship structure data of the ingredients of Niushe Kaiqiao Powder to obtain the synergistic network of the ingredients of Niushe Kaiqiao Powder;

[0092] In the present embodiment, sample relationship structure analysis is performed based on sample feature embedding vectors, and cluster analysis is performed using clustering algorithms such as K-means, hierarchical clustering or density-based DBSCAN algorithms to reveal the similarities and relationships between samples. By analyzing the clustering results, the component relationship structure data between samples can be obtained. Then, according to the component relationship structure data, a dynamic network of chemical components of Niushe Kaiqiao Powder is constructed, and a network analysis method, such as a weighted analysis of nodes and edges, is used to evaluate the mutual relationship and intensity between different components. This network can help reveal the synergy and interdependence between the components in Niushe Kaiqiao Powder, and identify key components and their importance in the overall formula.

[0093] Step S4: generating a global fingerprint spectrum based on the collaborative network of Niushe Kaiqiao Powder components, and screening the global fingerprint spectrum for potential abnormal points, thereby obtaining abnormal peak data of the global fingerprint spectrum; performing peak correction on the abnormal peak data of the global fingerprint spectrum, thereby obtaining an optimized fingerprint spectrum;

[0094] In this embodiment, a global fingerprint spectrum is generated based on a chemical component collaborative network, and the spectrum data of the chemical components and the collaborative network are used to perform weighted synthesis of the spectrum to generate a global fingerprint spectrum that can fully reflect the characteristics of the sample. This process can use the weighted average method or the support vector regression (SVR) method to fuse the spectrum data. Next, potential abnormal points are screened, and abnormal peaks that may exist in the global fingerprint spectrum are identified using anomaly detection algorithms such as Isolation Forest or the local outlier factor (LOF) algorithm. The abnormal peak data is corrected, and the spectrum peak is corrected using smoothing filtering or a fitting method based on the least squares method to remove abnormal data caused by noise or experimental errors, and finally an optimized fingerprint spectrum is obtained. The optimized fingerprint spectrum can not only more realistically reflect the chemical composition of Niushe Kaiqiao Powder, but also provide accurate spectrum information for quality evaluation.

[0095] Step S5: construct and verify the characteristic model according to the optimized fingerprint spectrum, so as to obtain the quality evaluation model of Niushe Kaiqiao Powder.

[0096] In this embodiment, the feature model is constructed and verified according to the optimized fingerprint spectrum. First, a machine learning method such as support vector machine (SVM), random forest (Random Forest) or gradient boosting tree (GBDT) is used to construct a quality evaluation model based on the characteristics of the optimized fingerprint spectrum. The input of the model is the optimized fingerprint spectrum data, and the output is the quality evaluation result of Niushe Kaiqiao Powder, which specifically includes indicators such as ingredient content and purity. In the model verification stage, the model is evaluated using methods such as cross-validation or leave-one-out (LOO) to ensure its generalization ability. Through this method, reliable data support and scientific basis can be provided for the quality control and evaluation of Niushe Kaiqiao Powder.

[0097] Optionally, step S1 specifically includes:

[0098] Step S11: obtaining high-dimensional spectrum data of Niushe Kaiqiao Powder, and performing spectrum data preprocessing on the high-dimensional spectrum data of Niushe Kaiqiao Powder, so as to obtain high-dimensional spectrum data of Niushe Kaiqiao Powder to be analyzed;

[0099] In this embodiment, high-dimensional spectral data of Niushe Kaiqiao Powder are obtained by liquid chromatography-mass spectrometry (LC-MS) technology. These data include multiple analysis dimensions, such as time, mass to charge ratio (m / z), and ion intensity. Subsequently, the original spectral data is preprocessed, and denoising methods such as wavelet denoising or Gaussian smoothing are used to remove background noise and interference signals. In addition, baseline correction techniques such as SNIP (Signal Noise Iterative Processing) algorithm are used to adjust the baseline of the spectrum to ensure accurate extraction of the peak. The processed data is the high-dimensional spectral data of Niushe Kaiqiao Powder to be analyzed, which can truly reflect the complex chemical composition characteristics of the sample and provide a basis for subsequent analysis.

[0100] Step S12: obtaining chemical component characteristic data based on a preset chemical library, and integrating the chemical component qualitative characteristics according to the chemical component characteristic data, thereby obtaining a chemical component characteristic matrix;

[0101] In this embodiment, the chemical component characteristic data related to Niushe Kaiqiao Powder is obtained by docking with the chemical component database. For example, the molecular structure information and molecular mass data provided by the chemical molecule library (such as PubChem, ChemSpider) are used. In combination with the preset chemical library, the key characteristics of each chemical component, such as polarity, solubility, molecular weight and molecular formula, can be extracted. Then, using data integration techniques such as weighted synthesis, principal component analysis (PCA), etc., these chemical characteristic data are combined together to form a chemical component feature matrix. This matrix contains the qualitative characteristics of each component in Niushe Kaiqiao Powder, which can help analyze the relationship between the components and their importance in the formula.

[0102] Step S13: performing chemical component spectrum decomposition on the high-dimensional spectrum data of Niushe Kaiqiao Powder to be analyzed, thereby obtaining a chemical component spectrum, and performing chemical component relative intensity calculation on the chemical component spectrum, thereby obtaining chemical component intensity data;

[0103] In the present embodiment, based on the acquired high-dimensional spectrogram data, chemical component spectrogram decomposition technology, such as principal component analysis (PCA) or non-negative matrix decomposition (NMF), is used to decompose the complex spectrogram of Niushe Kaiqiao Powder into spectrograms of different chemical components. These chemical component spectrograms can respectively reflect the peak characteristics of each chemical component in the sample. Subsequently, for each chemical component spectrogram, its relative intensity is calculated. The calculation of relative intensity can be obtained by dividing the intensity value of each peak by the total spectrum intensity, thereby obtaining chemical component intensity data. These data reflect the relative content of each chemical component in the sample, and provide quantitative information for subsequent analysis.

[0104] Step S14: performing chemical component-specific spectrum decomposition on the chemical component spectrum according to the chemical component characteristic matrix and the chemical component intensity data, thereby obtaining the chemical component spectrum of Niushe Kaiqiao Powder;

[0105] In this embodiment, specific decomposition of the chemical component spectrum is performed based on the obtained chemical component characteristic matrix and chemical component intensity data. The spectrum is decomposed into specific spectra related to specific chemical components through algorithms such as orthogonal least squares (OLS) or weighted least squares (WLS). This process helps to extract independent spectra of each chemical component from complex sample spectra and eliminate interference from other components. Through spectrum decomposition technology, the chemical spectrum of each component of Niushe Kaiqiao Powder can be accurately obtained, providing support for further analysis of its structure and function.

[0106] Step S15: Perform qualitative characteristic analysis of the chemical components of Niushe Kaiqiao Powder according to the chemical component spectrum of Niushe Kaiqiao Powder, so as to obtain the chemical component characteristic matrix of Niushe Kaiqiao Powder.

[0107] In this embodiment, qualitative feature analysis is performed based on the chemical component spectrum of Niushe Kaiqiao Powder. This analysis can parse the main peaks in the spectrum by Fourier transform, peak area analysis and other methods to identify the characteristic peaks of each chemical component, and match them by comparing the known spectra in the chemical component library. For example, a spectrum comparison algorithm (such as dynamic time warping DTW) is used to compare known standard spectra to determine the components in the sample. Based on these analysis results, the qualitative feature matrix of the chemical components of Niushe Kaiqiao Powder is further obtained to help identify the various active ingredients in the sample and provide data support for subsequent quality evaluation.

[0108] Optionally, step S14 is specifically:

[0109] Step S141: performing Pearson correlation coefficient feature matching on the chemical component feature matrix and the chemical component intensity data, thereby obtaining a preliminary matching matrix;

[0110] In this embodiment, the Pearson Correlation Coefficient (PCC) is used for feature matching based on the chemical component feature matrix (including the key characteristics of each chemical component) and the chemical component intensity data (such as the intensity of each spectral peak obtained by spectral analysis). The Pearson Correlation Coefficient determines whether there is a correlation between two variables by measuring the linear relationship between them. Specifically, a window size can be selected to calculate each pair of chemical component features and their intensity data to obtain a preliminary matching matrix. This preliminary matching matrix can help identify highly correlated chemical component pairs and provide a basis for subsequent factor separation and spectral feature extraction.

[0111] Step S142: performing sparse coding specific factor separation based on the preliminary matching matrix, thereby obtaining a specific factor matrix;

[0112] In this embodiment, after obtaining the preliminary matching matrix, sparse coding technology is used to separate specific factors. The goal of sparse coding is to represent a complex signal (such as spectrogram data) as a linear combination of a small number of basis functions, and these basis functions have obvious physical meanings. By introducing sparse constraints, factors specifically related to each chemical component can be effectively separated. By using techniques such as L1 regularization, factors with obvious characteristics and strong representativeness are extracted from the preliminary matching matrix to form a specific factor matrix. This step ensures that the core features of each component are extracted from the chemical component spectrum.

[0113] Step S143: extracting the spectrogram features of the specific factor matrix to obtain specific factor spectrogram feature data, and performing intensity normalization on the specific factor spectrogram feature data and the chemical component intensity data to obtain intensity normalized spectrogram feature data;

[0114] In the present embodiment, the obtained specific factor matrix is ​​subjected to the spectral feature extraction technology to extract its spectral feature data. The specific factors can be converted into feature data that can be used for spectral analysis by using techniques such as waveform feature extraction, wavelet transform, and Fourier transform. Then, the specific factor spectral feature data and chemical component intensity data are subjected to intensity normalization. Normalization is to eliminate the intensity deviation caused by instrument differences or different sample concentrations by linear transformation of the intensity of each spectra, so as to ensure that the comparison between different samples is more accurate. The intensity normalized spectral feature data finally obtained is convenient for further analysis and modeling.

[0115] Step S144: Calculating the relative contribution of components according to the intensity normalized spectrum feature data, thereby obtaining relative contribution data of components, and constructing a relative contribution matrix based on the relative contribution data of components;

[0116] In the present embodiment, after obtaining the intensity-normalized spectrum characteristic data, the relative contribution of each component is calculated. The contribution can be determined by calculating the ratio of the relative intensity of each chemical component in the sample to the total intensity. In the specific implementation, the total intensity of all peak values ​​of each chemical component in the sample is first determined, and then the proportion of a single peak value in the total intensity is calculated. In this way, the contribution of each component to the overall spectrum can be calculated. The chemical component contributions of all samples are summarized and constructed into a relative contribution matrix. The rows of the matrix represent different samples, the columns represent different chemical components, and each element represents the relative contribution of the chemical component in a specific sample. Each element in the matrix can be filled by calculating the contribution of the chemical component in the sample, thereby providing quantitative data support for further optimizing and analyzing the impact of each component.

[0117] Step S145: reconstructing the specific spectrum based on the relative contribution matrix and the specific factor matrix, thereby obtaining the specific spectrum of the chemical components of Niushe Kaiqiao Powder;

[0118] In the present embodiment, the specificity spectrum is reconstructed based on the relative contribution matrix and specificity factor matrix obtained before. Through the specificity spectrum reconstruction technology, the relative contribution between the components can be combined with their specificity factors to obtain a spectrum reflecting the characteristics of the chemical components and their contributions. The weighted reconstruction technology is adopted, and the factors in the specificity factor matrix and the contribution in the relative contribution matrix are weightedly combined to reconstruct the chemical component spectrum of Niushe Kaiqiao Powder. In the reconstruction process, the synergistic effect between the different chemical components in the spectrum is taken into account to ensure that the obtained spectrum reflects the chemical component characteristics in the sample as realistically as possible.

[0119] Step S146: According to the chemical component specific spectrum of Niushe Kaiqiao Powder, the original chemical component spectrum is specifically tested and optimized to obtain the chemical component spectrum of Niushe Kaiqiao Powder.

[0120] In the present embodiment, after the specificity spectrum reconstruction is completed, the original chemical composition spectrum is subjected to specificity inspection and optimization. This inspection includes comparing with the standard spectrum in the chemical composition characteristic library, and its specificity is determined by comparing the matching degree, peak shape, peak position and other parameters of the spectrum peak. In order to further improve the accuracy of the spectrum, data optimization techniques, such as nonlinear least squares method, genetic algorithm, etc., can also be used to adjust the spectrum peak in the spectrum to correct potential deviations and noise. The optimized spectrum can more accurately reflect the chemical composition of Niushe Kaiqiao Powder, and provide a reliable basis for subsequent quality control, component analysis and formulation optimization.

[0121] Optionally, step S15 is specifically:

[0122] Step S151: extracting peak features from the chemical component spectrum of Niushe Kaiqiao Powder to obtain a peak feature matrix;

[0123] In this embodiment, a spectrum peak detection algorithm (such as a derivative-based peak detection method) is used to identify all valid peaks in the spectrum. An effective peak refers to a peak that can clearly reflect a chemical component under the interference of a noise signal. By setting a threshold, noise peaks below a certain signal intensity are removed. For each identified peak, its related parameters are extracted, including peak position (corresponding to the characteristic wavelength of the chemical component), peak height (reflecting the concentration information of the component), peak width (reflecting molecular weight information), etc. Gaussian fitting and Lorentz fitting can be used to accurately extract the shape parameters of the peak. The parameters of each peak (such as peak position, peak height, peak width, etc.) are organized in a matrix form to construct a peak feature matrix. Each row of the matrix represents a sample, and the column represents the characteristics of each peak.

[0124] Step S152: performing correlation matching according to the chemical component feature matrix and the peak feature matrix, thereby obtaining a chemical component annotation matrix;

[0125] In this embodiment, by calculating the correlation between each element in the chemical component feature matrix and the peak feature matrix, a statistical method such as the Pearson correlation coefficient is used to calculate the similarity between the characteristics of each chemical component and the characteristics of the spectrum peak. The characteristics of each chemical component (such as chemical properties, molecular structure) are matched one-to-one with the characteristics of each peak in the spectrum (such as peak position, intensity, etc.). For peaks and chemical components with high correlation, they are marked as matched to form a preliminary chemical component annotation matrix. Through the correlation calculation and matching process, a chemical component annotation matrix is ​​finally constructed, in which each row represents a chemical component, each column corresponds to a spectrum peak, and the matrix elements represent the matching relationship between the chemical component and the spectrum peak.

[0126] Step S153: performing spectrum characteristic pattern recognition on the chemical component annotation matrix to obtain spectrum characteristic pattern data; performing characteristic pattern component distribution analysis based on the spectrum characteristic pattern data to obtain spectrum characteristic pattern component distribution data;

[0127] In this embodiment, a cluster analysis method (such as K-means, hierarchical clustering, etc.) is used to perform pattern recognition on the data in the chemical component annotation matrix. In this process, similar spectral patterns are classified into the same category by analyzing the similarity between the chemical components and the spectral peaks in the annotation matrix. On the basis of pattern recognition, the distribution of chemical components in each spectral feature pattern is further analyzed. For example, according to the component content of each spectral feature pattern, their distribution in the sample can be analyzed to determine which components dominate in different spectral patterns. The results of the spectral feature pattern and the component distribution are organized into a data format, and the spectral feature pattern data and the spectral feature pattern component distribution data are output as input for the subsequent integration of the qualitative characteristics of the chemical components.

[0128] Step S154: integrating the chemical component qualitative characteristics based on the spectrum characteristic pattern data and the spectrum characteristic pattern component distribution data, thereby obtaining the chemical component qualitative characteristic data;

[0129] In this embodiment, each chemical component is qualitatively analyzed based on the spectral characteristic pattern data and the spectral characteristic pattern component distribution data, and the behavioral characteristics of each chemical component in the sample are integrated. For example, it is analyzed which components appear more frequently in certain specific spectral patterns, and which component changes are highly correlated with certain specific spectral characteristic patterns. Combining the component distribution data and the spectral characteristic pattern, a qualitative feature description of each chemical component is constructed. These qualitative features may include information such as the presence or absence of the component, the intensity change in a specific spectral pattern, etc. The qualitative characteristics of these chemical components are organized into chemical component qualitative feature data to provide effective information for subsequent model construction.

[0130] Step S155: construct the chemical component characteristic matrix of Niushe Kaiqiao Powder based on the peak characteristic matrix, the chemical component annotation matrix and the chemical component qualitative characteristic data.

[0131] In this embodiment, the obtained peak feature matrix, chemical component annotation matrix and chemical component qualitative feature data are integrated to form a final chemical component feature matrix. Each row of the matrix represents a sample, and each column represents a chemical component or a spectral feature. In the fusion process, the chemical component annotation information is integrated with the peak feature data and the qualitative feature data using methods such as weighted average method or multiple regression to ensure that the various feature data of each chemical component are fully reflected. Output a matrix containing all chemical component features, and each element in the matrix represents the feature of the chemical component in the sample. This matrix can be used to establish a subsequent quality evaluation model.

[0132] Optionally, step S2 specifically includes:

[0133] Step S21: acquiring the multimodal production data of Niushe Kaiqiao Powder, and performing data preprocessing on the multimodal production data of Niushe Kaiqiao Powder, so as to obtain the multimodal production data of Niushe Kaiqiao Powder to be analyzed;

[0134] In this embodiment, multimodal production data of Niushe Kaiqiao Powder is collected from a variety of data sources (such as laboratory testing equipment, production line automation system, environmental monitoring system, etc.). These data include different types of production data such as temperature, humidity, material ratio, processing time, equipment operation status, and output. The collected multimodal production data is preprocessed, including removing missing values, outlier processing, normalization, etc. In particular, for data of different modes, it is necessary to convert the data into a unified format, such as uniformly processing time series data and tabular data so that they can be compatible with each other. Data cleaning techniques are used, such as removing invalid data, interpolating missing values, etc., to ensure data quality. Through data standardization or normalization processing (such as Z-score standardization or Min-Max normalization), the dimensional differences between different modal data are eliminated to enhance their comparability.

[0135] Step S22: performing a principal component analysis of global sample source characteristics on the multimodal production data of Niushe Kaiqiao Powder to be analyzed, thereby obtaining principal component multimodal production data, and performing nonlinear distribution feature dimensionality reduction on the principal component multimodal production data, thereby obtaining Niushe Kaiqiao Powder sample source data;

[0136] In this embodiment, principal component analysis (PCA) is performed on the pre-processed multimodal production data of Niushe Kaiqiao Powder. PCA projects the original data onto a set of new coordinate axes through linear transformation, which are the directions with the largest variance in the data, namely the principal components. Through principal component analysis, the dimension of the feature space is reduced, and the information that best represents the data change is retained. In order to further reduce the dimension and reveal the nonlinear characteristics of the data, nonlinear dimensionality reduction algorithms such as t-SNE (t-distributed random neighbor embedding) or UMAP (uniform manifold approximation and projection) are used. Unlike PCA, these methods can capture complex nonlinear structures in the data, thereby revealing potential patterns and associations in the data. Through PCA and nonlinear dimensionality reduction processing, the principal component multimodal production data is finally obtained, and the key characteristics of the sample source can be displayed as the basic data for the next step of analysis.

[0137] Step S23: performing feature alignment on the Niushe Kaiqiao Powder sample source data to obtain aligned sample source data, and performing unified feature space mapping based on the aligned sample source data to obtain the Niushe Kaiqiao Powder sample source feature space;

[0138] In this embodiment, different features of the sample source are aligned. Features from different data sources or different modalities are mapped to a unified standard space. For example, if there is data from different devices, the timestamps of the device data need to be aligned to ensure that data from different modalities can be compared on the same timeline. Use mapping techniques (such as linear mapping or Gaussian process regression) to convert the aligned data into a unified feature space. The samples are mapped so that all features have the same scale and reference frame, thereby providing a comparable feature space for subsequent analysis. After feature alignment and feature space mapping, the feature space of the source of the Niushe Kaiqiao Powder samples is obtained, and the eigenvalues ​​corresponding to each sample data point in the space are at the same scale, which is convenient for further analysis and modeling.

[0139] Step S24: aligning the dimension and scale of the chemical component feature matrix of Niushe Kaiqiao Powder and the feature space of the source of Niushe Kaiqiao Powder samples, so as to obtain a joint feature matrix;

[0140] In this embodiment, the chemical component feature matrix of Niushe Kaiqiao Powder and the sample source feature space come from different data sources and have different dimensions. In order to ensure the compatibility of the data, dimensional alignment is required. For example, the dimensions of the two can be unified by selecting common features or using dimensionality reduction methods (such as PCA) to ensure that their number of dimensions is consistent. After dimensional alignment, it is also necessary to ensure that the scale of the data is consistent. This usually includes standardizing the data in the two matrices (such as Z-score standardization) so that different feature data are compared at the same scale to eliminate the impact of dimensional differences on the analysis results. Through dimensional and scale alignment, a joint feature matrix is ​​finally obtained, which contains the chemical component characteristics and sample source characteristics of Niushe Kaiqiao Powder, so that the data from the two sources can be seamlessly combined to form a unified feature representation.

[0141] Step S25: Perform high-dimensional feature fusion according to the joint feature matrix to obtain a sample feature embedding vector.

[0142] In this embodiment, each feature in the joint feature matrix is ​​fused. Common fusion methods include weighted average method, principal component analysis (PCA), automatic encoders in deep learning, etc. Through these methods, high-dimensional feature vectors can be compressed into low-dimensional embedding vectors, thereby effectively reducing the computational complexity and retaining the main feature information. The fused features are processed using machine learning methods (such as support vector machines, K-means, etc.), and the similarities and differences between samples are further explored to finally obtain the sample feature embedding vector. The embedding vector will contain the comprehensive feature information of the sample and can be used for classification, regression or other analysis tasks. The obtained sample feature embedding vector can be used as input data for subsequent modeling, and can be used for tasks such as quality assessment and component analysis of Niushe Kaiqiao Powder.

[0143] Optionally, step S25 is specifically:

[0144] Step S251: performing feature importance evaluation on the joint feature matrix to obtain feature importance data;

[0145] In this embodiment, a tree-based model (such as a random forest or a gradient boosting tree) is used to evaluate the importance of each feature in the joint feature matrix. This method can calculate the importance of each feature based on the contribution of the feature to the prediction accuracy of the model. The importance of the feature is usually obtained by calculating the degree of influence of each feature on the prediction result and ranking them. The contribution of the feature to the prediction result is quantified by calculating the change of the "information gain" or "Gini index" of each feature during the model training process. For example, if the introduction of a feature greatly improves the classification or regression accuracy of the model, it means that the feature has a high importance. The feature importance evaluation results are presented in the form of data, such as a ranking table, listing the features and their corresponding importance scores. The data will serve as the basis for subsequent weighted processing.

[0146] Step S252: performing feature weighting processing on the joint feature matrix according to the feature importance data, thereby obtaining a weighted joint feature matrix;

[0147] In this embodiment, each feature in the joint feature matrix is ​​weighted using the obtained feature importance data. The weighting method can be to use the importance score of each feature as a weight and multiply it by the corresponding feature value, thereby enhancing the influence of high-importance features and reducing the influence of low-importance features. If the importance score of a feature is 0.8, the value of the feature will be multiplied by 0.8. Similarly, low-importance features (such as a score of 0.2) will be multiplied by a lower corresponding weight. A weighted joint feature matrix is ​​obtained, in which the value of each feature has been adjusted according to its importance.

[0148] Step S253: performing feature concatenation on the weighted joint feature matrix to obtain an initial high-dimensional feature vector;

[0149] In this embodiment, data from different modalities are spliced ​​on the basis of the weighted joint feature matrix. The feature columns of each data source are arranged in order to form a high-dimensional vector containing all feature information. For example, feature data from different sources such as chemical composition data, production environment data, and equipment status data are spliced ​​to form a high-dimensional vector representing all information of a single sample. The splicing operation can be completed by vector connection or more complex matrix stacking. The spliced ​​high-dimensional feature vector contains all the feature information of the multi-modal and is the basis for the subsequent deep learning model input.

[0150] Step S254: performing nonlinear feature combination processing on the initial high-dimensional feature vector to obtain a preliminary sample feature embedding vector;

[0151] In this embodiment, a nonlinear combination method (such as a neural network, a support vector machine, or a decision tree) is used to process the initial high-dimensional feature vector. These methods can capture the nonlinear relationship between features and optimize the combination of multiple features through model learning. The high-dimensional feature vector is processed by training a shallow neural network or a multi-layer perceptron (MLP). The network can learn the complex nonlinear relationship between features through nonlinear activation functions (such as ReLU, sigmoid), thereby generating a low-dimensional embedding vector representing the important features of the sample. The feature vector after nonlinear combination will be used as a preliminary sample feature embedding vector, which contains the comprehensive information of the sample and provides a basis for the subsequent optimization process.

[0152] Step S255: Perform embedding vector regularization overfitting optimization on the preliminary sample feature embedding vector to obtain a sample feature embedding vector.

[0153] In this embodiment, L2 regularization (or other regularization techniques) is used to process the preliminary sample feature embedding vector to avoid overfitting of the model. Regularization limits the excessive size of model parameters by adding a penalty term, thereby improving generalization ability. In order to avoid overfitting of the embedding vector on the training data, a cross-validation or early stopping strategy is used to select the appropriate regularization strength. These strategies help to find the best embedding vector during the training process while avoiding the model from being too dependent on the noise in the training data. After regularization optimization, the final sample feature embedding vector has good generalization ability and can be used for downstream tasks such as classification and regression.

[0154] Optionally, step S251 is specifically:

[0155] Perform feature screening based on the joint feature matrix to obtain the joint feature matrix to be analyzed;

[0156] In this embodiment, a statistical test method (such as chi-square test, analysis of variance, etc.) or a feature selection algorithm in machine learning (such as recursive feature elimination RFE or L1 regularization) is used to evaluate the correlation between each feature and the target variable (the target variable can be a specific component content, a quality score of the sample, or other indicators related to Niushe Kaiqiao Powder). Features that are highly correlated with the target variable are screened out according to a preset threshold (for example, a p value less than 0.05 or a feature importance greater than a certain threshold). The screened matrix contains only the features most relevant to the target task as the joint feature matrix to be analyzed.

[0157] The chemical component proportion of Niushe Kaiqiao Powder is calculated based on the component distribution data of the spectrum characteristic pattern, so as to obtain the chemical component proportion data of Niushe Kaiqiao Powder;

[0158] In this embodiment, the distribution of each chemical component of Niushe Kaiqiao Powder is extracted from the spectrum feature pattern data. For example, the spectrum characteristics of the chemical components are analyzed using mass spectrometry or liquid chromatography. According to the peak intensity or signal intensity in the spectrum, the proportion of each chemical component is calculated. The calculation formula for the proportion is: \text{Proportion}=\frac{\text{Peak intensity of a chemical component}}{\text{Total peak intensity}}\times 100\%. The calculation result obtains the relative proportion data of each chemical component in Niushe Kaiqiao Powder.

[0159] According to the chemical component proportion data of Niushe Kaiqiao Powder, the chemical components of the joint feature matrix to be analyzed are mapped to obtain the feature weight factor;

[0160] In this embodiment, according to the obtained chemical composition proportion data, the proportion value of each component is mapped to the corresponding feature in the joint feature matrix. The proportion data can be associated with each feature in the feature matrix by weighted mapping, and the feature weight factor is calculated. Each chemical component proportion data is multiplied by the eigenvalue in the joint feature matrix to form a corresponding feature weight factor. The feature weight factor finally obtained represents the importance of each feature in Niushe Kaiqiao San, reflecting its contribution to the overall chemical composition.

[0161] Calculate the standard importance weight of the joint feature matrix to be analyzed, so as to obtain the feature standard importance weight;

[0162] In this embodiment, supervised learning algorithms such as random forests and support vector machines (SVM) are used to calculate the importance of features based on the joint feature matrix to be analyzed. The contribution of each feature to the target variable is determined by training the model. Common methods include feature importance assessment based on decision trees. The calculated feature importance weights are standardized so that the weight values ​​of all features are compared under the same standard. Normalization methods are usually used to adjust the weight values ​​to between 0 and 1. After calculation, the standardized importance weight data of each feature is obtained.

[0163] Based on the feature weight factor and the feature standard importance weight, weighted fusion is performed to obtain the modified comprehensive feature importance value;

[0164] In this embodiment, for each feature, the feature weight factor and the standard importance weight are combined to calculate the comprehensive feature importance value using weighted average or other fusion algorithms (such as principal component analysis PCA). The formula for weighted fusion can be expressed as follows: Modified comprehensive feature importance value = A × feature weight factor + (1-A) × standard importance weight; where A is the fusion weight coefficient, which can usually be determined by cross-validation. By adjusting the weight coefficient (A), the contribution of the two weights can be flexibly controlled to optimize the final comprehensive feature importance value. The obtained modified comprehensive feature importance value is used to characterize the contribution of each feature to the overall task, which helps in subsequent analysis and model training.

[0165] The feature importance is assigned to the joint feature matrix according to the modified comprehensive feature importance value to obtain the feature importance data.

[0166] In this embodiment, an importance weight is assigned to each feature in the joint feature matrix based on the corrected comprehensive feature importance value. According to the obtained corrected comprehensive feature importance value, a corresponding weight value is assigned to each feature in the joint feature matrix. This weight value indicates the contribution of the feature to model training and prediction. These importance values ​​are multiplied by the features in the joint feature matrix to adjust the influence of each feature in the model. The weight data of each feature is obtained, which will provide a basis for subsequent model training.

[0167] Optionally, step S3 specifically includes:

[0168] Step S31: performing sample similarity calculation according to the sample feature embedding vector to obtain a sample similarity matrix, and constructing a sample relationship network based on the sample similarity matrix to obtain a sample relationship graph;

[0169] In this embodiment, the sample similarity matrix is ​​obtained by calculating the similarity between the sample feature embedding vectors. Common similarity calculation methods include cosine similarity, Euclidean distance or Manhattan distance, etc. For example, the embedding vectors can be compared using cosine similarity to obtain the similarity value between each pair of samples, thereby forming a sample similarity matrix. Based on this matrix, a sample relationship graph can be constructed by a graph theory method, in which each node represents a sample, the edges between the nodes represent the similarity between the samples, and the weights of the edges represent the size of the similarity. The graph can be further analyzed by a community detection algorithm to reveal the potential relationship between the samples.

[0170] Step S32: performing sample component co-occurrence frequency analysis on the sample relationship graph to obtain sample component co-occurrence data, and constructing a component co-occurrence matrix based on the sample component contribution data;

[0171] In this embodiment, sample component co-occurrence data is constructed by analyzing the co-occurrence frequency of chemical components between samples. For example, a frequent item set mining algorithm (such as the Apriori algorithm) can be used to mine the co-occurrence pattern of chemical components in samples to obtain component co-occurrence frequency data. These co-occurrence data reflect the occurrence of different chemical components in different samples. Subsequently, based on these co-occurrence frequency data, a component co-occurrence matrix is ​​constructed, and each item in the matrix represents the co-occurrence frequency of two components in the sample. This matrix can help understand which components often appear together in a sample, thereby revealing the potential association between them.

[0172] Step S33: Calculating the component Pearson correlation coefficient of the component co-occurrence matrix to obtain co-occurrence component correlation data, and constructing a component relationship structure based on the co-occurrence component correlation data to obtain the component relationship structure data of Niushe Kaiqiao Powder;

[0173] In the present embodiment, the Pearson correlation coefficient is calculated for the component co-occurrence matrix to evaluate the linear correlation between different chemical components. The Pearson correlation coefficient is a common statistical method to measure the strength of the linear relationship between two variables. The calculation result will provide a correlation value for each pair of components, which is between -1 and +1. Based on these correlation data, a component relationship structure diagram can be constructed, in which each node represents a chemical component, the edge between the nodes represents the correlation strength of the two components, and the weight of the edge represents their degree of correlation. This component relationship structure diagram helps to further understand the synergy between different components.

[0174] Step S34: performing dynamic network modeling based on the component relationship structure data of Niushe Kaiqiao Powder to obtain a chemical component dynamic network, and performing chemical component synergy evaluation on the chemical component dynamic network to obtain chemical component synergy data;

[0175] In this embodiment, a dynamic network of chemical components is constructed based on the structural data of the relationship between the components of Niushe Kaiqiao Powder. This network reflects the synergistic relationship between different chemical components under different conditions. Dynamic network modeling can capture the relationship between components that change over time or conditions through methods such as time series analysis and system dynamics models. Subsequently, the synergistic effect of chemical components is evaluated to evaluate the effect of the interaction between different components in the network. The evaluation method may include methods such as centrality analysis based on the graph (such as degree centrality, proximity centrality) or modular detection in the network to determine which components play a key role in the synergy. The construction of the dynamic network model takes into account the changes in the interactive relationship between chemical components at different times or under different experimental conditions. For example, through time series analysis, based on sample data at different time points, the volatility of the relationship between components is evaluated. The interactive changes of chemical components under different conditions can be simulated, such as the effects of temperature, humidity or other environmental factors on the interaction between components. In the network, the weight of the edge can be dynamically adjusted based on the degree of synergy between the components. For each pair of components, the mutual influence and synergistic effect between them can be calculated, such as under certain conditions, some components may enhance the effects of other components, or inhibit each other. Dynamic network modeling will show the changing relationships between these components and help to gain a deeper understanding of how they work together. After the dynamic network is established, the next step is to evaluate the synergy of the chemical components. The synergy evaluation aims to analyze how the different chemical components work together in Niushe Kaiqiao Powder to produce an overall effect. The importance of each chemical component in the dynamic network can be evaluated, and indicators such as degree centrality (the number of node connections) and betweenness centrality (the mediating role of paths between nodes) can be used to quantify the core role of the components. For example, some components produce synergistic effects under certain conditions and jointly enhance the efficacy; while some components interact with each other, resulting in a weakened or inhibited effect.

[0176] Step S35: Calculating the strength of the relationship between components based on the relationship structure data of the Niushe Kaiqiao Powder components, thereby obtaining the strength data of the relationship between chemical components;

[0177] In the present embodiment, according to the component relationship structure data of Niushe Kaiqiao Powder, the calculation of the component relationship strength is performed. This can be achieved by a weighted network model, in which the relationship strength of each component can be determined by the analysis results of the similarity, co-occurrence frequency or synergy between the components. For example, the relationship strength between the components can be measured by the path length, co-occurrence frequency or mutual influence degree of the dynamic network of each pair of components in the network. The calculated relationship strength data helps to identify which components have a stronger relationship, and these components may play a more important role in the formula of Niushe Kaiqiao Powder.

[0178] Step S36: construct a synergistic network of the ingredients of the musk deer awakening powder according to the chemical ingredient synergistic effect data and the chemical ingredient relationship strength data.

[0179] In the present embodiment, according to the chemical composition synergy data and the relationship strength data, the chemical composition synergy network of Niushe Kaiqiao Powder is constructed. The nodes in the network represent chemical compositions, the edges represent the synergy relationship between the components, and the weights of the edges can be set based on the synergy data or the relationship strength of the components. Through this network construction, the interaction and influence between the chemical compositions can be understood more intuitively. For example, the clustering algorithm in graph theory can be used to analyze which components often appear together and have an important impact on sample characteristics. The final component synergy network will help reveal the complex relationship between the components in the formula, and provide a scientific basis for optimizing the formula of Niushe Kaiqiao Powder and improving its quality.

[0180] Optionally, step S4 is specifically:

[0181] Based on the synergistic network of Niushe Kaiqiao Powder ingredients, the relative importance of ingredients and the synergistic effect characteristics of ingredients were extracted, so as to obtain the synergistic effect data of Niushe Kaiqiao Powder ingredients and the relative importance data of Niushe Kaiqiao Powder ingredients;

[0182] In the present embodiment, by analyzing the synergistic network of the chemical components of Niushe Kaiqiao Powder, the relative importance of each component and its synergistic characteristics are extracted. Based on the dynamic network of chemical components constructed above, graph theory methods such as centrality analysis and influence propagation models can be used to calculate the relative importance of each component. The nodes of each component in the network are evaluated according to the connection strength, interaction mode and contribution to the overall effect with other components. For example, assuming that components A, B and C in Niushe Kaiqiao Powder have complex synergies with each other, the role of each component in the synergy can be quantified through dynamic network modeling. Component A shows a strong synergistic effect by enhancing the efficacy of component B, while component C inhibits the effects of A and B under certain conditions. By evaluating the intensity of synergy, the relative importance and synergy data of each component are obtained.

[0183] Peak intensity mapping is performed based on the synergistic effect data of the ingredients of Niushe Kaiqiao Powder and the relative importance data of the ingredients of Niushe Kaiqiao Powder, thereby obtaining the vertical axis of the fingerprint spectrum;

[0184] In this embodiment, based on the obtained component synergy data and relative importance data, it can be further mapped to the fingerprint spectrum of the chemical components, especially the vertical axis (i.e., intensity axis) of the spectrum. By weighted mapping the intensity of each chemical component, the component relative importance data can be used to adjust the intensity of each peak and highlight the spectrum peak of the important component. For example, assuming that the importance of component A is higher, the corresponding spectrum peak intensity on the vertical axis of the fingerprint spectrum should also increase accordingly, while the spectrum peak intensity of the low-importance component is weakened. The vertical axis of the fingerprint spectrum finally obtained reflects the contribution of each component to the efficacy of the drug.

[0185] Sort the chemical component characteristics according to the chemical component annotation matrix to obtain the molecular characteristic axis;

[0186] In this embodiment, the molecular characteristics of different chemical components, such as polarity, molecular weight, functional groups, etc., can be extracted through the chemical component annotation matrix. These characteristics provide a basis for sorting the components. For example, the components can be sorted according to their polarity or molecular weight to obtain a molecular characteristic axis, reflecting the differences in the molecular properties of different chemical components. For example, if component A has a higher polarity and component B has a lower polarity, the chemical component annotation matrix can sort the two components, thereby mapping their characteristics to the molecular characteristic axis.

[0187] Construct a global fingerprint spectrum based on the vertical axis of the fingerprint spectrum and the molecular feature axis;

[0188] In this embodiment, a global fingerprint spectrum can be constructed by combining the vertical axis of the fingerprint spectrum and the molecular characteristic axis. The fingerprint spectrum takes into account the relative importance of each component in the drug effect and its molecular characteristics, thereby providing a global chemical component spectrum. For example, the horizontal axis of the fingerprint spectrum represents the molecular characteristics of different components (such as polarity, molecular weight), and the vertical axis represents the relative importance of these components in the drug and the strength of synergy. In this way, the multidimensional information of chemical components can be effectively visualized, providing richer reference data for subsequent analysis.

[0189] Screening the global fingerprint spectrum for potential abnormal points, thereby obtaining abnormal peak data of the global fingerprint spectrum;

[0190] In this embodiment, after the global fingerprint spectrum is constructed, the next step is to screen the spectrum for potential outliers. It aims to identify abnormal peaks in the spectrum, such as abnormal peaks caused by experimental errors or sample problems. By applying anomaly detection algorithms (such as Z-score-based outlier detection, or clustering-based outlier analysis), peaks that are beyond the expected range can be identified. For example, if the intensity of a spectral peak is much higher than the normal range and has no obvious relationship with other spectral peaks, it is an abnormal peak. The screening process helps to ensure the accuracy of the global fingerprint spectrum and eliminate irrelevant or interfering signals.

[0191] The abnormal peak data of the global fingerprint spectrum is corrected to obtain an optimized fingerprint spectrum.

[0192] In this embodiment, for the abnormal peaks screened out, a spectrum peak correction strategy can be adopted for processing. Specifically, data smoothing, peak correction or interpolation correction methods can be used to achieve the technical effect of spectrum peak correction. For example, if the intensity of a peak is detected to be abnormal (such as too high), the intensity of the peak can be adjusted by a smoothing algorithm to match it with other similar peaks, thereby ensuring the stability and accuracy of the fingerprint spectrum.

[0193] Optionally, the potential abnormal point screening in step S4 is specifically as follows:

[0194] Performing baseline correction on the global fingerprint spectrum to obtain the global fingerprint spectrum to be analyzed;

[0195] In the present embodiment, baseline correction is intended to remove the offset caused by background noise or instrument drift during the experiment to ensure the accuracy of the spectrogram. Common methods include linear baseline correction and nonlinear baseline correction. For example, the baseline of the global fingerprint spectrogram can be fitted by the least square method, and by removing the fitted baseline, a revised fingerprint spectrogram is obtained. If low-frequency drift occurs in the spectrum, a low-pass filter is used to remove the drift component, and then baseline correction is performed. In this process, the vertical axis intensity of the spectrum will be adjusted so that the peak and noise parts are clearer, thereby providing more accurate data for subsequent analysis. For example, if in the fingerprint spectrogram of Niushe Kaiqiao Powder, the baseline in some areas rises or falls significantly, these baseline offsets can be removed by linear baseline correction to obtain a more realistic fingerprint spectrogram.

[0196] According to the preset division ratio, the global fingerprint spectrum to be analyzed is divided into regions to obtain a local fingerprint spectrum;

[0197] In this embodiment, the global fingerprint spectrum is divided into several local areas so that each area can be analyzed in detail. The division ratio can be set as needed, for example, the spectrum is evenly divided into several sections according to the equal division method, or the spectrum is divided according to the characteristic peak distribution. In each local area, the spectrum is analyzed separately to find abnormal or significant peaks in the area. For example, if the total length of the fingerprint spectrum is 1000 data points, it can be divided into 10 areas, each containing 100 data points. The division standard can be appropriately adjusted according to the characteristic distribution of the spectrum to ensure that each local fingerprint spectrum contains enough information for further processing. Through this division, you can effectively focus on analyzing the area with special properties in the fingerprint spectrum without analyzing the entire spectrum, thereby improving the efficiency and accuracy of the analysis.

[0198] Performing local peak intensity distribution statistics on the local fingerprint spectrum to obtain local peak intensity distribution data, and performing significant deviation peak point identification on the local peak intensity distribution data to obtain abnormal local significant deviation peak point data;

[0199] In this embodiment, the peak values ​​in the local fingerprint spectrum are extracted, and the peak intensity distribution in each local area is statistically analyzed. A peak detection algorithm (such as the first derivative method or the second derivative method) can be used to locate the peak value, and its intensity is statistically analyzed. According to the deviation of the intensity distribution, a statistical method (such as Z-score or standard deviation) is used to identify the peak points that deviate significantly. For example, if the intensity of a peak value is much higher than other peak values ​​in the local area, it can be considered as an abnormal peak point. The deviation standard can be set to a peak point whose intensity exceeds 2 times the standard deviation of the average intensity of the local area. For example, in the fingerprint spectrum of Niushe Kaiqiao Powder, the intensity of some peak values ​​in the local area is abnormal, which may be due to the special reaction of the sample components or instrument errors. These abnormal peaks can be located by intensity distribution statistics and deviation peak point identification methods.

[0200] The intensity value variation of the points adjacent to the peak point is calculated for the local fingerprint spectrum according to the preset window range, so as to obtain the intensity value variation data of the points adjacent to the peak point, and the intensity mutation variation data of the intensity value variation data of the points adjacent to the peak point is statistically analyzed, so as to obtain the abnormal intensity mutation peak point data;

[0201] In this embodiment, for the peak points identified in the local fingerprint spectrum, a window range (such as 5 data points) is set to calculate the intensity change of the neighboring points around the peak point. The intensity mutation is calculated by comparing the intensity difference of the data adjacent to the peak point. For example, if the neighboring intensity value of a peak point changes significantly (exceeds the threshold setting, such as the change is greater than a certain percentage), it can be considered that the peak point has an intensity mutation. At this time, gradient calculation or differential method can be used to evaluate the intensity change of the neighboring points of the peak point. For example, assuming that in the fingerprint spectrum of Niushe Kaiqiao Powder, the intensity of the data points around a peak has significant fluctuations, through this variation statistical method, abnormal intensity mutation peaks can be identified.

[0202] The peak point intersection operation is performed on the abnormal local significant deviation peak point data and the abnormal intensity mutation peak point data to obtain the global fingerprint spectrum abnormal peak data.

[0203] In the present embodiment, the abnormal peak points identified are subjected to intersection operation to obtain the final abnormal peak point data. In the process, the local significant deviation peak point and the intensity mutation peak point are subjected to intersection processing to ensure that only the peak points that appear simultaneously in these two cases are marked as abnormal. For example, the peak points that deviate significantly in the local peak intensity distribution are first screened out, and then the peak points that mutate in the adjacent point intensity value variation statistics are screened out, and the final abnormal peak points are obtained by intersection operation. These abnormal peak points reflect the abnormal changes in the spectrogram and can be further processed or corrected. For example, in the fingerprint spectrogram of Niushe Kaiqiao Powder, the abnormal peak points found by intersection operation may be caused by a special reaction between a certain component in the sample and the instrument, and these peak points can be marked as abnormal data for subsequent quality control and data correction.

[0204] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0205] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for detecting a Chinese medicine fingerprint of a Chinese medicine preparation, characterized in that: The following steps are involved: Step S1: obtaining high-dimensional spectrum data of Niu Musk Kaiqiao Powder, and performing chemical component spectrum decomposition on the high-dimensional spectrum data of Niu Musk Kaiqiao Powder, so as to obtain a chemical component spectrum of Niu Musk Kaiqiao Powder; performing qualitative characteristic analysis of the chemical components of Niu Musk Kaiqiao Powder according to the chemical component spectrum of Niu Musk Kaiqiao Powder, so as to obtain a chemical component characteristic matrix of Niu Musk Kaiqiao Powder; Step S2: Obtain the multimodal production data of Niu Musk Kaiqiao Powder, and unify the Niu Musk Kaiqiao Powder sample source feature space of the multimodal production data of Niu Musk Kaiqiao Powder, so as to obtain the Niu Musk Kaiqiao Powder sample source feature space; perform high-dimensional feature fusion on the Niu Musk Kaiqiao Powder chemical component feature matrix and the Niu Musk Kaiqiao Powder sample source feature space, so as to obtain the sample feature embedding vector; Step S2 is specifically: Step S21: acquiring the multimodal production data of Niushe Kaiqiao Powder, and performing data preprocessing on the multimodal production data of Niushe Kaiqiao Powder, so as to obtain the multimodal production data of Niushe Kaiqiao Powder to be analyzed; Step S22: performing a principal component analysis of global sample source characteristics on the multimodal production data of Niushe Kaiqiao Powder to be analyzed, thereby obtaining principal component multimodal production data, and performing nonlinear distribution feature dimensionality reduction on the principal component multimodal production data, thereby obtaining Niushe Kaiqiao Powder sample source data; Step S23: performing feature alignment on the Niushe Kaiqiao Powder sample source data to obtain aligned sample source data, and performing unified feature space mapping based on the aligned sample source data to obtain the Niushe Kaiqiao Powder sample source feature space; Step S24: aligning the dimension and scale of the chemical component feature matrix of Niushe Kaiqiao Powder and the feature space of the source of Niushe Kaiqiao Powder samples, so as to obtain a joint feature matrix; Step S25: performing high-dimensional feature fusion according to the joint feature matrix to obtain a sample feature embedding vector; Step S3: performing sample relationship structure analysis according to the sample feature embedding vector to obtain the relationship structure data of the ingredients of Niushe Kaiqiao Powder, and constructing a chemical component dynamic network based on the relationship structure data of the ingredients of Niushe Kaiqiao Powder to obtain the synergistic network of the ingredients of Niushe Kaiqiao Powder; Step S4: generating a global fingerprint spectrum based on the collaborative network of Niushe Kaiqiao Powder components, and screening the global fingerprint spectrum for potential abnormal points, thereby obtaining abnormal peak data of the global fingerprint spectrum; performing peak correction on the abnormal peak data of the global fingerprint spectrum, thereby obtaining an optimized fingerprint spectrum; Step S5: construct and verify the characteristic model according to the optimized fingerprint spectrum, so as to obtain the quality evaluation model of Niushe Kaiqiao Powder.

2. The method for detecting the fingerprint of a Chinese medicine preparation according to claim 1, characterized in that: Step S1 is specifically as follows: Step S11: obtaining high-dimensional spectrum data of Niushe Kaiqiao Powder, and performing spectrum data preprocessing on the high-dimensional spectrum data of Niushe Kaiqiao Powder, so as to obtain high-dimensional spectrum data of Niushe Kaiqiao Powder to be analyzed; Step S12: obtaining chemical component characteristic data based on a preset chemical library, and integrating the chemical component qualitative characteristics according to the chemical component characteristic data, thereby obtaining a chemical component characteristic matrix; Step S13: performing chemical component spectrum decomposition on the high-dimensional spectrum data of Niushe Kaiqiao Powder to be analyzed, thereby obtaining a chemical component spectrum, and performing chemical component relative intensity calculation on the chemical component spectrum, thereby obtaining chemical component intensity data; Step S14: performing chemical component-specific spectrum decomposition on the chemical component spectrum according to the chemical component characteristic matrix and the chemical component intensity data, thereby obtaining the chemical component spectrum of Niushe Kaiqiao Powder; Step S15: Perform qualitative characteristic analysis of the chemical components of Niushe Kaiqiao Powder according to the chemical component spectrum of Niushe Kaiqiao Powder, so as to obtain the chemical component characteristic matrix of Niushe Kaiqiao Powder.

3. The method for detecting the fingerprint of a Chinese medicine preparation according to claim 2, characterized in that: Step S14 is specifically as follows: Step S141: performing Pearson correlation coefficient feature matching on the chemical component feature matrix and the chemical component intensity data, thereby obtaining a preliminary matching matrix; Step S142: performing sparse coding specific factor separation based on the preliminary matching matrix, thereby obtaining a specific factor matrix; Step S143: extracting the spectrogram features of the specific factor matrix to obtain specific factor spectrogram feature data, and performing intensity normalization on the specific factor spectrogram feature data and the chemical component intensity data to obtain intensity normalized spectrogram feature data; Step S144: Calculating the relative contribution of components according to the intensity normalized spectrum feature data, thereby obtaining relative contribution data of components, and constructing a relative contribution matrix based on the relative contribution data of components; Step S145: reconstructing the specific spectrum based on the relative contribution matrix and the specific factor matrix, thereby obtaining the specific spectrum of the chemical components of Niushe Kaiqiao Powder; Step S146: According to the chemical component specific spectrum of Niushe Kaiqiao Powder, the original chemical component spectrum is specifically tested and optimized to obtain the chemical component spectrum of Niushe Kaiqiao Powder.

4. The method for detecting the fingerprint of a Chinese medicine preparation according to claim 2, characterized in that: Step S15 is specifically as follows: Step S151: extracting peak features from the chemical component spectrum of Niushe Kaiqiao Powder to obtain a peak feature matrix; Step S152: performing correlation matching according to the chemical component feature matrix and the peak feature matrix, thereby obtaining a chemical component annotation matrix; Step S153: performing spectrum characteristic pattern recognition on the chemical component annotation matrix to obtain spectrum characteristic pattern data; performing characteristic pattern component distribution analysis based on the spectrum characteristic pattern data to obtain spectrum characteristic pattern component distribution data; Step S154: integrating the chemical component qualitative characteristics based on the spectrum characteristic pattern data and the spectrum characteristic pattern component distribution data, thereby obtaining the chemical component qualitative characteristic data; Step S155: construct the chemical component characteristic matrix of Niushe Kaiqiao Powder based on the peak characteristic matrix, the chemical component annotation matrix and the chemical component qualitative characteristic data.

5. The method for detecting the fingerprint of a Chinese medicine preparation according to claim 1, characterized in that: Step S25 is specifically as follows: Step S251: performing feature importance evaluation on the joint feature matrix to obtain feature importance data; Step S252: performing feature weighting processing on the joint feature matrix according to the feature importance data, thereby obtaining a weighted joint feature matrix; Step S253: performing feature concatenation on the weighted joint feature matrix to obtain an initial high-dimensional feature vector; Step S254: performing nonlinear feature combination processing on the initial high-dimensional feature vector to obtain a preliminary sample feature embedding vector; Step S255: Perform embedding vector regularization overfitting optimization on the preliminary sample feature embedding vector to obtain a sample feature embedding vector.

6. The method for detecting the fingerprint of a Chinese medicine preparation according to claim 5, characterized in that: Step S251 is specifically as follows: Perform feature screening based on the joint feature matrix to obtain the joint feature matrix to be analyzed; The chemical component proportion of Niushe Kaiqiao Powder is calculated based on the component distribution data of the spectrum characteristic pattern, so as to obtain the chemical component proportion data of Niushe Kaiqiao Powder; According to the chemical component proportion data of Niushe Kaiqiao Powder, the chemical components of the joint feature matrix to be analyzed are mapped to obtain the feature weight factor; Calculate the standard importance weight of the joint feature matrix to be analyzed, so as to obtain the feature standard importance weight; Based on the feature weight factor and the feature standard importance weight, weighted fusion is performed to obtain the modified comprehensive feature importance value; The feature importance is assigned to the joint feature matrix according to the modified comprehensive feature importance value to obtain the feature importance data.

7. The method for detecting the fingerprint of a Chinese medicine preparation according to claim 1, characterized in that: Step S3 is specifically as follows: Step S31: performing sample similarity calculation according to the sample feature embedding vector to obtain a sample similarity matrix, and constructing a sample relationship network based on the sample similarity matrix to obtain a sample relationship graph; Step S32: performing sample component co-occurrence frequency analysis on the sample relationship graph to obtain sample component co-occurrence data, and constructing a component co-occurrence matrix based on the sample component contribution data; Step S33: Calculating the component Pearson correlation coefficient of the component co-occurrence matrix to obtain co-occurrence component correlation data, and constructing a component relationship structure based on the co-occurrence component correlation data to obtain the component relationship structure data of Niushe Kaiqiao Powder; Step S34: performing dynamic network modeling based on the component relationship structure data of Niushe Kaiqiao Powder to obtain a chemical component dynamic network, and performing chemical component synergy evaluation on the chemical component dynamic network to obtain chemical component synergy data; Step S35: Calculating the strength of the relationship between components based on the relationship structure data of the Niushe Kaiqiao Powder components, thereby obtaining the strength data of the relationship between chemical components; Step S36: construct a synergistic network of the ingredients of the musk deer awakening powder according to the chemical ingredient synergistic effect data and the chemical ingredient relationship strength data.

8. The method for detecting the fingerprint of a Chinese medicine preparation according to claim 1, characterized in that: Step S4 is specifically as follows: Based on the synergistic network of Niushe Kaiqiao Powder ingredients, the relative importance of ingredients and the synergistic effect characteristics of ingredients were extracted, so as to obtain the synergistic effect data of Niushe Kaiqiao Powder ingredients and the relative importance data of Niushe Kaiqiao Powder ingredients; Peak intensity mapping is performed based on the synergistic effect data of the ingredients of Niushe Kaiqiao Powder and the relative importance data of the ingredients of Niushe Kaiqiao Powder, thereby obtaining the vertical axis of the fingerprint spectrum; Sort the chemical component characteristics according to the chemical component annotation matrix to obtain the molecular characteristic axis; Construct a global fingerprint spectrum based on the vertical axis of the fingerprint spectrum and the molecular feature axis; Screening the global fingerprint spectrum for potential abnormal points, thereby obtaining abnormal peak data of the global fingerprint spectrum; The abnormal peak data of the global fingerprint spectrum is corrected to obtain an optimized fingerprint spectrum.

9. The method for detecting the fingerprint of a Chinese medicine preparation according to claim 8, characterized in that: The potential outlier screening in step S4 is specifically as follows: Performing baseline correction on the global fingerprint spectrum to obtain the global fingerprint spectrum to be analyzed; According to the preset division ratio, the global fingerprint spectrum to be analyzed is divided into regions to obtain a local fingerprint spectrum; Performing local peak intensity distribution statistics on the local fingerprint spectrum to obtain local peak intensity distribution data, and performing significant deviation peak point identification on the local peak intensity distribution data to obtain abnormal local significant deviation peak point data; The intensity value variation of the points adjacent to the peak point is calculated for the local fingerprint spectrum according to the preset window range, so as to obtain the intensity value variation data of the points adjacent to the peak point, and the intensity mutation variation data of the intensity value variation data of the points adjacent to the peak point is statistically analyzed, so as to obtain the abnormal intensity mutation peak point data; The peak point intersection operation is performed on the abnormal local significant deviation peak point data and the abnormal intensity mutation peak point data to obtain the global fingerprint spectrum abnormal peak data.

Images