Artificial intelligence-oriented chromatography mass spectrometry data management method and system

Abstract

This invention discloses a method and system for chromatographic mass spectrometry data management based on artificial intelligence. The method includes: performing dual-domain collaborative adaptive filtering on sample chromatograms to obtain purified sample chromatograms; matching standard chromatograms according to chromatographic conditions; calculating the first similarity between the purified sample chromatograms and the standard chromatograms to obtain a standard screening chromatogram set; calculating the relative retention times of each characteristic peak in the purified sample chromatograms; calculating the cosine similarity between the relative retention times and the corresponding standard retention time values ​​of the standard screening chromatograms to obtain a second similarity; determining sample component prediction; generating a sample chromatogram matrix; embedding chromatographic metadata into the sample chromatogram matrix to obtain chromatographic mass spectrometry managed data; and storing the data according to the detection information and generating storage addresses. This method not only improves the efficiency and accuracy of chromatographic mass spectrometry data management but also has good interpretability and can be directly applied to chromatographic mass spectrometry data management systems.

Description

Technical Field

[0001] This invention relates to the fields of analytical chemistry and data governance technology, and in particular to a method and system for chromatographic and mass spectrometric data governance for artificial intelligence. Background Technology

[0002] Chromatography-mass spectrometry (GC-MS) technology is widely used in the fields of quality control of Chinese medicinal materials, food safety testing, environmental pollutant monitoring and metabolomics analysis. With the in-depth application of artificial intelligence technology in intelligent compound identification, quality prediction and process control, higher requirements are placed on the accuracy, standardization and adaptability of GC-MS data to AI models.

[0003] However, existing chromatographic mass spectrometry data management technologies have significant limitations: First, traditional baseline correction and noise removal often employ separate processing methods, which can easily lead to distortion of chromatographic peak areas or sharp peak shape distortion, and lacks dynamic compensation for physical factors such as instrument drift; second, existing normalization methods are mostly static statistical methods, failing to consider the gradient propagation requirements and structural characteristics of subsequent deep neural networks (DNNs), and chromatographic metadata is not involved in data management and AI input; at the same time, traditional methods lack quantitative analysis of multi-scale morphological characteristics of peak shapes and do not fully utilize mass spectrometry fragment ion information for structural confirmation, resulting in a high false positive matching rate; finally, existing solutions mostly use physical storage based on filenames, making it difficult to guarantee data integrity verification and tamper-proof traceability. Therefore, this invention proposes a method and system for managing chromatographic and mass spectrometry data for artificial intelligence. It ensures peak shape preservation by jointly optimizing chromatograms through dual-domain collaborative adaptive filtering, improves matching accuracy through multi-scale morphological features, mass spectrometry verification, and multi-level similarity verification, enhances the cross-batch generalization capability of AI models through dynamic embedding of metadata, and achieves reliable storage of AI-ready data through spatiotemporal-content hybrid addressing. This provides a standardized and high-quality data foundation for the intelligent application of chromatographic and mass spectrometry data. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for managing chromatographic and mass spectrometric data oriented towards artificial intelligence.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution: This invention includes the following steps: Acquire chromatographic and mass spectrometric data and chromatographic metadata of the submitted sample, perform dual-domain collaborative adaptive filtering on the sample chromatogram to obtain a purified sample chromatogram, and finally perform peak extraction and peak alignment operations. Match the standard chromatograms with the chromatographic conditions corresponding to the chromatograms of the purified samples, extract the chromatographic feature vectors and full-scan mass spectra of the purified sample chromatograms, and calculate the first similarity between the purified sample chromatograms and the standard chromatograms to obtain the standard screening chromatogram set. Extract the reference peaks and relative retention time standard values ​​of the standard screening chromatogram, calculate the relative retention time of each characteristic peak of the purified sample chromatogram, calculate the cosine similarity between the relative retention time standard value and the relative retention time to obtain the second similarity, and determine the chromatogram prediction information of the purified sample. Sample component prediction is obtained from the chromatogram prediction information of all purified samples, a sample chromatogram matrix is ​​generated, chromatographic metadata is embedded into the sample chromatogram matrix to obtain chromatographic mass spectrometry treatment data, and storage is performed according to the detection information and the storage address is generated. The chromatographic mass spectrometry data includes chromatograms, chromatographic conditions, and detection information; the chromatographic conditions include mobile phase composition, flow rate, detection wavelength, and column temperature; the detection information includes sampling time, sampling location, detection time, and detection location. The chromatographic metadata includes instrument drift coefficient, signal-to-noise ratio, and column efficiency parameters; The standard chromatogram is obtained from a specific standard substance under standard chromatographic conditions; The chromatographic feature vector includes basic chromatographic features and speciation chromatographic features; the basic chromatographic features include the number of characteristic peaks, the characteristic peak area vector, and the distribution density; the speciation chromatographic features include the peak sharpness index, the peak asymmetry factor, and the peak skewness coefficient. The sample chromatogram prediction information includes predicted raw materials and predicted content.

[0006] 3. Furthermore, the method for performing dual-domain collaborative adaptive filtering includes: The chromatographic metadata is input as a priori initialization parameter into the dual-domain collaborative adaptive filtering module, and the sample chromatogram is input into the dual-domain collaborative adaptive filtering module for dual-domain collaborative adaptive filtering. The dual-domain collaborative adaptive filtering module includes a physical domain estimator and a data domain denoiser. The physical domain estimator initializes the decay rate using the chromatographic data instrument drift coefficient, constructs a baseline drift function, obtains an initial baseline estimate by least squares fitting of the physical domain parameter vector, and obtains the intermediate signal by subtracting the initial baseline estimate from the sample chromatogram. The baseline function expression is: ; in For baseline drift function, To retain the time, The number of baseline components, For physical domain parameter vectors, For the first Each index component amplitude, For the first An exponential decay rate, For periodic interference amplitude, For periodic interference angular frequency, This is the DC offset; The data domain denoiser performs multi-level stationary wavelet transform on the intermediate signal to obtain approximation coefficients and detail coefficients. Based on the second derivative of the chromatographic peak and the signal-to-noise ratio of the chromatographic data, it calculates the peak density index and adaptive threshold. The detail coefficients are adjusted according to the adaptive threshold to obtain the denoised detail coefficients. Then, the inverse wavelet transform is performed to reconstruct the data domain denoised signal. The expressions for calculating the peak density index and the adaptive threshold are as follows: ; ; in for Layer adaptive threshold, for Standard deviation of layer detail factor For signal length, For signal-to-noise ratio, The peak protection strength coefficient, This is the peak density attenuation coefficient. for Layer adaptive threshold, This is an intermediate signal; The method for calculating the denoising detail coefficients by adjusting the detail coefficients based on the adaptive threshold is as follows: ; in for Layer denoising detail coefficients, for Layer detail factor; The dual-domain collaborative adaptive filtering module uses a dual-domain joint loss function to optimize the adaptive threshold and physical domain parameter vector, ultimately outputting the chromatogram of the purified sample, expressed as follows: ; ; in For the joint loss function of two domains, For an adaptive threshold set, This is the original signal, corresponding to the sample chromatogram. For the current iteration baseline estimate, For the current iteration signal estimation, For physical domain regularization weights, For data domain regularization weights, For baseline smoothing regularization, The wavelet decomposition level is denoted as . This is the filtered signal, corresponding to the chromatogram of the purified sample. For the optimal physical domain parameter vector The corresponding optimal baseline estimate.

[0007] Furthermore, the method for obtaining the standard screening chromatogram set includes: Standard chromatograms are acquired, categorized, and stored according to substance type to establish a standard chromatogram library. Each standard chromatogram is associated with its corresponding quasi-substance, standard chromatogram, and chromatographic metadata. First, match the standard chromatograms in the standard chromatogram library with mobile phase polarity parameters that deviate within ±15% and have the same elution mode as the chromatogram of the purified sample. Then, select any standard chromatogram with an instrument parameter deviation rate of no more than 20%. The instrument parameter deviation rate includes column size deviation rate, column temperature deviation rate, flow rate deviation rate, and detection wavelength deviation rate. The basic chromatographic features are directly extracted from the chromatogram of the purified sample. The speciation chromatographic features are then calculated based on the detail factor in the chromatogram of the purified sample. The expression is as follows: ; ; ; in for Stratification sharpness index. For peak asymmetry factor, This is the peak skewness coefficient. The number of characteristic peaks, To preserve the time variable, for Characteristic peak retention time interval For stationary wavelet transform Layer detail factor, for layer The characteristic peak width at 5% is shown. for layer The distance of the leading edge of the characteristic peak, for Retention time of characteristic peaks; The full-scan mass spectra corresponding to each characteristic peak of the purified sample chromatogram are extracted and compared with the mass spectra of the corresponding standard substances in the standard chromatogram to calculate the mass spectrometry fragment ion matching degree. Chromatographic features are composed of basic chromatographic features and speciation chromatographic features. The feature matching degree of the chromatographic feature vector between the purified sample chromatogram and the candidate standard chromatogram is calculated. The first similarity is obtained by weighting the mass spectrometry fragment ion matching degree and feature matching degree, expressed as: ; ; in Chromatogram of purified sample and First similarity of standard chromatograms To purify the sample chromatogram and Mass spectrometric fragment ion matching degree of standard chromatograms. , For similarity weights, This is the feature vector of the sample chromatogram. for The characteristic vector of a standard chromatogram For the feature covariance matrix of the training set, This represents the upper limit of the mass spectrometry scanning range. To purify the sample chromatogram Mass-to-charge ratio ionic strength For the first Weighting factors for the mass-to-charge ratio, for Standard chromatogram No. Individual mass-to-charge ratio ionic strength; A standard screening chromatogram set is constructed by selecting standard chromatograms that meet the first similarity threshold.

[0008] Furthermore, the method for determining sample chromatogram prediction information includes: Extract the reference peaks and corresponding relative retention time values ​​of each standard screening chromatogram. Calculate the relative retention time of each characteristic peak based on the reference peaks of the purified sample chromatogram. Calculate the cosine similarity between the vector of relative retention time values ​​of the standard screening chromatogram and the vector of relative retention time corresponding to each characteristic peak of the purified sample chromatogram as the second similarity. Select standard screening chromatograms with a second similarity greater than the second similarity threshold as sample component chromatograms. Use the standard substances corresponding to the sample component chromatograms as predictive raw materials. The predicted content of the corresponding standard substance is calculated based on the characteristic peak areas and detection solution preparation parameters in the chromatograms of the purified sample and the chromatograms of the sample components.

[0009] Furthermore, the method for obtaining chromatographic mass spectrometry processing data includes: The chromatograms of each purified sample of the same sample were normalized in the time-mass-charge ratio dimension to obtain a single-sample tensor. The sample component prediction was then encoded and spatiotemporally mapped. The component tensor is obtained by concatenating the single sample tensor with the component tensor, and the sample spectral matrix is ​​obtained by stacking all sub-samples in the same batch; the same batch includes sub-samples of the same sample, the same sampling time, and different chromatographic conditions. The chromatographic metadata is subjected to linear and nonlinear projections respectively, and the projection results are fused to obtain a multi-scale metadata embedding vector, expressed as: ; in For scale metadata embedding vectors, As a linear projection matrix, the chromatographic metadata is... Mapped to the chromatogram channel dimension, As the first nonlinear projection matrix, the chromatographic data... Mapping to hidden dimensions to extract non-linear features. The second nonlinear projection matrix maps the hidden dimensions back to the chromatogram channel dimensions; The mean and variance of the sample spectral matrix are calculated along the batch dimension. Multi-scale metadata is embedded into a vector input to a lightweight fully connected network to generate spatially and channel-adaptive scaling and offset parameters. Finally, adaptive normalization is performed on the batch spectral matrix to obtain chromatographic and mass spectrometric treatment data. The expression is: ; ; in For chromatographic and mass spectrometric processing data, , Spatial-channel adaptive scaling and offset parameters, This is the sample spectrum matrix. , The mean and variance of the sample spectrum matrix are given. It is the numerical stability constant. Generate a weight matrix for the affine parameters. Generate a bias vector for the affine parameters.

[0010] Furthermore, the method for generating the storage address includes: The chromatographic mass spectrometry processing data and the corresponding chromatographic mass spectrometry data and chromatographic metadata are encapsulated into a structured data package; A basic path is generated based on the sample batch number and sampling time and location from the testing information. A content identifier is generated based on the chromatographic mass spectrometry processing data and component prediction results. The storage address is obtained using the basic path and content identifier, and the structured data package is stored according to the storage address. The expression is: ; ; ; in For storage address, Basic path, For content identification, For sampling locations, For sampling time, For batch number, for Chromatographic and mass spectrometric processing data of the sample, For component prediction feature vectors, For hash functions, It uses the SHA-256 secure hash algorithm. This is a Merkel tree root hash.

[0011] Secondly, the chromatographic mass spectrometry data management system oriented towards artificial intelligence includes: Chromatography purification module: used to acquire chromatographic and mass spectrometric data and chromatographic metadata of the submitted sample, perform dual-domain collaborative adaptive filtering on the sample chromatogram to obtain a purified sample chromatogram, and finally perform peak extraction and peak alignment operations; Standard substance screening module: It is used to match each standard chromatogram with the chromatographic conditions corresponding to the chromatogram of the purified sample, extract the chromatographic feature vector and full scan mass spectrum of the purified sample chromatogram, and calculate the first similarity between the purified sample chromatogram and the standard chromatogram to obtain the standard screening chromatogram set. Component prediction module: used to extract reference peaks and relative retention time standard values ​​of standard screening chromatograms, calculate the relative retention time of each characteristic peak of the purified sample chromatogram, calculate the cosine similarity between the relative retention time standard value and the relative retention time to obtain the second similarity, and determine the chromatogram prediction information of the purified sample. The sample component prediction is obtained from the chromatogram prediction information of all purified samples. Data governance module: used to generate sample spectral matrix, embed chromatographic metadata into the sample spectral matrix to obtain chromatographic mass spectrometry governance data, and generate storage addresses according to the detection information for storage.

[0012] The beneficial effects of this invention are: This invention relates to a method and system for managing chromatographic and mass spectrometric data oriented towards artificial intelligence. Compared with existing technologies, this invention has the following technical advantages: This invention improves data purification quality and peak shape fidelity by constructing a dual-domain collaborative filtering mechanism of a physical domain baseline drift motion mechanics model and a data domain adaptive wavelet threshold decomposition, and by establishing a dual-domain joint loss function for joint optimization. This invention introduces multi-scale morphological features and mass spectrometry fragment ion matching degree as feature vector dimensions, and combines a two-level similarity verification mechanism of weighted Mahalanobis distance and cosine similarity, which can effectively distinguish substances with similar chromatographic conditions but different chemical structures in complex matrix samples. This invention maps chromatographic metadata into dynamic embedding vectors through a multi-scale metadata embedding mechanism, constructs a spatial-channel adaptive normalization module, and enables the normalization parameters to be adaptively adjusted according to the instrument status. This can meet the needs of gradient propagation in deep neural networks and improve the generalization ability of subsequent AI models. This invention achieves the upgrade of chromatographic mass spectrometry data from laboratory standards to AI-ready standards through a complete technology chain, from raw data acquisition, dual-domain collaborative purification, multi-level similarity component prediction, metadata embedding adaptive normalization to hybrid addressing storage. It provides a high-quality and standardized data governance solution for intelligent analysis in fields such as quality control of traditional Chinese medicine and food safety monitoring. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the steps of the chromatographic mass spectrometry data management method for artificial intelligence according to the present invention. Detailed Implementation

[0014] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0015] The present invention provides a method and system for managing chromatographic and mass spectrometry data based on artificial intelligence, comprising the following steps: like Figure 1 As shown, this embodiment includes the following steps: Acquire chromatographic and mass spectrometric data and chromatographic metadata of the submitted sample, perform dual-domain collaborative adaptive filtering on the sample chromatogram to obtain a purified sample chromatogram, and finally perform peak extraction and peak alignment operations. Match the standard chromatograms with the chromatographic conditions corresponding to the chromatograms of the purified samples, extract the chromatographic feature vectors and full-scan mass spectra of the purified sample chromatograms, and calculate the first similarity between the purified sample chromatograms and the standard chromatograms to obtain the standard screening chromatogram set. Extract the reference peaks and relative retention time standard values ​​of the standard screening chromatogram, calculate the relative retention time of each characteristic peak of the purified sample chromatogram, calculate the cosine similarity between the relative retention time standard value and the relative retention time to obtain the second similarity, and determine the chromatogram prediction information of the purified sample. Sample component prediction is obtained from the chromatogram prediction information of all purified samples, a sample chromatogram matrix is ​​generated, chromatographic metadata is embedded into the sample chromatogram matrix to obtain chromatographic mass spectrometry treatment data, and storage is performed according to the detection information and the storage address is generated. The chromatographic mass spectrometry data includes chromatograms, chromatographic conditions, and detection information; the chromatographic conditions include mobile phase composition, flow rate, detection wavelength, and column temperature; the detection information includes sampling time, sampling location, detection time, and detection location. The chromatographic metadata includes instrument drift coefficient, signal-to-noise ratio, and column efficiency parameters; The standard chromatogram is obtained from a specific standard substance under standard chromatographic conditions; The chromatographic feature vector includes basic chromatographic features and speciation chromatographic features; the basic chromatographic features include the number of characteristic peaks, the characteristic peak area vector, and the distribution density; the speciation chromatographic features include the peak sharpness index, the peak asymmetry factor, and the peak skewness coefficient. The sample chromatogram prediction information includes predicted raw materials and predicted content.

[0016] 4. In this embodiment, the method for performing dual-domain collaborative adaptive filtering includes: The chromatographic metadata is input as a priori initialization parameter into the dual-domain collaborative adaptive filtering module, and the sample chromatogram is input into the dual-domain collaborative adaptive filtering module for dual-domain collaborative adaptive filtering. The dual-domain collaborative adaptive filtering module includes a physical domain estimator and a data domain denoiser. The physical domain estimator initializes the decay rate using the chromatographic data instrument drift coefficient, constructs a baseline drift function, obtains an initial baseline estimate by least squares fitting of the physical domain parameter vector, and obtains the intermediate signal by subtracting the initial baseline estimate from the sample chromatogram. The baseline function expression is: ; in For baseline drift function, To retain the time, The number of baseline components, For physical domain parameter vectors, For the first Each index component amplitude, For the first An exponential decay rate, For periodic interference amplitude, For periodic interference angular frequency, This is the DC offset; The data domain denoiser performs multi-level stationary wavelet transform on the intermediate signal to obtain approximation coefficients and detail coefficients. Based on the second derivative of the chromatographic peak and the signal-to-noise ratio of the chromatographic data, it calculates the peak density index and adaptive threshold. The detail coefficients are adjusted according to the adaptive threshold to obtain the denoised detail coefficients. Then, the inverse wavelet transform is performed to reconstruct the data domain denoised signal. The expressions for calculating the peak density index and the adaptive threshold are as follows: ; ; in for Layer adaptive threshold, for Standard deviation of layer detail factor For signal length, For signal-to-noise ratio, The peak protection strength coefficient, This is the peak density attenuation coefficient. for Layer adaptive threshold, This is an intermediate signal; The method for calculating the denoising detail coefficients by adjusting the detail coefficients based on the adaptive threshold is as follows: ; in for Layer denoising detail coefficients, for Layer detail factor; The dual-domain collaborative adaptive filtering module uses a dual-domain joint loss function to optimize the adaptive threshold and physical domain parameter vector, ultimately outputting the chromatogram of the purified sample, expressed as follows: ; ; in For the joint loss function of two domains, For an adaptive threshold set, This is the original signal, corresponding to the sample chromatogram. For the current iteration baseline estimate, For the current iteration signal estimation, For physical domain regularization weights, For data domain regularization weights, For baseline smoothing regularization, The wavelet decomposition level is denoted as . This is the filtered signal, corresponding to the chromatogram of the purified sample. For the optimal physical domain parameter vector The corresponding optimal baseline estimate; In actual evaluation, taking the chromatographic and mass spectrometric data processing of flavonoids and triterpenoid saponins in the extract of a certain Chinese medicinal herb licorice as an example, the two types of components have essential differences in polarity and ultraviolet absorption wavelength. Therefore, they are divided into two independent samples (sample 1 and sample 2), and optimized exclusive chromatographic conditions are used for determination and data processing respectively. Taking the chromatographic conditions of sample 1 (for the detection of flavonoid components) as an example: the detection system was a Waters ACQUITYUPLC H-Class system, the chromatographic column was a Waters BEH C18 (2.1×100 mm, 1.7 μm), the mobile phase was acetonitrile (A)-0.1% formic acid aqueous solution (B) gradient elution (0-3 min, 15%A; 3-10 min, 15%-35%A; 10-15 min, 35%-60%A), the flow rate was 0.3 mL / min, the column temperature was 35℃, the detection wavelength was 330 nm, and the injection volume was 2 μL; the corresponding chromatographic data were: instrument drift coefficient 0.002 min / h, signal-to-noise ratio 320, and theoretical plate number 18500; Chromatographic detection was performed to obtain the chromatogram of sample 1. The baseline component number was set to 3. The physical domain parameter vector was initialized in the physical domain estimator to obtain the initial baseline estimate and calculate the intermediate signal. The input data domain denoiser was subjected to a 4-level stationary wavelet transform to obtain the approximation coefficient and detail coefficient. The peak protection intensity coefficient and peak density attenuation coefficient were set to 0.12 / 0.88. The denoising detail coefficient was calculated and inverse wavelet transform was performed to reconstruct the data domain denoised signal. The physical domain parameter vector was iteratively optimized by the physical domain estimator and the data domain denoiser through the dual-domain joint loss function to minimize the dual-domain joint loss function and output the optimal physical domain parameter vector. The optimal physical domain parameter vector was substituted into the physical domain estimator to calculate the optimal baseline estimate. The optimal baseline estimate was subtracted from the sample chromatogram (original signal) to obtain the chromatogram of the flavonoid purified sample. The first derivative method was used to extract the main flavonoid chromatographic peaks, such as glycyrrhizin (retention time approximately 8.2 min) and isoglycyrrhizin (retention time approximately 9.5 min). The retention time of the chromatograms of three consecutive batches of sample A was aligned using the DTW algorithm to eliminate drift caused by minor fluctuations in the low flow rate system. For sample 2 (used for the detection of triterpenoid saponins), chromatographic conditions were determined and chromatographic detection was performed to obtain the chromatogram of sample 2. The above method was repeated to obtain the chromatogram of the purified sample of triterpenoid saponins.

[0017] In this embodiment, the method for obtaining a standard screening chromatogram set includes: Standard chromatograms are acquired, categorized, and stored according to substance type to establish a standard chromatogram library. Each standard chromatogram is associated with its corresponding quasi-substance, standard chromatogram, and chromatographic metadata. First, match the standard chromatograms in the standard chromatogram library with mobile phase polarity parameters that deviate within ±15% and have the same elution mode as the chromatogram of the purified sample. Then, select any standard chromatogram with an instrument parameter deviation rate of no more than 20%. The instrument parameter deviation rate includes column size deviation rate, column temperature deviation rate, flow rate deviation rate, and detection wavelength deviation rate. The basic chromatographic features are directly extracted from the chromatogram of the purified sample. The speciation chromatographic features are then calculated based on the detail factor in the chromatogram of the purified sample. The expression is as follows: ; ; ; in for Stratification sharpness index. For peak asymmetry factor, This is the peak skewness coefficient. The number of characteristic peaks, To preserve the time variable, for Characteristic peak retention time interval For stationary wavelet transform Layer detail factor, for layer The characteristic peak width at 5% is shown. for layer The distance of the leading edge of the characteristic peak, for Retention time of characteristic peaks; The full-scan mass spectra corresponding to each characteristic peak of the purified sample chromatogram are extracted and compared with the mass spectra of the corresponding standard substances in the standard chromatogram to calculate the mass spectrometry fragment ion matching degree. Chromatographic features are composed of basic chromatographic features and speciation chromatographic features. The feature matching degree of the chromatographic feature vector between the purified sample chromatogram and the candidate standard chromatogram is calculated. The first similarity is obtained by weighting the mass spectrometry fragment ion matching degree and feature matching degree, expressed as: ; ; in Chromatogram of purified sample and First similarity of standard chromatograms To purify the sample chromatogram and Mass spectrometric fragment ion matching degree of standard chromatograms. , For similarity weights, This is the feature vector of the sample chromatogram. for The characteristic vector of a standard chromatogram For the feature covariance matrix of the training set, This represents the upper limit of the mass spectrometry scanning range. To purify the sample chromatogram Mass-to-charge ratio ionic strength For the first Weighting factors for the mass-to-charge ratio, for Standard chromatogram No. Individual mass-to-charge ratio ionic strength; A standard screening chromatogram set is constructed by selecting standard chromatograms that meet the first similarity threshold. In actual evaluation, taking the standard chromatogram matching of the flavonoid purification sample chromatogram of Sample 1 as an example, the same standard chromatogram is directly matched according to the mobile phase polarity parameter (deviation less than 15%) and elution mode. Further screening is carried out according to the instrument parameter deviation rate, and three sets of standard chromatogram chromatogram chromatogram matching examples are provided: Case 1 (standard substance: quercetin) detection wavelength deviation rate is 23% > 20% / matching failed; Case 2 (standard substance: glycyrrhizin) all instrument parameter deviation rate ≤ 20% / matching successful; Case 3 (standard substance: baicalin) all instrument parameter deviation rate ≤ 20% / matching successful. Before calculating the first similarity, the purified chromatogram and the standard chromatogram are peak-aligned (the sample contains more components, so the purified chromatogram is longer than the standard chromatogram). A purified chromatogram segment is then cut according to the length of the standard chromatogram (the purified chromatogram segment corresponding to each standard chromatogram is different; the subsequent second similarity calculation also uses the corresponding purified chromatogram segment). The chromatographic feature vector (normalized number of characteristic peaks / distribution density / average peak area, first / second / third layer sharpness, first / second / third layer asymmetry factor, first / second / third layer skewness coefficient) of the purified chromatogram segment of sample 1 is extracted as [0.75, 0.8, 0.713, 1.28, 0.95, 0.72, 1.05, 1]. [1, 1.08, 0.18, 0.25, 0.2], taking the first similarity weight coefficient as 0.6 / 0.4, the first similarity of the chromatographic feature vector between the standard chromatogram and the purified chromatogram of Case 2 is calculated to be 0.908 (Madara distance index 0.88, mass spectrometry fragment ion matching degree 0.95), which is greater than the first similarity threshold of 0.75. The standard chromatogram of Case 2 is included in the standard screening chromatogram set. The first similarity of the chromatographic feature vector between the standard chromatogram and the purified chromatogram of Case 3 is calculated to be 0.59 (Madara distance index 0.75, mass spectrometry fragment ion matching degree 0.35, the sample does not contain baicalin, so the feature ion matching degree is low), which is less than the first similarity threshold of 0.75. Case 3 matching fails. Similarly, repeat the above steps to match the remaining standard chromatograms according to the chromatographic conditions to obtain a standard screening chromatogram set.

[0018] In this embodiment, the method for determining sample chromatogram prediction information includes: Extract the reference peaks and corresponding relative retention time values ​​of each standard screening chromatogram. Calculate the relative retention time of each characteristic peak based on the reference peaks of the purified sample chromatogram. Calculate the cosine similarity between the vector of relative retention time values ​​of the standard screening chromatogram and the vector of relative retention time corresponding to each characteristic peak of the purified sample chromatogram as the second similarity. Select standard screening chromatograms with a second similarity greater than the second similarity threshold as sample component chromatograms. Use the standard substances corresponding to the sample component chromatograms as predictive raw materials. The predicted content of the corresponding standard substance is calculated based on the characteristic peak areas and detection solution preparation parameters in the chromatograms of the purified sample and the chromatograms of the sample components. In the actual evaluation, taking the determination of the chromatogram prediction information of the flavonoid purification sample of Sample 1 as an example, the second similarity verification and content prediction were carried out on the standard screening chromatogram of Case 2 (standard substance: glycyrrhizin) that passed the first similarity screening. According to the characteristic chromatogram requirements of licorice formula granules in the Chinese Pharmacopoeia, the reference peak (the 3rd peak, the characteristic peak of glycyrrhizin) and the specified values ​​of relative retention time (0.83, 0.98, 1, 1.16, 1.) of the standard screening chromatogram of Case 2 (standard substance: glycyrrhizin) were extracted. 23, 1.42, 1.55, 1.68), using the actually detected glycyrrhizin peak as a reference peak in the chromatogram segment of the purified sample, the relative retention times of each characteristic peak were calculated (0.89, 0.91, 1, 1.21, 1.17, 1.38, 1.68, 1.58). Based on this, the second similarity was calculated to be 0.972 (greater than the second similarity threshold of 0.85), indicating the presence of glycyrrhizin in the sample. According to the chromatographic identification results, the glycyrrhizin content was determined to be 2.35%. The specific calculation is expressed as follows: ; in This refers to the content of glycyrrhizin. This represents the peak area of ​​glycyrrhizin in the sample solution. The peak area of ​​the standard substance (glycyrrhizin) during the standard screening chromatogram determination. The concentration of the sample solution. The mass of the standard substance to be weighed during the standard screening chromatogram determination. This is the final volume of the sample solution; Similarly, repeat the above operation to perform a second similarity verification and content prediction on the remaining standard screening chromatograms matched by sample 1 and the standard screening chromatograms matched by sample 2, and obtain the predicted components of the submitted sample: glycyrrhizin 2.35%, isoglycyrrhizin 1.12%, glycyrrhizic acid 3.56%.

[0019] In this embodiment, the method for obtaining chromatography-mass spectrometry treatment data includes: The chromatograms of each purified sample of the same sample were normalized in the time-mass-charge ratio dimension to obtain a single-sample tensor. The sample component prediction was then encoded and spatiotemporally mapped. The component tensor is obtained by concatenating the single sample tensor with the component tensor, and the sample spectral matrix is ​​obtained by stacking all sub-samples in the same batch; the same batch includes sub-samples of the same sample, the same sampling time, and different chromatographic conditions. The chromatographic metadata is subjected to linear and nonlinear projections respectively, and the projection results are fused to obtain a multi-scale metadata embedding vector, expressed as: ; in For scale metadata embedding vectors, As a linear projection matrix, the chromatographic metadata is... Mapped to the chromatogram channel dimension, As the first nonlinear projection matrix, the chromatographic data... Mapping to hidden dimensions to extract non-linear features. The second nonlinear projection matrix maps the hidden dimensions back to the chromatogram channel dimensions; The mean and variance of the sample spectral matrix are calculated along the batch dimension. Multi-scale metadata is embedded into a vector input to a lightweight fully connected network to generate spatially and channel-adaptive scaling and offset parameters. Finally, adaptive normalization is performed on the batch spectral matrix to obtain chromatographic and mass spectrometric treatment data. The expression is: ; ; in For chromatographic and mass spectrometric processing data, , Spatial-channel adaptive scaling and offset parameters, This is the sample spectrum matrix. , The mean and variance of the sample spectrum matrix are given. It is the numerical stability constant. Generate a weight matrix for the affine parameters. Generate a bias vector for the affine parameters; In practical evaluation, firstly, the chromatograms of each purified sample are standardized in the time-mass-charge ratio dimension: the chromatograms are uniformly resampled along the retention time axis to a fixed length T (e.g., 1024 time points), and the mass spectrometry data are resampled along the m / z axis to a fixed dimension M (e.g., 512 mass-charge ratio channels), forming a single-sample tensor. ; Subsequently, the sample component prediction encoding (including predicted raw material identification and predicted content) is used to construct a component prediction feature vector. This prediction feature vector is then mapped to a component tensor that matches the spatiotemporal dimension of the spectrum through a learnable embedding layer. quantity, This represents the number of channels in the original sample spectrum. To determine the number of channels for component prediction encoding, the single-sample tensor is concatenated with the component tensor to obtain the enhanced spectrum tensor. The sample spectral matrix is ​​obtained by stacking all sub-samples from the same batch. , Batch size (e.g.) (This represents two sub-samples: flavonoids and saponins). This represents the total number of channels; Chromatography metadata shallow linear features are obtained by inputting a linear projection layer. This is used to directly transmit the raw scale information of metadata (such as specific SNR values) to ensure that basic instrument parameters are not lost; and to transmit chromatographic metadata. Input the first nonlinear projection matrix to obtain the nonlinear features of the hidden layer The ReLU activation function introduces nonlinearity to characterize the complex interactions of metadata (such as the synergistic effect of high SNR and low drift). Then, the nonlinear features of the hidden layer are input into the second nonlinear projection matrix to obtain the deep nonlinear features. The hidden layer nonlinear features are mapped back to the chromatogram channel dimension, and higher-order combination patterns of metadata are learned (such as the combined effect of high SNR and low column efficiency), enhancing the model's ability to perceive complex instrument conditions. and The output dimensions are all ; Affine parameter generation weight matrix will Dimensional embedding mapping to The affine parameters generate the bias vector. The scaling and offset parameters of each channel provide the initial offset. The scaling parameter acts as a multiplicative gain, which dynamically adjusts the magnification / scaling of each channel based on metadata (such as SNR). The offset parameter acts as an additive bias, which dynamically adjusts the reference value of each channel based on metadata.

[0020] In this embodiment, the method for generating storage addresses includes: The chromatographic mass spectrometry processing data and the corresponding chromatographic mass spectrometry data and chromatographic metadata are encapsulated into a structured data package; A basic path is generated based on the sample batch number and sampling time and location from the testing information. A content identifier is generated based on the chromatographic mass spectrometry processing data and component prediction results. The storage address is obtained using the basic path and content identifier, and the structured data package is stored according to the storage address. The expression is: ; ; ; in For storage address, Basic path, For content identification, For sampling locations, For sampling time, For batch number, for Chromatographic and mass spectrometric processing data of the sample, For component prediction feature vectors, For hash functions, It uses the SHA-256 secure hash algorithm. For Merkel root hash; In actual assessments, within the basic pathway, This indicates that the sampling location is encoded using a hash function, and the first 8 characters are used in the content identifier. This represents a set of hash values ​​from chromatographic and mass spectrometric data processing of samples. In the case of licorice extract, it includes flavonoids and triterpenoid saponins, i.e., according to batch dimension. Generate hash values ​​for the chromatogram and mass spectrometry processed data of Sample 1 and Sample 2 (corresponding to two leaf nodes), and aggregate the hash values ​​of the two leaf nodes layer by layer to generate a Merkle root hash. This indicates that the first 16 characters of the component prediction vector are hashed using the SHA-256 secure hash algorithm. The licorice extract (batch number GL-20260328-A) sampled in Chifeng, Inner Mongolia on March 28, 2026, was divided into two samples (Sample 1 for detecting flavonoids and Sample 2 for detecting triterpenoid saponins). The base path "2026-03-28 / a3f9b2d1 / GL-20260328-A / " was generated. The SHA-256 hashes of the chromatographic and mass spectrometric data of both samples were calculated, and a Merkle tree was constructed to calculate the root hash as "7e8c2f4a...". The SHA-256 hash of the component prediction feature vector is calculated, and the first 16 bits "b8d9e2f1a4c5d6e7" are taken to generate the content identifier "7e8c2f4a... / b8d9e2f1a4c5d6e7". Finally, the storage address "2026-03-28 / a3f9b2d1 / GL-20260328-A / 7e8c2f4a... / b8d9e2f1a4c5d6e7" is generated. The encapsulated structured data packet is stored according to this address.

[0021] Secondly, the chromatographic mass spectrometry data management system oriented towards artificial intelligence includes: Chromatography purification module: used to acquire chromatographic and mass spectrometric data and chromatographic metadata of the submitted sample, perform dual-domain collaborative adaptive filtering on the sample chromatogram to obtain a purified sample chromatogram, and finally perform peak extraction and peak alignment operations; Standard substance screening module: It is used to match each standard chromatogram with the chromatographic conditions corresponding to the chromatogram of the purified sample, extract the chromatographic feature vector and full scan mass spectrum of the purified sample chromatogram, and calculate the first similarity between the purified sample chromatogram and the standard chromatogram to obtain the standard screening chromatogram set. Component prediction module: used to extract reference peaks and relative retention time standard values ​​of standard screening chromatograms, calculate the relative retention time of each characteristic peak of the purified sample chromatogram, calculate the cosine similarity between the relative retention time standard value and the relative retention time to obtain the second similarity, and determine the chromatogram prediction information of the purified sample. The sample component prediction is obtained from the chromatogram prediction information of all purified samples. Data governance module: used to generate sample spectral matrix, embed chromatographic metadata into the sample spectral matrix to obtain chromatographic mass spectrometry governance data, and generate storage addresses according to the detection information for storage.

[0022] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for managing chromatographic and mass spectrometric data based on artificial intelligence, characterized in that, Includes the following steps: S1. Obtain the chromatographic mass spectrometry data and chromatographic metadata of the submitted sample, perform dual-domain collaborative adaptive filtering on the sample chromatogram to obtain the purified sample chromatogram, and finally perform peak extraction and peak alignment operations. S2. Match each standard chromatogram with the chromatographic conditions corresponding to the chromatogram of the purified sample, extract the chromatographic feature vector and full scan mass spectrum of the purified sample chromatogram, and calculate the first similarity between the purified sample chromatogram and the standard chromatogram to obtain the standard screening chromatogram set. S3. Extract the reference peaks and relative retention time standard values ​​of the standard screening chromatogram, calculate the relative retention time of each characteristic peak of the purified sample chromatogram, calculate the cosine similarity between the relative retention time standard value and the relative retention time to obtain the second similarity, and determine the chromatogram prediction information of the purified sample. S4. Obtain sample component prediction from the chromatogram prediction information of all purified samples, generate a sample chromatogram matrix, embed chromatographic metadata into the sample chromatogram matrix to obtain chromatographic mass spectrometry treatment data, and store it according to the storage address generated by the detection information. The chromatographic mass spectrometry data includes chromatograms, chromatographic conditions, and detection information; the chromatographic conditions include mobile phase composition, flow rate, detection wavelength, and column temperature; the detection information includes sampling time, sampling location, detection time, and detection location. The chromatographic metadata includes instrument drift coefficient, signal-to-noise ratio, and column efficiency parameters; The standard chromatogram is obtained from a specific standard substance under standard chromatographic conditions; The chromatographic feature vector includes basic chromatographic features and speciation chromatographic features; the basic chromatographic features include the number of characteristic peaks, the characteristic peak area vector, and the distribution density; the speciation chromatographic features include the peak sharpness index, the peak asymmetry factor, and the peak skewness coefficient. The sample chromatogram prediction information includes predicted raw materials and predicted content.

2. The method for managing chromatographic and mass spectrometry data based on artificial intelligence according to claim 1, characterized in that, The method for performing dual-domain collaborative adaptive filtering includes: The chromatographic metadata is input as a priori initialization parameter into the dual-domain collaborative adaptive filtering module, and the sample chromatogram is input into the dual-domain collaborative adaptive filtering module for dual-domain collaborative adaptive filtering. The dual-domain collaborative adaptive filtering module includes a physical domain estimator and a data domain denoiser. The physical domain estimator initializes the decay rate using the chromatographic data instrument drift coefficient, constructs a baseline drift function, obtains an initial baseline estimate by least squares fitting of the physical domain parameter vector, and obtains the intermediate signal by subtracting the initial baseline estimate from the sample chromatogram. The baseline function expression is: ； in For baseline drift function, To retain the time, The number of baseline components, For physical domain parameter vectors, For the first Each index component amplitude, For the first An exponential decay rate, For periodic interference amplitude, For periodic interference angular frequency, This is the DC offset; The data domain denoiser performs multi-level stationary wavelet transform on the intermediate signal to obtain approximation coefficients and detail coefficients. Based on the second derivative of the chromatographic peak and the signal-to-noise ratio of the chromatographic data, it calculates the peak density index and adaptive threshold. The detail coefficients are adjusted according to the adaptive threshold to obtain the denoised detail coefficients. Then, the inverse wavelet transform is performed to reconstruct the data domain denoised signal. The expressions for calculating the peak density index and the adaptive threshold are as follows: ； ； in for Layer adaptive threshold, for Standard deviation of layer detail factor For signal length, For signal-to-noise ratio, The peak protection strength coefficient, This is the peak density attenuation coefficient. for Layer adaptive threshold, This is an intermediate signal; The method for calculating the denoising detail coefficients by adjusting the detail coefficients based on the adaptive threshold is as follows: ； in for Layer denoising detail coefficients, for Layer detail factor; The dual-domain collaborative adaptive filtering module uses a dual-domain joint loss function to optimize the adaptive threshold and physical domain parameter vector, ultimately outputting the chromatogram of the purified sample, expressed as follows: ； ； in For the joint loss function of two domains, For an adaptive threshold set, This is the original signal, corresponding to the sample chromatogram. For the current iteration baseline estimate, For the current iteration signal estimation, For physical domain regularization weights, For data domain regularization weights, For baseline smoothing regularization, The wavelet decomposition level is denoted as . This is the filtered signal, corresponding to the chromatogram of the purified sample. For the optimal physical domain parameter vector The corresponding optimal baseline estimate.

3. The method for managing chromatographic and mass spectrometry data based on artificial intelligence according to claim 1, characterized in that, The method for obtaining a standard screening chromatogram set includes: Standard chromatograms are acquired, categorized, and stored according to substance type to establish a standard chromatogram library. Each standard chromatogram is associated with its corresponding quasi-substance, standard chromatogram, and chromatographic metadata. First, match the standard chromatograms in the standard chromatogram library with mobile phase polarity parameters that deviate within ±15% and have the same elution mode as the chromatogram of the purified sample. Then, select any standard chromatogram with an instrument parameter deviation rate of no more than 20%. The instrument parameter deviation rate includes column size deviation rate, column temperature deviation rate, flow rate deviation rate, and detection wavelength deviation rate. The basic chromatographic features are directly extracted from the chromatogram of the purified sample. The speciation chromatographic features are then calculated based on the detail factor in the chromatogram of the purified sample. The expression is as follows: ； ； ； in for Stratification sharpness index. For peak asymmetry factor, This is the peak skewness coefficient. The number of characteristic peaks, To preserve the time variable, for Characteristic peak retention time interval For stationary wavelet transform Layer detail factor, for layer The characteristic peak width at 5% is shown. for layer The distance of the leading edge of the characteristic peak, for Retention time of characteristic peaks; The full-scan mass spectra corresponding to each characteristic peak of the purified sample chromatogram are extracted and compared with the mass spectra of the corresponding standard substances in the standard chromatogram to calculate the mass spectrometry fragment ion matching degree. Chromatographic features are composed of basic chromatographic features and speciation chromatographic features. The feature matching degree of the chromatographic feature vector between the purified sample chromatogram and the candidate standard chromatogram is calculated. The first similarity is obtained by weighting the mass spectrometry fragment ion matching degree and feature matching degree, expressed as: ； ； in Chromatogram of purified sample and First similarity of standard chromatograms To purify the sample chromatogram and Mass spectrometric fragment ion matching degree of standard chromatograms. , For similarity weights, This is the feature vector of the sample chromatogram. for The characteristic vector of a standard chromatogram For the feature covariance matrix of the training set, This represents the upper limit of the mass spectrometry scanning range. To purify the sample chromatogram Mass-to-charge ratio ionic strength For the first Weighting factors for the mass-to-charge ratio, for Standard chromatogram No. Individual mass-to-charge ratio ionic strength; A standard screening chromatogram set is constructed by selecting standard chromatograms that meet the first similarity threshold.

4. The method for managing chromatographic and mass spectrometry data oriented towards artificial intelligence according to claim 1, characterized in that, The method for determining sample chromatogram prediction information includes: Extract the reference peaks and corresponding relative retention time values ​​of each standard screening chromatogram. Calculate the relative retention time of each characteristic peak based on the reference peaks of the purified sample chromatogram. Calculate the cosine similarity between the vector of relative retention time values ​​of the standard screening chromatogram and the vector of relative retention time corresponding to each characteristic peak of the purified sample chromatogram as the second similarity. Select standard screening chromatograms with a second similarity greater than the second similarity threshold as sample component chromatograms. Use the standard substances corresponding to the sample component chromatograms as predictive raw materials. The predicted content of the corresponding standard substance is calculated based on the characteristic peak areas and detection solution preparation parameters in the chromatograms of the purified sample and the chromatograms of the sample components.

5. The method for managing chromatographic and mass spectrometry data oriented towards artificial intelligence according to claim 1, characterized in that, The method for obtaining chromatography-mass spectrometry processing data includes: The chromatograms of each purified sample of the same sample are normalized in the time-mass-charge ratio dimension to obtain a single sample tensor. The sample component prediction is encoded and spatiotemporally mapped to obtain a component tensor. The single sample tensor and the component tensor are concatenated to obtain an enhanced chromatogram tensor. All sub-samples of the same batch are stacked to obtain a sample chromatogram matrix. The same batch includes sub-samples of the same sample, the same sampling time, and different chromatographic conditions. The chromatographic metadata is subjected to linear and nonlinear projections respectively, and the projection results are fused to obtain a multi-scale metadata embedding vector, expressed as: ； in For scale metadata embedding vectors, As a linear projection matrix, the chromatographic metadata is... Mapped to the chromatogram channel dimension, As the first nonlinear projection matrix, the chromatographic data... Mapping to hidden dimensions to extract non-linear features. The second nonlinear projection matrix maps the hidden dimensions back to the chromatogram channel dimensions; The mean and variance of the sample spectral matrix are calculated along the batch dimension. Multi-scale metadata is embedded into a vector input to a lightweight fully connected network to generate spatially and channel-adaptive scaling and offset parameters. Finally, adaptive normalization is performed on the batch spectral matrix to obtain chromatographic and mass spectrometric treatment data. The expression is: ； ； in For chromatographic and mass spectrometric processing data, , Spatial-channel adaptive scaling and offset parameters, This is the sample spectrum matrix. , The mean and variance of the sample spectrum matrix are given. It is the numerical stability constant. Generate a weight matrix for the affine parameters. Generate a bias vector for the affine parameters.

6. The method for managing chromatographic and mass spectrometry data oriented towards artificial intelligence according to claim 1, characterized in that, The method for generating storage addresses includes: The chromatographic mass spectrometry processing data and the corresponding chromatographic mass spectrometry data and chromatographic metadata are encapsulated into a structured data package; A basic path is generated based on the sample batch number and sampling time and location from the testing information. A content identifier is generated based on the chromatographic mass spectrometry processing data and component prediction results. The storage address is obtained using the basic path and content identifier, and the structured data package is stored according to the storage address. The expression is: ； ； ； in For storage address, Basic path, For content identification, For sampling locations, For sampling time, For batch number, for Chromatographic and mass spectrometric processing data of the sample, For component prediction feature vectors, For hash functions, It uses the SHA-256 secure hash algorithm. This is a Merkel tree root hash.

7. A chromatography-mass spectrometry data management system for artificial intelligence, used to perform the method according to any one of claims 1-6, characterized in that, include: Chromatography purification module: used to acquire chromatographic and mass spectrometric data and chromatographic metadata of the submitted sample, perform dual-domain collaborative adaptive filtering on the sample chromatogram to obtain a purified sample chromatogram, and finally perform peak extraction and peak alignment operations; Standard substance screening module: It is used to match each standard chromatogram with the chromatographic conditions corresponding to the chromatogram of the purified sample, extract the chromatographic feature vector and full scan mass spectrum of the purified sample chromatogram, and calculate the first similarity between the purified sample chromatogram and the standard chromatogram to obtain the standard screening chromatogram set. Component prediction module: used to extract reference peaks and relative retention time standard values ​​of standard screening chromatograms, calculate the relative retention time of each characteristic peak of the purified sample chromatogram, calculate the cosine similarity between the relative retention time standard value and the relative retention time to obtain the second similarity, and determine the chromatogram prediction information of the purified sample. The sample component prediction is obtained from the chromatogram prediction information of all purified samples. Data governance module: used to generate sample spectral matrix, embed chromatographic metadata into the sample spectral matrix to obtain chromatographic mass spectrometry governance data, and generate storage addresses according to the detection information for storage.

Images