Method for detecting folic acid metabolites of red blood cells by liquid chromatography-mass spectrometry tandem technology

Abstract

The invention relates to the technical field of liquid chromatography-mass spectrometry series connection, and particularly discloses a method for detecting folic acid metabolites of red blood cells through the liquid chromatography-mass spectrometry series connection technology. And an image analysis algorithm based on artificial intelligence and deep learning is utilized to extract a chromatographic separation state, so that gradient parameters are dynamically adjusted, the requirement of manual intervention is reduced, and the separation process of a target object in a complex matrix is favorably optimized. Particularly, key features in a chromatogram can be automatically identified and quantified based on a deep learning method, and gradient elution parameters are intelligently and dynamically adjusted according to the information, so that more optimized substance separation and detection can be realized. For example, if separation of a certain component is not ideal, the separation effect of the component can be improved by adjusting the proportion of the mobile phase B, so that the gradient elution process is optimized.

Description

Technical Field

[0001] The present application relates to the technical field of liquid chromatography-mass spectrometry tandem technology, and more specifically, to a method for detecting red blood cell folate metabolites using liquid chromatography-mass spectrometry tandem technology. Background Art

[0002] As an essential B vitamin for the human body, the metabolic status of folic acid is closely related to a variety of diseases and pregnancy health. The level of folate in red blood cells is an important indicator for assessing the body's folate reserves, but because folate in red blood cells is mostly in the form of polyglutamic acid, it needs to be hydrolyzed and converted into monoglutamic acid for accurate detection. In the existing technology, high-performance liquid chromatography tandem mass spectrometry (HPLC-MS / MS) has become the mainstream method for detecting folic acid and its metabolites due to its high sensitivity and specificity.

[0003] Chinese patent CN116223693A provides a method for determining folic acid and its metabolites in red blood cells by high performance liquid chromatography tandem mass spectrometry, which can use HPLC-MS / MS analysis method to determine folic acid and its metabolites in red blood cells. However, the HPLC-MS / MS analysis method in the above patent adopts a fixed gradient elution program, which is difficult to adapt to the dynamic changes of interfering components in different sample matrices, resulting in unstable separation efficiency, redundant analysis time and other problems. Specifically, it can achieve multi-component synchronous detection by presetting gradient parameters based on experience, but when facing complex biological samples, fixed gradients may cause insufficient separation of target and interfering substances, and conditions need to be repeatedly optimized. In addition, existing dynamic gradient adjustment technologies rely more on artificial experience or simple feedback mechanisms, lack intelligent optimization capabilities based on real-time data, and are difficult to adaptively adjust the mobile phase ratio in a single run, limiting the analysis throughput and accuracy.

[0004] Therefore, an optimized liquid chromatography-mass spectrometry-tandem technique for detecting erythrocyte folate metabolites is desired. Summary of the invention

[0005] The present application provides a method for detecting red blood cell folate metabolites using liquid chromatography-mass spectrometry tandem technology, which can automatically identify and quantify key features in the chromatogram and intelligently and dynamically adjust gradient elution parameters based on this information, thereby facilitating more optimized substance separation and detection.

[0006] According to one aspect of the present application, a method for detecting erythrocyte folate metabolites by liquid chromatography-mass spectrometry tandem technology is provided, comprising: collecting a blood sample containing an EDTA anticoagulant, and subjecting the sample to centrifugation to remove plasma to obtain a sample to be tested; pre-treating the sample to be tested to obtain a supernatant to be injected; performing liquid chromatography analysis on the supernatant to be injected to determine the mobile phase ratio; performing gradient elution on the supernatant to be injected based on the mobile phase ratio to separate and remove interference groups in the sample matrix to obtain a target analyte solution; detecting the mass-to-charge ratio of the target analyte and its corresponding isotope internal standard in the target analyte solution by a mass spectrometer, and using the isotope internal standard method for quantitative determination of the content of erythrocyte folate metabolites; wherein, performing liquid chromatography analysis on the supernatant to be injected to determine the mobile phase ratio comprises: obtaining a liquid chromatogram of the supernatant to be injected; pre-treating the liquid chromatogram; and The invention relates to a liquid chromatogram method for extracting multi-dimensional features of the liquid chromatogram and obtaining shallow features of the liquid chromatogram and semantic features of the liquid chromatogram; performing liquid chromatogram multi-dimensional feature joint encoding on the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram to obtain a significant information interaction representation between the shallow and semantic features of the liquid chromatogram, including: performing core clue extraction on the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram respectively to obtain the semantic core clue encoding features of the liquid chromatogram and the shallow core clue encoding features of the liquid chromatogram; based on the semantic core clue encoding features of the liquid chromatogram and the shallow core clue encoding features of the liquid chromatogram, performing fine-grained feature interaction analysis on the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram to obtain a significant information interaction representation between the shallow and semantic features of the liquid chromatogram; and determining the mobile phase ratio based on the significant information interaction representation between the shallow and semantic features of the liquid chromatogram.

[0007] In the above-mentioned method for detecting red blood cell folate metabolites using liquid chromatography-mass spectrometry tandem technology, the liquid chromatogram is preprocessed and multidimensional features of the liquid chromatogram are extracted to obtain shallow features of the liquid chromatogram and semantic features of the liquid chromatogram, including: denoising and baseline correction are performed on the liquid chromatogram to obtain a corrected liquid chromatogram; the corrected liquid chromatogram is passed through a liquid chromatography feature extractor based on a diaphanous space pyramid pooling module to obtain a liquid chromatogram shallow feature map as the liquid chromatogram shallow features and liquid chromatogram semantic feature map; the liquid chromatogram semantic feature map is passed through a liquid chromatogram semantic global perception enhancement module based on a fully connected layer to obtain a liquid chromatogram semantic feature vector as the liquid chromatogram semantic feature.

[0008] In the method for detecting red blood cell folate metabolites using the above-mentioned liquid chromatography-mass spectrometry tandem technology, core clues are extracted from the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram respectively to obtain the liquid chromatogram semantic core clue coding features and the liquid chromatogram shallow core clue coding features, including: performing core clue extraction based on point convolution coding on the semantic feature vector of the liquid chromatogram to obtain the liquid chromatogram semantic core clue coding vector as the liquid chromatogram semantic core clue coding feature; extracting the liquid chromatogram shallow core clue coding vector from the liquid chromatogram shallow feature map as the liquid chromatogram shallow core clue coding feature.

[0009] In the above-mentioned method for detecting red blood cell folate metabolites using liquid chromatography-mass spectrometry tandem technology, a shallow core clue coding vector of the liquid chromatogram is extracted from the shallow feature map of the liquid chromatogram as the shallow core clue coding feature of the liquid chromatogram, including: splitting the shallow feature map of the liquid chromatogram to obtain a sequence of shallow local feature maps of the liquid chromatogram; performing an attention-based significant area attention analysis on each shallow local feature map of the liquid chromatogram in the sequence of shallow local feature maps of the liquid chromatogram to obtain a sequence of significant area attention coefficients; based on the sequence of significant area attention coefficients, performing a core clue search on the shallow feature map of the liquid chromatogram to obtain the shallow core clue coding vector of the liquid chromatogram.

[0010] In the method for detecting red blood cell folate metabolites using the above-mentioned liquid chromatography-mass spectrometry tandem technology, based on the semantic core clue coding features of the liquid chromatogram and the shallow core clue coding features of the liquid chromatogram, fine-grained feature interaction analysis is performed on the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram to obtain a significant information interaction representation between the shallow and semantic features of the liquid chromatogram, including: based on the semantic core clue coding vector of the liquid chromatogram and the shallow core clue coding vector of the liquid chromatogram, a core clue weaving template matrix between the shallow and semantic features of the liquid chromatogram is constructed; based on the core clue weaving template matrix between the shallow and semantic features of the liquid chromatogram, a cross-modal fine-grained feature interaction analysis is performed on the semantic feature vector of the liquid chromatogram and the shallow feature map of the liquid chromatogram to obtain a significant information interaction coding vector between the shallow and semantic features of the liquid chromatogram as a significant information interaction representation between the shallow and semantic features of the liquid chromatogram.

[0011] In the method for detecting red blood cell folate metabolites using the above-mentioned liquid chromatography-mass spectrometry tandem technology, a core clue weaving template matrix between shallow and semantic features of the liquid chromatogram is constructed based on the semantic core clue coding vector of the liquid chromatogram and the shallow core clue coding vector of the liquid chromatogram, including: constructing an inter-modal core clue weaving template between the semantic core clue coding vector of the liquid chromatogram and the shallow core clue coding vector of the liquid chromatogram to obtain the core clue weaving template matrix between shallow and semantic features of the liquid chromatogram.

[0012] In the method for detecting red blood cell folate metabolites using the above-mentioned liquid chromatography-mass spectrometry tandem technology, an inter-modal core clue weaving template between the semantic core clue coding vector of the liquid chromatogram and the shallow core clue coding vector of the liquid chromatogram is constructed to obtain the core clue weaving template matrix between the shallow layer and semantic features of the liquid chromatogram, including: using a feature mapping function to process the semantic core clue coding vector of the liquid chromatogram and the shallow core clue coding vector of the liquid chromatogram to obtain a core association matrix between the shallow layer and semantic features of the liquid chromatogram; and performing semantic alignment compensation on the core association matrix between the shallow layer and semantic features of the liquid chromatogram to obtain the core clue weaving template matrix between the shallow layer and semantic features of the liquid chromatogram.

[0013] In the above-mentioned method for detecting red blood cell folate metabolites using liquid chromatography-mass spectrometry tandem technology, semantic alignment compensation is performed on the core association matrix between shallow and semantic features of the liquid chromatogram to obtain the core clue weaving template matrix between shallow and semantic features of the liquid chromatogram, including: using the residual mechanism and the feature mapping function to perform structured semantic alignment on the semantic core clue encoding vector of the liquid chromatogram and the shallow core clue encoding vector of the liquid chromatogram to obtain the liquid chromatogram semantic alignment vector and the liquid chromatogram shallow alignment vector; calculating the explicit correlation matrix between shallow and semantic features of the liquid chromatogram between the semantic alignment vector of the liquid chromatogram and the shallow alignment vector of the liquid chromatogram; based on the explicit correlation matrix between shallow and semantic features of the liquid chromatogram, significant compensation is performed on the core association matrix between shallow and semantic features of the liquid chromatogram to obtain the core clue weaving template matrix between shallow and semantic features of the liquid chromatogram.

[0014] A method for detecting erythrocyte folate metabolites by a liquid chromatography-mass spectrometry tandem technology provided in the present application can monitor the liquid chromatogram to be injected by the supernatant in real time, and use an image analysis algorithm based on artificial intelligence and deep learning to extract the chromatographic separation state, so as to dynamically adjust the gradient parameters and reduce the need for manual intervention, thereby helping to optimize the separation process of the target in a complex matrix. In particular, the method based on deep learning can automatically identify and quantify the key features in the chromatogram, and intelligently and dynamically adjust the gradient elution parameters based on this information, so as to facilitate the realization of more optimized material separation and detection. For example, if it is found that the separation of a certain component is not ideal, the separation effect can be improved by adjusting the ratio of mobile phase B, thereby optimizing the gradient elution process. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0016] Figure 1 This is a schematic flow chart of a method for detecting erythrocyte folate metabolites using liquid chromatography-mass spectrometry tandem technology according to an embodiment of the present application.

[0017] Figure 2 This is a schematic flow chart of step S3 in the method for detecting red blood cell folate metabolites using liquid chromatography-mass spectrometry tandem technology according to an embodiment of the present application.

[0018] Figure 3 This is a schematic flow chart of step S32 in the method for detecting red blood cell folate metabolites using liquid chromatography-mass spectrometry tandem technology according to an embodiment of the present application.

[0019] Figure 4 This is a schematic flow chart of step S33 in the method for detecting red blood cell folate metabolites using liquid chromatography-mass spectrometry tandem technology according to an embodiment of the present application.

[0020] Figure 5 This is a schematic flow chart of step S331 in the method for detecting red blood cell folate metabolites using liquid chromatography-mass spectrometry tandem technology according to an embodiment of the present application.

[0021] Figure 6 This is a schematic flow chart of step S332 in the method for detecting red blood cell folate metabolites using liquid chromatography-mass spectrometry tandem technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0023] With the increasing application of machine learning algorithms in analytical chemistry, it is possible to optimize the chromatographic separation process using these advanced computational tools. However, existing research focuses on offline data modeling, and has not yet achieved online monitoring and control optimization with HPLC-MS / MS systems. In addition, the ability to jointly analyze multi-dimensional chromatographic features (such as peak shape, retention time drift, and co-elution interference) is insufficient, which limits its application in actual detection.

[0024] Based on this, in the technical solution of the present application, a method for detecting red blood cell folate metabolites by liquid chromatography-mass spectrometry tandem technology is proposed, such as Figure 1 As shown, the method for detecting red blood cell folate metabolites by liquid chromatography-mass spectrometry tandem technology comprises the following steps: S1, collecting a blood sample containing an EDTA anticoagulant, and subjecting the sample to centrifugal separation to remove plasma to obtain a sample to be tested; S2, pre-treating the sample to be tested to obtain a supernatant to be injected; S3, performing liquid chromatography analysis on the supernatant to be injected to determine the mobile phase ratio; S4, performing gradient elution on the supernatant to be injected based on the mobile phase ratio to separate and remove interference groups in the sample matrix to obtain a target analyte solution; S5, detecting the mass-to-charge ratio of the target analyte and its corresponding isotope internal standard in the target analyte solution by a mass spectrometer, and using the isotope internal standard method for quantitative determination of the content of red blood cell folate metabolites.

[0025] Exemplarily, in step S1, a blood sample containing an EDTA anticoagulant is collected, and centrifuged to remove the plasma to obtain a sample to be tested. It should be understood that the blood sample containing an EDTA anticoagulant is selected to prevent the blood from coagulating during the collection and processing process. EDTA is an effective chelating agent that can bind calcium ions, thereby preventing the coagulation factors in the blood from functioning, ensuring that the blood remains in a flowing state for subsequent processing. Once the blood sample containing an EDTA anticoagulant is collected, the plasma needs to be removed by centrifugation to obtain a red blood cell sample. This process is necessary because it is necessary to specifically detect folic acid and its metabolites in red blood cells. Red blood cells are the main place in the human body to store folic acid. They contain a large amount of polyglutamic acid forms of folic acid, and these forms must be converted into monoglutamic acid forms to be accurately detected. Therefore, in order to ensure the accuracy of the analysis results, red blood cells must be separated from whole blood and further processed.

[0026] In a specific example of the present application, the collected blood sample is placed in a high-speed centrifuge and centrifuged under specific speed and time conditions. For example, commonly used conditions may be 3000 to 5000 revolutions per minute (rpm) for 10 to 20 minutes. This centrifugation process causes the blood components to be stratified according to density differences: the lighter plasma is located on the upper layer, while the heavier red blood cells are deposited at the bottom. By carefully removing the upper layer of plasma, a pure red blood cell sample can be obtained. It is worth noting that special care should be taken when removing plasma to avoid mixing in any residual plasma or other components such as white blood cells, so as not to affect the accuracy of subsequent experiments.

[0027] Exemplarily, in step S2, the sample to be tested is pre-treated to obtain a supernatant to be injected. It should be understood that because the folic acid in red blood cells mainly exists in the form of polyglutamic acid, these forms need to be converted into monoglutamic acid forms to be accurately detected. In addition, the red blood cell sample may contain a variety of proteins and other impurities, which will interfere with subsequent chromatography and mass spectrometry analysis. Therefore, these interfering substances can be effectively removed by pre-treatment, and folic acid and its metabolites can be converted into a form suitable for detection.

[0028] In a specific example of the present application, after collecting a pure red blood cell sample, a mixed internal standard working solution needs to be added first. This internal standard solution contains isotope-labeled compounds with similar structures but different masses to the target analyte, such as folic acid-d4, 5-methyltetrahydrofolate-d4, etc. The addition of the internal standard helps to correct errors that may occur during sample preparation and analysis, thereby improving the accuracy of the quantitative results. Next, in order to protect folic acid and its metabolites from oxidative degradation, a protective agent, usually a 1% to 3% ascorbic acid (VC) solution, needs to be added. At the same time, in order to adjust the pH value to make it suitable for subsequent enzymatic hydrolysis reactions, an appropriate amount of sodium hydroxide (NaOH) solution needs to be added. These reagents work together to provide a stable chemical environment for subsequent steps. Subsequently, in order to convert the polyglutamic acid form of folic acid into the monoglutamic acid form, the red blood cell sample must be treated with a lysing enzyme. The specific method is to add an aqueous solution containing 0.3 to 1.0 units / ml of glutamyl hydrolase and vortex mix. Then, the mixture is incubated at 37°C in the dark for a period of time (e.g., 1 hour) to fully hydrolyze folic acid polyglutamate. After the enzymatic hydrolysis reaction is completed, a protein precipitant is added to precipitate the protein and further purify the sample. The commonly used precipitant is a methanol-acetonitrile mixed solution with a volume ratio of 2:1. After oscillation and mixing, high-speed centrifugation is performed (e.g., 14,800 rpm for 5 minutes). The supernatant after centrifugation is the required liquid to be injected, which contains folic acid and its metabolites that have been converted into monoglutamate, and most of the protein and other impurities have been removed.

[0029] Exemplarily, in step S3, the supernatant to be injected is subjected to liquid chromatography analysis to determine the mobile phase ratio. It should be understood that the optimization of the mobile phase ratio is crucial to achieving efficient chromatographic separation. Different compounds exhibit different retention behaviors under different mobile phase conditions, so selecting a suitable mobile phase ratio can significantly improve the separation effect of the target compound. Specifically, by adjusting the ratio of mobile phase A (usually an aqueous solution) and mobile phase B (usually an organic solvent such as methanol or acetonitrile), the retention time and elution order of the target compound in the chromatographic column can be controlled. This is particularly important for the simultaneous detection of multiple targets in a complex matrix, because different targets may have different polarities and chemical properties, and a specific mobile phase ratio is required to achieve the best separation effect. Secondly, liquid chromatography analysis helps to identify and quantify key features in the chromatogram, thereby providing a basis for dynamically adjusting gradient parameters. By real-time monitoring and analysis of the liquid chromatogram of the supernatant to be injected, multi-dimensional information such as peak shape, retention time drift, and co-elution interference can be obtained. This information can help researchers understand the separation effect under the current separation conditions and make adjustments according to actual conditions. For example, if the separation of a certain component is found to be unsatisfactory, the separation effect can be improved by adjusting the proportion of mobile phase B.

[0030] Exemplarily, in step S4, the supernatant to be injected is subjected to gradient elution based on the mobile phase ratio to separate and remove the interfering groups in the sample matrix to obtain the target analyte solution. It should be understood that, in addition to the target compound (such as folic acid and its metabolites), the red blood cell sample also contains a large number of other components, which may interfere with the detection of the target compound. For example, proteins, lipids and other small molecule metabolites may co-elute with the target compound, resulting in signal overlap or increased background noise, thereby affecting the detection results. Through gradient elution, the composition of the mobile phase can be gradually changed so that compounds of different polarities are eluted at different time points, thereby achieving effective separation. This method is particularly suitable for complex matrix samples because it can dynamically adjust the mobile phase conditions according to the characteristics of the target compound to optimize the separation effect.

[0031] Exemplarily, in step S5, the mass-to-charge ratio of the target analyte in the target analyte solution and its corresponding isotope internal standard is detected by a mass spectrometer, and the isotope internal standard method is used for quantitative determination of the content of erythrocyte folate metabolites. It should be understood that the mass spectrometer can provide extremely high sensitivity and specificity, which is crucial for the detection of low-concentration target compounds in complex biological matrices. In particular, in red blood cell samples, the concentration of folic acid and its metabolites is low, and there are a large number of interfering substances. The mass spectrometer can accurately identify and quantify the target compound by accurately measuring the mass-to-charge ratio (m / z). The mass spectrometer usually uses a multiple reaction monitoring (MRM) mode for positive ion scanning. In this mode, the instrument selects specific precursor ions and product ions for detection, thereby greatly improving the selectivity and sensitivity of the detection. This method is particularly suitable for the simultaneous detection of multiple compounds, such as folic acid, 5-methyltetrahydrofolate, 5-formyltetrahydrofolate, etc. In order to reduce the errors introduced during sample preparation and analysis, isotope-labeled internal standards are usually added. These internal standards have similar chemical properties to the target analytes, but have different mass-to-charge ratios when detected by mass spectrometry. By comparing the response intensity of the target analyte with that of the internal standard, errors caused by factors such as sample processing and injection volume changes can be corrected, thereby improving the accuracy of quantification.

[0032] In a specific example of the present application, first, the target analyte solution separated by high performance liquid chromatography (HPLC) enters the mass spectrometer. The mass spectrometer ionizes the compounds in the solution by electrospray ionization (ESI) mode. In this process, the compound molecules are ionized into charged particles, and then separated and detected in the mass spectrometer according to their mass-to-charge ratio (m / z). Specifically, for each target compound and its corresponding isotope internal standard, the mass spectrometer scans within a specific mass-to-charge ratio range and records their signal intensities. For example, in this method, the target compounds of folic acid and its metabolites include folic acid (VB9), 5-methyltetrahydrofolate (5-MTHF), 5-formyltetrahydrofolate (5-FoTHF), flavin adenine dinucleotide (FAD) and vitamin B3 (VB3). The corresponding isotopic internal standards are folic acid-d4 (VB9-d4), 5-methyltetrahydrofolate-d4 (5-MTHF-d4), 5-formyltetrahydrofolate-d4 (5-FoTHF-d4), flavin adenine dinucleotide-d5 (FAD-d5) and vitamin B3-d4 (VB3-d4). During mass spectrometer detection, the capillary voltage was set to 3.0 kV, the nozzle voltage was 200 V, the drying gas temperature was 300 °C, the sheath gas temperature was 350 °C, the drying gas flow rate was 10 L / min, the sheath gas flow rate was 11 L / min, and the nebulizer gas pressure was 45 psi. The optimization of these parameters helps to improve ionization efficiency and detection sensitivity. Next, the mass spectrometer records the mass-to-charge ratio response of the target compound and its isotopic internal standard. By comparing the response intensity of the target compound and the internal standard, the concentration of the target compound can be calculated. Specifically, the concentration of the target compound is calculated using the internal standard method based on the pre-established standard curve. The standard curve is obtained by adding known concentrations of standards and internal standards to the blank matrix, measuring their response intensity and plotting them. In this way, even if there are some minor changes in the sample processing, they can be corrected by the internal standard to ensure the accuracy of the quantitative results. Finally, by repeating the detection and data processing of multiple samples, the accurate content of folic acid and its metabolites in red blood cells can be obtained. This method is not only suitable for laboratory research, but also can be used for clinical diagnosis and health assessment, helping doctors better understand the nutritional status and metabolic status of patients.

[0033] Specifically, in the above method, the process of liquid chromatography analysis to determine the mobile phase ratio of the supernatant to be injected is crucial, but the gradient elution program in the existing scheme is fixed, which limits the separation and detection process of the target object facing different sample characteristics. Based on this, the technical concept of the present application is to be able to monitor the liquid chromatogram of the supernatant to be injected in real time, and use the image analysis algorithm based on artificial intelligence and deep learning to extract the chromatographic separation state, so as to dynamically adjust the gradient parameters and reduce the need for manual intervention, thereby helping to optimize the separation process of the target object in the complex matrix. In particular, the method based on deep learning can automatically identify and quantify the key features in the chromatogram, and intelligently and dynamically adjust the gradient elution parameters according to this information, so as to facilitate the realization of more optimized material separation and detection. For example, if it is found that the separation of a certain component is not ideal, the separation effect can be improved by adjusting the ratio of mobile phase B, thereby optimizing the gradient elution process.

[0034] In one embodiment, Figure 2 As shown, in step S3, the supernatant to be injected is subjected to liquid chromatography analysis to determine the mobile phase ratio, including: S31, obtaining a liquid chromatogram of the supernatant to be injected; S32, preprocessing the liquid chromatogram and extracting multidimensional features of the liquid chromatogram to obtain shallow features of the liquid chromatogram and semantic features of the liquid chromatogram; S33, performing liquid chromatogram multidimensional feature joint encoding on the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram to obtain a significant information interaction representation between the shallow and semantic features of the liquid chromatogram; S34, determining the mobile phase ratio based on the significant information interaction representation between the shallow and semantic features of the liquid chromatogram.

[0035] Exemplarily, in step S31, a liquid chromatogram of the supernatant to be injected is obtained. It should be understood that obtaining a liquid chromatogram of the supernatant to be injected is to ensure that the target compound can be effectively separated and detected in a complex matrix. In the red blood cell sample, in addition to the target folic acid and its metabolites, there are a large number of other components, such as proteins, lipids and other small molecule metabolites. These interfering substances may co-elute with the target compound, resulting in signal overlap or increased background noise, thereby affecting the accuracy and sensitivity of the test results. By obtaining a liquid chromatogram, the separation of each component can be intuitively observed, potential interfering peaks can be identified, and corresponding measures can be taken to optimize. Secondly, the liquid chromatogram can provide important information about the sample components. For example, by analyzing the peak shape, retention time, peak area and other parameters in the chromatogram, the separation effect of different compounds and whether there is a co-elution phenomenon can be evaluated. This is particularly important for the simultaneous detection of multiple components in a complex matrix, because different target compounds may have different polarities and chemical properties, and a specific mobile phase ratio is required to achieve the best separation effect. In addition, the liquid chromatogram can also be used to calibrate the experimental conditions to ensure the consistency and reliability of each analysis. Finally, the analysis of liquid chromatograms in combination with artificial intelligence and deep learning algorithms can further improve the separation effect. Modern computational tools can automatically identify and quantify key features in chromatograms and intelligently and dynamically adjust gradient elution parameters based on this information. This method can not only capture the significant interactive information between shallow features (such as peak shape, retention time, etc.) and semantic features (such as the chemical properties of target compounds, metabolic pathways, etc.) in chromatograms, but also provide more comprehensive data support for subsequent mobile phase ratio optimization. In this way, the mobile phase ratio can be adaptively adjusted in a single run, thereby achieving more optimized separation and detection of substances.

[0036] In a specific example of this application, Figure 3 As shown, in step S32, the liquid chromatogram is preprocessed and multi-dimensional features of the liquid chromatogram are extracted to obtain shallow features of the liquid chromatogram and semantic features of the liquid chromatogram, including: S321, denoising and baseline correction are performed on the liquid chromatogram to obtain a corrected liquid chromatogram; S322, the corrected liquid chromatogram is passed through a liquid chromatogram feature extractor based on a dilated space pyramid pooling module to obtain a liquid chromatogram shallow feature map as the liquid chromatogram shallow feature and liquid chromatogram semantic feature map; S323, the liquid chromatogram semantic feature map is passed through a liquid chromatogram semantic global perception enhancement module based on a fully connected layer to obtain a liquid chromatogram semantic feature vector as the liquid chromatogram semantic feature.

[0037] Exemplarily, in step S321, it is considered that in liquid chromatography analysis, the liquid chromatogram is an important data source for detecting target compounds. However, due to factors such as instrument noise, baseline drift, background interference, etc., the original liquid chromatogram often contains a large amount of noise and irregular baseline fluctuations, which seriously affect the subsequent data analysis and result accuracy. Therefore, the liquid chromatogram is denoised and baseline corrected to obtain a corrected liquid chromatogram. Specifically, the noise in the liquid chromatogram may come from instrument electronic noise, detector fluctuations, mobile phase inhomogeneity, etc. Noise can cover the signal of the target compound, especially the signal of the low-concentration target may be submerged by noise. Through denoising, the noise level can be effectively reduced, the peak signal of the target compound can be highlighted, and the signal-to-noise ratio can be improved, so as to more accurately identify and quantify the target. In addition, baseline drift may be caused by factors such as mobile phase gradient changes, column temperature fluctuations, and detector response changes. Baseline drift can cause peak shape distortion, affecting the accurate judgment of the integral area and retention time of the peak. Through baseline correction, the baseline can be restored to a stable state, ensuring the accuracy of the starting and ending points of each peak, avoiding peak area calculation errors caused by baseline drift, and thus improving the accuracy of quantitative analysis.

[0038] In a specific example of the present application, first, the original liquid chromatogram containing noise and baseline drift is obtained. These noises may come from instrument electronic noise, detector fluctuations, mobile phase inhomogeneity, etc., while baseline drift may be caused by factors such as mobile phase gradient changes, column temperature fluctuations, or detector response changes. In order to improve the signal-to-noise ratio and ensure the accuracy of the peak shape, these interfering factors need to be processed. In order to remove background noise, a smoothing filtering method is used. Specifically, after loading the original liquid chromatogram in a data analysis software (such as Agilent OpenLab CDS or Thermo Scientific Chromeleon), a suitable smoothing algorithm is selected, such as a Savitzky-Golay filter. The filter can reduce the impact of high-frequency noise while retaining the main features of the signal. Setting an appropriate window width (such as 5 or 7 points) can effectively reduce the noise level and make the peak signal of the target compound clearer. Another method is to use wavelet transform technology. This method can decompose signals at different scales, thereby effectively distinguishing noise from useful signals. By selecting a suitable wavelet type (such as db4) and the number of decomposition layers (such as 4 layers), the liquid chromatogram can be decomposed. Then, thresholding is applied to each decomposition coefficient to remove noise components below a certain threshold. Finally, the processed coefficients are reconstructed into a denoised liquid chromatogram. The processed chromatogram not only reduces noise but also maintains the main features of the target peak.

[0039] After the denoising process is completed, the next step is to perform baseline correction. Baseline drift will cause peak shape distortion, affecting the accurate judgment of the peak integral area and retention time. Therefore, the baseline must be restored to a stable state. Common baseline correction methods include manual correction and automatic correction. In manual correction, you can observe the baseline trend in the chromatogram, select a relatively flat area as the reference baseline, and then adjust the baseline of other parts to make it consistent with the reference baseline. This method is suitable for simple chromatograms, but for complex multi-component samples, manual correction may take a lot of time and effort. In contrast, automatic baseline correction is more efficient. Modern data analysis software usually has a built-in automatic baseline correction function that can intelligently identify the baseline and correct it according to the characteristics of the chromatogram. For example, in Agilent OpenLab CDS, you can select the "Automatic Baseline Correction" option, and the software will automatically identify the part with baseline drift and adjust it to a stable state. In addition, you can set some parameters to optimize the correction effect, such as baseline threshold, smoothing window size, etc.

[0040] Exemplarily, in step S322, considering that the liquid chromatogram is corrected, in order to be able to capture the features related to the target in the complex matrix in the liquid chromatogram, and to dynamically adjust the mobile phase ratio by real-time monitoring of the separation situation, it is necessary to extract the features of the corrected liquid chromatogram. However, in the traditional convolutional neural network, the receptive field size is fixed, which limits the model's understanding of the contextual information of the liquid chromatogram over a larger range. Therefore, in the technical solution of the present application, the corrected liquid chromatogram is further passed through a liquid chromatogram feature extractor based on a hollow space pyramid pooling module to obtain a shallow feature map of the liquid chromatogram and a semantic feature map of the liquid chromatogram. Through the processing of the liquid chromatogram feature extractor based on the hollow space pyramid pooling module, the network can be allowed to capture liquid chromatogram features at different scales, which is particularly important for complex and variable liquid chromatogram data. This is because the information such as peak shape, retention time drift, and co-elution interference in the liquid chromatogram may show different characteristics at different scales, which helps to enhance the model's ability to understand different sample characteristics, thereby better realizing the control of the gradient elution process. It is worth mentioning that the shallow feature map of the liquid chromatogram usually contains more original detailed information about the input liquid chromatogram, such as the position, width and preliminary morphological characteristics of the peak, which is very important for identifying specific compound peaks. The semantic feature map of the liquid chromatogram focuses more on high-level abstract features, such as the relationship between different compounds, the overall distribution pattern, etc., which is critical for understanding the composition of complex mixtures and their dynamic changes.

[0041] Exemplarily, in step S323, since the liquid chromatogram semantic feature map contains multi-scale information about the liquid chromatogram semantic features, these different scales of liquid chromatogram semantics are related to the characteristic perception of different samples and the subsequent gradient elution control, but this information may be relatively scattered, and it is difficult to present the global representation information of the semantic features in the liquid chromatogram semantic feature map. Based on this, in the technical solution of the present application, the liquid chromatogram semantic feature map is further passed through a liquid chromatogram semantic global perception enhancement module based on a fully connected layer to obtain a liquid chromatogram semantic feature vector. Through the processing of the liquid chromatogram semantic global perception enhancement module based on a fully connected layer, the complex relationships and semantic association features between different regions in the liquid chromatogram can be captured, which is particularly important for understanding complex liquid chromatograms. Such a more global representation of liquid chromatogram semantic features is crucial for distinguishing different compounds and their metabolites.

[0042] Exemplarily, in step S33, the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram are jointly encoded with liquid chromatogram multidimensional features to obtain a significant information interaction representation between the shallow and semantic features of the liquid chromatogram. It should be understood that in liquid chromatography analysis, the liquid chromatogram is the core data source for detecting target compounds. In order to extract richer and more accurate feature information from the chromatogram, it is necessary not only to analyze and encode the chromatogram from multiple dimensions (such as shallow features and semantic features), but also to capture the significant interaction information between the shallow features (such as peak shape, retention time, etc.) and semantic features (such as the chemical properties of the target compound, metabolic pathways, etc.) in the chromatogram, so as to provide more accurate guidance for the subsequent optimization of the mobile phase ratio. Specifically, since the shallow features of the liquid chromatogram mainly include intuitive information such as peak retention time, peak height, peak area, peak shape, etc. These features can reflect the physicochemical properties (such as polarity, molecular weight, etc.) and separation state (such as whether it overlaps with other peaks) of the target compound. The semantic features of the liquid chromatogram are a deep understanding of the chromatogram, such as the chemical structure of the target compound, metabolic pathways, interactions with other compounds, etc. Based on this, in the technical solution of the present application, the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram are further jointly encoded with liquid chromatogram multidimensional features to obtain a significant information interaction representation between the shallow and semantic features of the liquid chromatogram. Through the processing of the joint encoding of multidimensional features of liquid chromatography, the significant interaction between the shallow features and semantic features in the liquid chromatogram can be captured, such as whether the separation effect of a peak is affected by the chemical structure of the target compound, or whether a baseline drift is related to the change in the mobile phase gradient. These interactions can provide more comprehensive data support for the subsequent optimization of the mobile phase ratio, thereby helping to optimize the gradient elution program and dynamically adjust the mobile phase ratio. For example, if it is found that the separation effect of a peak is not ideal, the model can adjust the proportion of mobile phase B according to the information in the interactive coding vector, which helps to improve the separation effect.

[0043] In one embodiment, Figure 4 As shown, in step S33, the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram are jointly encoded with liquid chromatogram multi-dimensional features to obtain a significant information interaction representation between the shallow and semantic features of the liquid chromatogram, including: S331, core clues are extracted from the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram respectively to obtain the semantic core clue encoding features of the liquid chromatogram and the shallow core clue encoding features of the liquid chromatogram; S332, based on the semantic core clue encoding features of the liquid chromatogram and the shallow core clue encoding features of the liquid chromatogram, fine-grained feature interaction analysis is performed on the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram to obtain a significant information interaction representation between the shallow and semantic features of the liquid chromatogram.

[0044] In a specific example of this application, Figure 5 As shown, in step S331, core clues are extracted from the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram respectively to obtain the liquid chromatogram semantic core clue coding features and the liquid chromatogram shallow core clue coding features, including: S3311, core clues extraction based on point convolution coding is performed on the semantic feature vector of the liquid chromatogram to obtain the liquid chromatogram semantic core clue coding vector as the liquid chromatogram semantic core clue coding feature; S3312, the liquid chromatogram shallow core clue coding vector is extracted from the shallow feature map of the liquid chromatogram as the liquid chromatogram shallow core clue coding feature.

[0045] Exemplarily, in step S3311, the process of extracting the core clues of the semantic feature vector of the liquid chromatogram based on point convolution coding to obtain the semantic core clue coding vector of the liquid chromatogram is intended to extract the most representative information fragments from the complex raw data. This process is like an experienced analyst looking for the most critical signals in a large amount of data, which can most effectively describe the characteristics of the sample and the existence of each component therein. Specifically, directly using these liquid chromatogram semantic feature vectors for subsequent analysis may face many challenges, including but not limited to noise interference, the complexity of co-elution peaks, and variability under different experimental conditions. In order to solve these problems, it is crucial to use the method of point convolution coding to extract the core clues of the semantic feature vector. The point convolution operation plays the role of a refiner here, which can not only maintain the integrity of high-dimensional information, but also strengthen those features that are semantically significant. This means that after point convolution coding, the originally complex chromatogram is simplified into a highly condensed and abstract representation-the semantic core clue coding vector. This representation method allows subsequent analysis to focus more on those truly important features, rather than being overwhelmed by irrelevant information. In addition, point convolution effectively avoids redundant expression of information by focusing on important elements of the modality (chromatogram in this case). In this process, point convolution can identify and emphasize those parts that are most critical to understanding the composition of the sample, such as the unique peak shape of a specific metabolite or its interaction pattern with adjacent components. Doing so can not only improve the sensitivity of detection, but also enhance the specificity of the results, thereby more accurately quantifying the concentration of the target compound. Specifically, the calculation process of the semantic core clue encoding vector of the liquid chromatogram can be expressed as follows:

[0046] in, is the semantic feature vector of the liquid chromatogram, It is point convolution processing, is the bias vector, yes Activation function, is the semantic core clue encoding vector of the liquid chromatography graph.

[0047] In a specific example of the present application, in step S3312, extracting a shallow core clue coding vector of the liquid chromatogram from the shallow feature map of the liquid chromatogram as the shallow core clue coding feature of the liquid chromatogram includes: splitting the shallow feature map of the liquid chromatogram to obtain a sequence of shallow local feature maps of the liquid chromatogram. Specifically, the process can be expressed by a formula as follows:

[0048] in, is the shallow characteristic graph of the liquid chromatogram, represents feature map splitting, Respectively represent the first, second, and third in the sequence of shallow local feature graphs of liquid chromatograms and A shallow local feature map of the liquid chromatogram.

[0049] The attention-based salient region attention analysis is performed on each of the shallow local feature maps of the liquid chromatogram in the sequence of shallow local feature maps of the liquid chromatogram to obtain a sequence of salient region attention coefficients. Specifically, the process can be expressed by the formula:

[0050]

[0051]

[0052] in, is a convolution layer with a convolution kernel of 3×3. for Activation function, For the A liquid chromatogram with shallow local hidden feature graphs, To perform global pooling on each feature matrix along the channel dimension in the feature map, For the The shallow local feature vector of the liquid chromatogram corresponding to the shallow local implicit feature map of the liquid chromatogram, and are the maximum and minimum values ​​in the vector, respectively. , and are weight coefficients, and are the mean and variance of the vector, respectively. is the attention coefficient of the significant area.

[0053] Based on the sequence of the attention coefficients of the salient regions, a core clue search is performed on the shallow feature map of the liquid chromatogram to obtain a shallow core clue encoding vector of the liquid chromatogram. Specifically, the process can be expressed as follows:

[0054] in, is the exponential function value with the natural constant e as the base, is the number of characteristic graphs in the sequence of shallow local characteristic graphs of the liquid chromatogram, is the shallow core clue encoding vector of the liquid chromatogram.

[0055] Exemplarily, in step S3312, the entire process of splitting the shallow feature map of the liquid chromatogram, focusing on the analysis of the significant areas, and searching for core clues is intended to extract the most critical information fragments from a complex and multidimensional data set. This process is essential for accurately identifying and quantifying folic acid and its metabolites in red blood cells. First, the original shallow feature map of the liquid chromatogram is split to better understand and process the data. Imagine that if the entire chromatogram is processed directly, it may be difficult to capture subtle but important details due to the large amount of information. By splitting, a complex image can be decomposed into multiple local feature maps, each focusing on a specific time period or component. This is like dividing a large painting into several small pieces for careful study, each of which can be observed and analyzed more carefully, making it easier to discover valuable information hidden in a large amount of data.

[0056] Next, attention-based salient area attention analysis is applied to the shallow local feature maps of these split liquid chromatograms. Here, the attention mechanism is like installing a pair of eyes that can automatically focus on the system, which can automatically identify the most important parts among a lot of information. In this process, the algorithm evaluates each local feature map and gives different weights according to its importance. For example, when analyzing red blood cell samples, peaks related to the target folate and its metabolites will be given higher attention because they are key indicators for judging the quality of the sample. This method not only improves the efficiency of data analysis, but also ensures that the most relevant information is prioritized and reduces the interference of noise and other irrelevant factors.

[0057] Finally, based on the sequence of attention coefficients of the salient regions, a core clue search is performed on the shallow feature map of the liquid chromatogram, with the aim of extracting the most representative feature vector, namely the shallow core clue encoding vector of the liquid chromatogram. This step is like the process of further refining gold after preliminary screening. By integrating the most salient parts of each local feature map, a highly condensed representation can be constructed, which contains the most important and useful information about the sample.

[0058] In a specific example of this application, Figure 6 As shown, in step S332, based on the liquid chromatogram semantic core clue encoding features and the liquid chromatogram shallow core clue encoding features, fine-grained feature interaction analysis is performed on the liquid chromatogram shallow features and the liquid chromatogram semantic features to obtain a significant information interaction representation between the shallow and semantic features of the liquid chromatogram, including: S3321, based on the liquid chromatogram semantic core clue encoding vector and the liquid chromatogram shallow core clue encoding vector, a core clue weaving template matrix between the shallow and semantic features of the liquid chromatogram is constructed; S3322, based on the core clue weaving template matrix between the shallow and semantic features of the liquid chromatogram, a cross-modal fine-grained feature interaction analysis is performed on the liquid chromatogram semantic feature vector and the liquid chromatogram shallow feature map to obtain a significant information interaction encoding vector between the shallow and semantic features of the liquid chromatogram as the significant information interaction representation between the shallow and semantic features of the liquid chromatogram.

[0059] In a specific example of the present application, in step S3321, based on the liquid chromatogram semantic core clue encoding vector and the liquid chromatogram shallow core clue encoding vector, a liquid chromatogram shallow-semantic feature core clue weaving template matrix is ​​constructed, including: constructing an inter-modal core clue weaving template between the liquid chromatogram semantic core clue encoding vector and the liquid chromatogram shallow core clue encoding vector to obtain the liquid chromatogram shallow-semantic feature core clue weaving template matrix.

[0060] In a specific example of the present application, constructing an inter-modal core clue weaving template between the liquid chromatogram semantic core clue encoding vector and the liquid chromatogram shallow core clue encoding vector to obtain the liquid chromatogram shallow-semantic feature inter-core clue weaving template matrix, including: using a feature mapping function to process the liquid chromatogram semantic core clue encoding vector and the liquid chromatogram shallow core clue encoding vector to obtain a liquid chromatogram shallow-semantic feature inter-core association matrix. Specifically, the process can be expressed by a formula:

[0061] in, and are feature mapping functions, such as linear mapping or nonlinear kernel function, represents transpose, is a core correlation matrix between shallow and semantic features of the liquid chromatogram. The core correlation matrix between shallow and semantic features of the liquid chromatogram is subjected to semantic alignment compensation to obtain a core clue weaving template matrix between shallow and semantic features of the liquid chromatogram.

[0062] In a specific example of the present application, semantic alignment compensation is performed on the core association matrix between the shallow layer and semantic features of the liquid chromatogram to obtain the core clue weaving template matrix between the shallow layer and semantic features of the liquid chromatogram, including: using the residual mechanism and the feature mapping function to perform structured semantic alignment on the liquid chromatogram semantic core clue encoding vector and the liquid chromatogram shallow core clue encoding vector to obtain the liquid chromatogram semantic alignment vector and the liquid chromatogram shallow alignment vector. Specifically, the process can be expressed by the formula:

[0063]

[0064] in, represents positional subtraction, and They represent the semantic alignment vector of liquid chromatogram and the shallow alignment vector of liquid chromatogram respectively.

[0065] The explicit correlation matrix between the shallow-semantic features of the liquid chromatogram is calculated between the semantic alignment vector of the liquid chromatogram and the shallow alignment vector of the liquid chromatogram. Specifically, the process can be expressed by the formula:

[0066] in, To calculate the square of the vector's two-norm, represents the logarithmic function value with base 2, It is the explicit correlation matrix between shallow-semantic features of the liquid chromatogram.

[0067] Based on the explicit correlation matrix between the shallow and semantic features of the liquid chromatogram, the core correlation matrix between the shallow and semantic features of the liquid chromatogram is significantly compensated to obtain the core clue weaving template matrix between the shallow and semantic features of the liquid chromatogram. Specifically, the process can be expressed as follows:

[0068] in, It is the core clue weaving template matrix between the shallow and semantic features of the liquid chromatography graph.

[0069] Exemplarily, in step S3321, first, the semantic core clue encoding vector of the liquid chromatogram and the shallow core clue encoding vector of the liquid chromatogram need to be integrated. This is like combining two different perspectives or two different sets of observations to form a more comprehensive view. For example, when analyzing red blood cell samples, on the one hand, there is confirmation information about the presence of folic acid and its metabolites (from semantic features), and on the other hand, there is a detailed description of the specific manifestations of these compounds in the chromatogram (from shallow features). By constructing an inter-modal core clue weaving template, a bridge can be built between the two types of information, so that each type of information can provide supplementary and support for the other type of information. Next, the process of using this weaving template to generate a core clue weaving template matrix is ​​actually creating a framework that can effectively organize and integrate data from different sources. It is like an intelligent map or index system that helps to quickly find the most useful parts from a large amount of information and present it in a structured way. Doing so not only helps to improve the efficiency of data analysis, but also enhances the reliability and accuracy of the results.

[0070] Preferably, in the above process, the feature vectors of the core clue representations of different modal features (such as semantic features and shallow features) are and , through its residual mechanism based on heterogeneous modal mapping, a preliminary structured semantic alignment of universal explicit associations is performed in the shared representation space to obtain the liquid chromatogram semantic alignment vector and the liquid chromatogram shallow alignment vector and Then, based on the global semantic spatialization representation of the graph structure under explicit modeling of mapping preference, the residual mechanism is further used to capture the "missing" information in the alignment process under different mapping modalities, and adaptive compensation is performed through explicit association to further capture the fine-grained significant difference information between modalities in the initial alignment process. In this way, the contradiction between the alignment complexity of the intrinsic driving mapping of cross-modal differences and the intuitiveness of the explicit association of the graph structure can be optimized, and the illusion of global semantic constraints between modalities under the condition of misalignment of fine-grained relationships between local salient regions can be reduced.

[0071] In a specific example of the present application, in step S3322, based on the core clue weaving template matrix between the shallow and semantic features of the liquid chromatogram, the semantic feature vector of the liquid chromatogram and the shallow feature map of the liquid chromatogram are subjected to cross-modal fine-grained feature interaction analysis to obtain a significant information interaction encoding vector between the shallow and semantic features of the liquid chromatogram as a significant information interaction representation between the shallow and semantic features of the liquid chromatogram, including: performing feature decoupling along the channel dimension on the shallow feature map of the liquid chromatogram to obtain a set of shallow local feature matrices of the liquid chromatogram. Specifically, the process can be expressed by the formula:

[0072] in, represents the feature decoupling along the channel dimension, Respectively represent each liquid chromatogram shallow local feature matrix in the sequence of liquid chromatogram shallow local feature matrices.

[0073] The semantic feature vector of the liquid chromatogram is used as the query vector, each of the shallow local feature matrices of the liquid chromatogram in the set of shallow local feature matrices of the liquid chromatogram is used as the key matrix, and the core clue weaving template matrix between the shallow and semantic features of the liquid chromatogram is used as the prior information constraint matrix, which is input into the cross-modal template constraint encoder based on the heterogeneous converter to obtain a set of fine-grained interaction encoding vectors between the shallow and semantic features of the liquid chromatogram. Specifically, the process can be expressed by the formula:

[0074] in, is the first in the sequence of the shallow local feature matrix of the liquid chromatogram A shallow local feature matrix of the liquid chromatogram, for The scale of the matrix is ​​the width times the height. is matrix multiplication, express function, is the first in the set of fine-grained interaction encoding vectors between shallow-semantic features of liquid chromatograms. A fine-grained interaction encoding vector between shallow-level semantic features of liquid chromatograms.

[0075] The positional mean vector of the set of fine-grained interaction coding vectors between the shallow and semantic features of the liquid chromatogram is calculated to obtain the significant information interaction coding vector between the shallow and semantic features of the liquid chromatogram. Specifically, the process can be expressed by the formula:

[0076] in, is the number of vectors in the set of fine-grained interaction encoding vectors between shallow-semantic features of the liquid chromatogram, It is the significant information interaction encoding vector between the shallow-semantic features of the liquid chromatogram.

[0077] Exemplarily, in step S3322, the shallow feature map of the liquid chromatogram is first feature decoupled along the channel dimension to obtain a set of shallow local feature matrices of the liquid chromatogram. This process decomposes the complex original feature map into multiple local feature matrices. Each local feature matrix represents the detailed information of the chromatogram in a specific time period or component, so that the subsequent analysis can capture the key features of each local area in more detail. Then, the semantic feature vector of the liquid chromatogram is used as the query vector, each matrix in the set of shallow local feature matrices of the liquid chromatogram is used as the key matrix, and the core clue weaving template matrix between the shallow and semantic features of the liquid chromatogram is combined as the prior information constraint matrix, which is input into the cross-modal template constraint encoder based on the heterogeneous converter. In this process, the semantic feature vector plays the role of query, helping the system to clarify the information that needs to be paid attention to; while the shallow local feature matrix provides specific detail information. Through the cross-modal template constraint encoder of the heterogeneous converter, this information is integrated to form a set of fine-grained interactive encoding vectors. This step not only ensures the effective interaction between different modalities, but also strengthens the standardization and accuracy of this interaction through the prior information constraint matrix, and improves the representation ability of the results. Finally, the positional mean vector of the set of fine-grained interaction encoding vectors between shallow and semantic features of the liquid chromatogram is calculated to obtain the significant information interaction encoding vector between shallow and semantic features of the liquid chromatogram. This process further compresses the processing cost and retains the main semantic information. Specifically, by calculating the positional mean of all fine-grained interaction encoding vectors, a highly condensed and information-rich representation is generated - the significant information interaction encoding vector between shallow and semantic features of the liquid chromatogram. This step not only simplifies the workload of subsequent analysis, but also improves the accuracy and reliability of the results.

[0078] Exemplarily, in step S34, the mobile phase ratio is determined based on the significant information interaction representation between the shallow and semantic features of the liquid chromatogram. In a specific example of the present application, the mobile phase ratio is determined based on the significant information interaction representation between the shallow and semantic features of the liquid chromatogram, including: the significant information interaction encoding vector between the shallow and semantic features of the liquid chromatogram is passed through a decoder-based mobile phase ratio generator to obtain a mobile phase ratio recommended decoding value. In other words, the semantics of the significant information interaction between the shallow features and semantic features of the liquid chromatogram is used to perform decoding regression, so as to optimize the gradient elution program and adjust the mobile phase ratio. In this way, the gradient parameters can be dynamically adjusted to reduce the need for manual intervention, thereby helping to optimize the separation process of the target in a complex matrix, and facilitating more optimized separation and detection of substances.

[0079] In a specific example of the present application, in order to convert the significant information interaction encoding vector between the shallow-semantic features of the liquid chromatogram into a recommended decoding value of the mobile phase ratio, a specific mobile phase ratio generator based on a fully connected decoder can be designed. The input layer of this decoder receives the significant information interaction encoding vector as input, and finally outputs the recommended value of the mobile phase ratio through a fully connected layer. Among them, the output layer determines the number of nodes in the output layer according to the number of mobile phase ratios that need to be predicted. For example, if you want to predict the ratio of two solvents A and B, the output layer should have two nodes representing the ratio of the two solvents respectively.

[0080] In a specific example of the present application, when performing reverse phase high performance liquid chromatography (RP-HPLC), water (containing 0.1% trifluoroacetic acid) is used as phase A and acetonitrile (containing 0.1% trifluoroacetic acid) is used as phase B. Under the initial setting, the mobile phase ratio may be 95:5 (A:B), which means 95% water and 5% acetonitrile. If the experimental results show that some compounds are not well separated, the system will automatically adjust the mobile phase ratio based on the significant information interaction encoding vector. After calculation by the mobile phase ratio generator based on the fully connected decoder, the new recommended value is 70:30 (A:B). This means that starting from the initial 95:5, it will gradually change to 70:30 in the next 20 minutes to enhance the elution ability of the solvent and ensure that those compounds that are difficult to dissolve in water can be effectively eluted. In this process, by intelligently adjusting the mobile phase ratio, not only can the separation conditions be optimized, the analysis efficiency and accuracy can be improved, but also the need for manual intervention is reduced, making the whole process more automated and intelligent.

[0081] In summary, the method for detecting erythrocyte folate metabolites by liquid chromatography-mass spectrometry tandem technology according to the embodiment of the present application is illustrated, which can be achieved by real-time monitoring of the liquid chromatogram to be injected by the supernatant, and using an image analysis algorithm based on artificial intelligence and deep learning to extract the chromatographic separation state, so as to dynamically adjust the gradient parameters and reduce the need for manual intervention, thereby helping to optimize the separation process of the target in a complex matrix. In particular, the method based on deep learning can automatically identify and quantify the key features in the chromatogram, and intelligently and dynamically adjust the gradient elution parameters based on this information, so as to facilitate the realization of more optimized material separation and detection. For example, if it is found that the separation of a certain component is not ideal, the separation effect can be improved by adjusting the ratio of mobile phase B, thereby optimizing the gradient elution process.

[0082] Finally, it should be noted that the embodiments described above are only some embodiments of the present invention, rather than all embodiments. The detailed description of the embodiments of the present invention is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

Claims

1. A method for detecting erythrocyte folate metabolites using liquid chromatography-mass spectrometry tandem technology, comprising: Collecting a blood sample containing EDTA anticoagulant, and centrifuging it to remove plasma to obtain a sample to be tested; Pre-treating the sample to be tested to obtain a supernatant to be injected; Performing liquid chromatography analysis on the supernatant to be injected to determine the mobile phase ratio; performing gradient elution on the supernatant to be injected based on the mobile phase ratio to separate and remove interfering groups in the sample matrix to obtain a target analyte solution; The method comprises detecting the mass-to-charge ratio of the target analyte in the target analyte solution and its corresponding isotope internal standard by a mass spectrometer, and using the isotope internal standard method for quantitative determination of the content of erythrocyte folate metabolites; characterized in that the supernatant to be injected is subjected to liquid chromatography analysis to determine the mobile phase ratio, including: Obtaining a liquid chromatogram of the supernatant to be injected; Preprocessing the liquid chromatogram and extracting multidimensional features of the liquid chromatogram to obtain shallow features of the liquid chromatogram and semantic features of the liquid chromatogram; The shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram are jointly encoded with multi-dimensional features of liquid chromatogram to obtain a significant information interaction representation between the shallow and semantic features of the liquid chromatogram, including: extracting core clues from the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram respectively to obtain the semantic core clue encoding features of the liquid chromatogram and the shallow core clue encoding features of the liquid chromatogram; based on the semantic core clue encoding features of the liquid chromatogram and the shallow core clue encoding features of the liquid chromatogram, fine-grained feature interaction analysis is performed on the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram to obtain a significant information interaction representation between the shallow and semantic features of the liquid chromatogram; The mobile phase ratio is determined based on the significant information interaction representation between the shallow-semantic features of the liquid chromatogram.

2. The method for detecting erythrocyte folate metabolites by liquid chromatography-mass spectrometry tandem technology according to claim 1, characterized in that: Preprocessing the liquid chromatogram and extracting multidimensional features of the liquid chromatogram to obtain shallow features of the liquid chromatogram and semantic features of the liquid chromatogram, including: De-noising and baseline correcting the liquid chromatogram to obtain a corrected liquid chromatogram; The corrected liquid chromatogram is passed through a liquid chromatogram feature extractor based on a dilated space pyramid pooling module to obtain a liquid chromatogram shallow feature map as the liquid chromatogram shallow feature and the liquid chromatogram semantic feature map; The liquid chromatogram semantic feature map is passed through a liquid chromatogram semantic global perception enhancement module based on a fully connected layer to obtain a liquid chromatogram semantic feature vector as the liquid chromatogram semantic feature.

3. The method for detecting erythrocyte folate metabolites by liquid chromatography-mass spectrometry tandem technology according to claim 2, characterized in that: The shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram are respectively subjected to core clue extraction to obtain the semantic core clue coding features of the liquid chromatogram and the shallow core clue coding features of the liquid chromatogram, including: Performing point convolution coding-based core clue extraction on the liquid chromatogram semantic feature vector to obtain a liquid chromatogram semantic core clue coding vector as the liquid chromatogram semantic core clue coding feature; A shallow core clue coding vector of the liquid chromatogram is extracted from the shallow feature map of the liquid chromatogram as the shallow core clue coding feature of the liquid chromatogram.

4. The method for detecting erythrocyte folate metabolites by liquid chromatography-mass spectrometry tandem technology according to claim 3, characterized in that: Extracting a shallow core clue coding vector of the liquid chromatogram from the shallow feature map of the liquid chromatogram as the shallow core clue coding feature of the liquid chromatogram includes: Splitting the shallow characteristic graph of the liquid chromatogram to obtain a sequence of shallow local characteristic graphs of the liquid chromatogram; Performing attention-based salient region attention analysis on each of the shallow local feature maps of the liquid chromatogram in the sequence of shallow local feature maps of the liquid chromatogram to obtain a sequence of salient region attention coefficients; Based on the sequence of the salient area attention coefficients, a core clue search is performed on the shallow feature map of the liquid chromatogram to obtain a shallow core clue encoding vector of the liquid chromatogram.

5. The method for detecting erythrocyte folate metabolites by liquid chromatography-mass spectrometry tandem technology according to claim 4, characterized in that: Based on the semantic core clue encoding features of the liquid chromatogram and the shallow core clue encoding features of the liquid chromatogram, fine-grained feature interaction analysis is performed on the shallow features of the liquid chromatogram and the semantic features of the liquid chromatogram to obtain a significant information interaction representation between the shallow and semantic features of the liquid chromatogram, including: Based on the liquid chromatogram semantic core clue encoding vector and the liquid chromatogram shallow core clue encoding vector, constructing a liquid chromatogram shallow-semantic feature core clue weaving template matrix; Based on the core clue weaving template matrix between the shallow and semantic features of the liquid chromatogram, a cross-modal fine-grained feature interaction analysis is performed on the semantic feature vector of the liquid chromatogram and the shallow feature map of the liquid chromatogram to obtain a significant information interaction encoding vector between the shallow and semantic features of the liquid chromatogram as a representation of the significant information interaction between the shallow and semantic features of the liquid chromatogram.

6. The method for detecting erythrocyte folate metabolites by liquid chromatography-mass spectrometry tandem technology according to claim 5, characterized in that: Based on the liquid chromatogram semantic core clue encoding vector and the liquid chromatogram shallow core clue encoding vector, a liquid chromatogram shallow-semantic feature core clue weaving template matrix is ​​constructed, including: constructing an inter-modal core clue weaving template between the liquid chromatogram semantic core clue encoding vector and the liquid chromatogram shallow core clue encoding vector to obtain the liquid chromatogram shallow-semantic feature core clue weaving template matrix.

7. The method for detecting erythrocyte folate metabolites by liquid chromatography-mass spectrometry tandem technology according to claim 6, characterized in that: Constructing an inter-modal core clue weaving template between the liquid chromatogram semantic core clue encoding vector and the liquid chromatogram shallow core clue encoding vector to obtain the liquid chromatogram shallow-semantic feature inter-core clue weaving template matrix, including: Using a feature mapping function, the liquid chromatogram semantic core clue encoding vector and the liquid chromatogram shallow core clue encoding vector are processed to obtain a liquid chromatogram shallow-semantic feature core association matrix; The core correlation matrix between the shallow layer and semantic features of the liquid chromatogram is subjected to semantic alignment compensation to obtain the core clue weaving template matrix between the shallow layer and semantic features of the liquid chromatogram.

8. The method for detecting erythrocyte folate metabolites by liquid chromatography-mass spectrometry tandem technology according to claim 7, characterized in that: The core correlation matrix between the shallow layer and semantic features of the liquid chromatogram is subjected to semantic alignment compensation to obtain a core clue weaving template matrix between the shallow layer and semantic features of the liquid chromatogram, including: Using the residual mechanism and the feature mapping function, the liquid chromatogram semantic core clue encoding vector and the liquid chromatogram shallow core clue encoding vector are structured semantically aligned to obtain a liquid chromatogram semantic alignment vector and a liquid chromatogram shallow alignment vector; Calculating an explicit correlation matrix between shallow-layer and semantic features of the liquid chromatogram between the semantic alignment vector of the liquid chromatogram and the shallow-layer alignment vector of the liquid chromatogram; Based on the explicit correlation matrix between the shallow and semantic features of the liquid chromatogram, the core association matrix between the shallow and semantic features of the liquid chromatogram is significantly compensated to obtain the core clue weaving template matrix between the shallow and semantic features of the liquid chromatogram.

9. The method for detecting erythrocyte folate metabolites by liquid chromatography-mass spectrometry tandem technology according to claim 8, characterized in that: Based on the core clue weaving template matrix between the shallow and semantic features of the liquid chromatogram, a cross-modal fine-grained feature interaction analysis is performed on the semantic feature vector of the liquid chromatogram and the shallow feature map of the liquid chromatogram to obtain a significant information interaction encoding vector between the shallow and semantic features of the liquid chromatogram as a significant information interaction representation between the shallow and semantic features of the liquid chromatogram, including: Performing feature decoupling along the channel dimension on the shallow feature map of the liquid chromatogram to obtain a set of shallow local feature matrices of the liquid chromatogram; Taking the semantic feature vector of the liquid chromatogram as a query vector, taking each shallow local feature matrix of the liquid chromatogram in the set of shallow local feature matrices of the liquid chromatogram as a key matrix, and taking the core clue weaving template matrix between shallow and semantic features of the liquid chromatogram as a prior information constraint matrix, inputting them into a cross-modal template constraint encoder based on a heterogeneous converter to obtain a set of fine-grained interaction encoding vectors between shallow and semantic features of the liquid chromatogram; The positional mean vector of the set of fine-grained interaction coding vectors between shallow-layer and semantic features of the liquid chromatogram is calculated to obtain a significant information interaction coding vector between shallow-layer and semantic features of the liquid chromatogram.

10. The method for detecting erythrocyte folate metabolites by liquid chromatography-mass spectrometry tandem technology according to claim 9, characterized in that: Based on the representation of significant information interaction between shallow-semantic features of the liquid chromatogram, the mobile phase ratio is determined, including: passing the significant information interaction encoding vector between shallow-semantic features of the liquid chromatogram through a decoder-based mobile phase ratio generator to obtain a recommended decoding value of the mobile phase ratio.

Images