A method for extracting, measuring and mass spectrometry imaging visualization of spatial metabolome high-order characteristic information

Abstract

The application belongs to the technical field of biological analysis and detection, and discloses a method for extracting, measuring and mass spectrum imaging visualizing high-order characteristic information of spatial metabolome. Specifically, the application relates to a mass spectrum imaging analysis method for evaluating the in-vivo active action site and action intensity of a preclinical candidate drug, which can be applied to the prediction of potential action sites, molecular targets and molecular mechanisms of a drug. The index "metabolic disturbance score" introduced in the application can be used to measure the change of high-order characteristics of spatial metabolome, and the metabolic changes of different spatial sites can be visually represented in images, so that the overall metabolic state change of each part of the body under the action of the candidate drug can be objectively evaluated. In addition, the application can be used to evaluate the influence of the drug on specific metabolic pathways, species, functions or biological transformation types, which is helpful to explain the potential molecular mechanism of the drug action and accelerate the discovery of new drug targets and clinical indications in the pharmaceutical industry.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of biological analysis and detection technology, and relates to a mass spectrometry imaging analysis method for evaluating the in vivo active site and intensity of a preclinical candidate drug or diagnosing a clinical disease, which can be applied to the prediction of potential action sites, molecular targets and molecular mechanisms of drugs and the molecular pathological diagnosis of clinical diseases. BACKGROUND

[0002] Mass spectrometry imaging (MSI) is a molecular imaging technology that combines desorption ionization technology with mass spectrometry detection and analysis. By desorbing, ionizing and collecting ion signals from the sample surface (mostly biological tissue samples) row by row or point by point, the signal intensity values of different mass-to-charge ratio ions are reconstructed into two-dimensional data arrays according to the physical space position and displayed in the form of images. It does not require radioisotopes, fluorescence or immunolabeling, and can not only directly obtain the spatial distribution information of drugs and their metabolites in biological samples, but also provide information on the molecular composition of endogenous metabolites in complex biological matrices. Therefore, mass spectrometry imaging has important application value in the preclinical drug development field, such as evaluating drug treatment effects, predicting potential toxic side effects and exploring molecular mechanisms.

[0003] Although mass spectrometry imaging can explore the changes in metabolite composition and the spatial distribution information of related metabolites at the lesion site under drug intervention, using any single metabolite as a molecular indicator of drug treatment or toxic side effects cannot fully describe the complex effects of drugs on the body. In addition to the relative abundance of metabolite ions, there are many more abstract biological information that cannot be directly read from the mass spectrum, such as the influence of drugs on the overall metabolic state of the body, metabolic pathways, certain structural metabolites or biological transformations, etc. These more abstract biological information helps to have a more basic understanding of the efficacy and safety of drugs, so it is necessary to extract new variables from a large number of metabolite ion information that can reflect the characteristics of the above abstract biological information and their spatial changes, in order to more objectively evaluate and compare the efficacy and potential toxic side effects of preclinical candidate drugs. Currently, there is no related mass spectrometry imaging method and software development to visually display the spatially distinguishable overall metabolic profile state and high-order metabolomic features of specific metabolic functions, therefore, the present application intends to establish a spatial metabolomic high-order feature extraction, measurement and mass spectrometry imaging visualization method based on multi-dimensional vector distance measurement. SUMMARY

[0004] The present application aims to solve three technical problems: (1) biological tissue homogeneous region division based on spatial metabolic profile difference; (2) establishment of a mass spectrometry imaging metabolite ion information library; (3) extraction of spatial metabolomic high-order features and measurement of metabolic disturbance score, the overall technical process diagram of the method is shown in the accompanyingFigure 1 .

[0005] To solve the above technical problems, the present application provides detailed technical solutions as follows:

[0006] Step 1. Mass spectrometry imaging data collection of biological tissue samples:

[0007] Desorption ionization is performed on the intact whole animal tissue frozen section or tumor tissue frozen section using a vacuum or atmospheric open ion source such as air force assisted ionization (AFAI), desorption electrospray ionization (DESI), matrix assisted laser desorption ionization (MALDI), secondary ion mass spectrometry (SIMS), and mass spectrometry imaging data is collected by methods such as single channel, positive or negative full scan collection, or multi-channel, alternatively positive and negative full scan collection.

[0008] Step 2. Preprocessing of mass spectrometry imaging data structure:

[0009] This step is divided into three sub-steps to preprocess the mass spectrometry imaging data collected in step 1

[0010] Data structure conversion: the mass spectrometry imaging raw data collected from the biological tissue sample is a multi-dimensional array composed of collection rows, collection columns, and different mass-to-charge ratios. First, the multi-dimensional array is reduced in dimension to a two-dimensional data array composed of all pixel points of the biological tissue sample and all metabolite ion intensities, and the physical space position of each pixel point is indexed and coded by row and column to facilitate extraction of certain column metabolite ion intensity information of the two-dimensional data array for image reconstruction.

[0011] Homogeneous region division: the metabolite ion intensity in the two-dimensional data array obtained in step 2.1 is used as the input characteristic variable, and the pixel point region identification of the entire biological tissue sample two-dimensional data matrix is performed by machine learning method, to realize the homogeneous region division of the entire biological tissue sample, and mark the homogeneous region class label of each pixel point;

[0012] Splitting and recombination of biological tissue sample matrix: according to the homogeneous region class label of each pixel point in the biological tissue sample obtained in step 2.2, the two-dimensional data matrix of each biological tissue sample is split into several homogeneous region sub-matrices, and the homogeneous region sub-matrices from different biological tissue samples are recombined into a homogeneous region matrix (see Figure 2 );

[0013] Step 3. Establishment of metabolite ion information library of mass spectrometry imaging: select the ion mass-to-charge ratio with a relative abundance greater than 0.1% in the average mass spectrum of the biological tissue sample, and after searching the open source metabolomics database, label and identify the metabolite information of all ions; construct the chemical and biological information data of the metabolite ion of mass spectrometry imaging, the entry includes ion accurate mass-to-charge ratio, adduct ion type, metabolite structure type, metabolic pathway, biological function, biological transformation type, and the information entries are edited and indexed according to categories, which are used for the extraction of a certain type of characteristic metabolite in the subsequent step. Specific examples of the information library are shown in the following table 2. Figure 3 The entries of biological function, metabolite type, biological transformation and metabolic pathway are shown in Table 1.

[0014] Step 4. Selection and extraction of high-order feature information, which is divided into two sub-steps:

[0015] 4.1 Selection of high-order feature information: there are two specific implementation methods for this step: one is non-target high-order feature information extraction, which is suitable for the preliminary stage of candidate drug in vivo action prediction and evaluation, i.e. there is no specific concern about metabolic pathways, biological functions, metabolite types or biological transformation types, so the whole metabolic profile needs to be represented by all ion information in the metabolite ion information library of mass spectrometry imaging constructed in step 3, as the regional whole metabolic profile matrix; the other is targeted high-order feature information extraction, i.e. only the retrieval index of the concerned metabolic pathway, biological function, biological transformation type, metabolite structure type and other biological information is extracted to form a retrieval vector, and the retrieval vector is used to extract the primary characteristic metabolite group corresponding to each pixel in the regional matrix to form a regional feature matrix.

[0016] 4.2 Extraction of high-order feature information: linear or nonlinear dimensionality reduction is performed on the ion intensity information of the whole metabolic profile or the primary characteristic metabolite group described in step 4.1 to generate new variables that can reflect the whole metabolic profile or specific metabolic feature differences. The dimensionality reduction methods include principal component extraction, non-negative matrix factorization and t-distributed neighborhood embedding.

[0017] Step 5. Measure the metabolic disturbance change in the high-order feature space:

[0018] After dimensionality reduction of the multi-dimensional feature vector composed of multiple metabolites to two-dimensional or three-dimensional high-order feature information space, the biological tissue sample pixels at different positions in the two-dimensional or three-dimensional high-order feature information space are clustered, the cluster center of the control group pixel cluster is determined and calculated, and the distance between each pixel of the other group of biological tissue samples and the cluster center is calculated as the metabolic disturbance score of the spatial position corresponding to each pixel under the action of drug intervention.

[0019] Step 6. Visualization of metabolic disturbance change:

[0020] After obtaining the pixel metabolic disturbance score in each homogeneous region matrix in step 5, the pixel metabolic disturbance score is filled into the corresponding physical space position according to the biological tissue sample and pixel position index of each pixel, so that the metabolic disturbance image of each biological tissue sample under the action of the drug can be obtained.

[0021] Beneficial technical effects:

[0022] The extraction, measurement and mass spectrometry imaging visualization method of the spatial metabolome high-order characteristics established by the application can produce beneficial technical effects in the preclinical efficacy and toxic and side effect evaluation of candidate drugs and molecular mechanism research. Specifically as follows: (1) visualizing the prediction of the action site and intensity of the candidate drug in the test animal body; (2) visualizing the prediction of the toxic and side effects of the candidate drug on non-target organs or tissues in the test animal body; (3) quantitatively evaluating the high-order metabolic characteristics of the metabolic pathways, metabolic functions, biological transformations and the like of the candidate drug in the test animal body; (4) helping to discover new drug action targets and potential molecular mechanisms. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 . The overall animal spatial metabolome high-level feature extraction, measurement and mass spectrometry imaging visualization flowchart. (a) Collecting spatially resolved metabolomics data from control group and test group animal sections under positive / negative switching mass spectrometry acquisition mode; (b) converting the mass spectrometry images of each pixel collected into control group and test group metabolomics data matrices, each pixel is assigned a label index belonging to different regions, and several rows of pixel data from the same type of region in the two matrices are extracted and reconstructed into a new homogeneous region matrix; (c) extracting several metabolite ion data (columns) belonging to the same biological function, metabolic pathway, biological transformation or metabolite category according to the metabolite retrieval vector in the pre-constructed metabolite information library, to constitute a regional feature matrix; (d) extracting two high-order characteristics from the high-dimensional data of the regional feature matrix using a dimension reduction method; (e) mapping each pixel to a two-dimensional or three-dimensional high-order characteristic space according to its high-order characteristic value, calculating the cluster center of the reference group pixel cluster in the space, and the distance between each pixel point of the test group and the cluster center; (f) obtaining the metabolic disturbance score of the main organ by calculating the average distance of the test group pixels; (g) obtaining the overall animal spatial metabolic disturbance image by reconstructing the distance values of all biological tissue pixel points according to the physical space position.

[0024] Figure 2 . Schematic diagram of mass spectrometry imaging data structure conversion.

[0025] Figure 3. Partial schematic diagram of information composition of the metabolite ion information library of mass spectrometry imaging.

[0026] Figure 4 . Homogeneous region identification results of whole animal and tumor tissue. (a) Whole animal homogeneous region division predicted by machine learning method; (b) Tumor tissue sub-region division predicted by machine learning method; (c) Prediction results of developed machine learning model on each homogeneous region of whole animal and tumor tissue sub-region. TPR: true positive rate; FNR: false negative rate; PPV: positive predictive value; FDR: false discovery rate.

[0027] Figure 5 . Metabolic disturbance score images for characterizing the effects of drugs on the overall metabolic state of each organ of the body. (a) Healthy control group; (b) tumor model group; (c) positive control drug group; (d) low-dose test drug group; (e) medium-dose test drug group; (f) high-dose test drug group

[0028] Figure 6 Metabolic disturbance score images for characterizing the strength of the anti-tumor metabolic effects of drugs. (a) Healthy control group; (b) tumor model group; (c) positive control drug group; (d) low-dose test drug group; (e) medium-dose test drug group; (f) high-dose test drug group

[0029] Figure 7 Metabolic disturbance score ranking of specific metabolic pathways of tumor. (a) Ranking of the effects of positive control drug and test drug on different types of biotransformation; (b) Ranking of the effects of positive control drug and test drug on different metabolic pathways; (c) Ranking of the effects of positive control drug and test drug on different types of metabolites. PTS: test drug group; PTX: positive control drug group DETAILED DESCRIPTION

[0030] The present application takes the anti-tumor drug paclitaxel (PTX) and the candidate drug paclitaxel derivative (PTR) as examples, and the overall process is as shown in Figure 1 The specific implementation steps are described in detail, but this embodiment does not limit the present application in any way.

[0031] 1. Preparation of biological tissue frozen sections

[0032] Leica CM3600 large-scale frozen section machine was used to prepare whole animal frozen sections with a thickness of 25 μm, and Leica CM1860 sectioning machine was used to prepare allogeneic tumor frozen sections with a thickness of 12 μm. The experimental samples were divided into healthy control group, tumor model animal group, positive control drug group (PTX) and low, medium and high dose test drug (PTR) group. The prepared biological tissue samples were first placed in a dry dish in a low temperature environment of -20 degrees Celsius and vacuum dried for 1 hour, then the dry dish was moved to room temperature environment and continued to dry for 2 hours. The temperature of the section was restored to room temperature and was ready for use.

[0033] 2. Mass spectrometry imaging data acquisition of whole animal and tumor tissue section samples

[0034] The AFAI ion source developed independently was combined with the electrostatic field orbitrap mass spectrometer (Q Orbitrap, Thermo Fisher Corporation) to perform data acquisition and imaging analysis of whole animal and tumor tissue section samples. The key parameters are as follows: the spray needle voltage and transmission tube voltage are set to ± 7.5 kV and ± 3.0 kV, respectively. The spray solvent is acetonitrile-water (7:3, v / v), and the spray solvent flow rate is 5 μL / min. The mass spectrometry data collection adopts an alternative acquisition method of positive full scan / negative full scan, and the mass-to-charge ratio range of metabolite ions is m / z 100-1000. The dynamic acquisition mode (AGC) is set to off to ensure constant mass spectrometry scanning speed, and the scanning speed is set to 100 ms. The lateral (X-axis) movement speed of the two-dimensional sample translation stage is 0.2 mm / sec, and the longitudinal (Y-axis) interline spacing is set to 0.2 mm, so that the metabolite components at different physical space positions of the biological tissue sample are sequentially desorbed, ionized and introduced into the mass spectrometer by the AFAI ion source spray needle.

[0035] The raw format original file is first converted to a cdf format file using the file converter provided with the Xcalibur software of Thermo Corporation, and then imported into the MATLAB2021a (MathWorks Corporation) environment. The cdfread function provided with MATLAB2021a is used to read and save the cdf file in batches. The subsequent steps are executed using the script code written by the author in the MATLAB2021a environment.

[0036] 3. Mass spectrometry imaging data structure conversion

[0037] A total of 1817 ions (1318 positive ions and 499 negative ions) included in the self-built mass spectrometry imaging metabolite ion information library were used as initial features, and the above-mentioned ions detected in 6 groups of whole animal and tumor tissue samples were reconstructed into images (mass-to-charge ratio tolerance was set to ± 0.005). First, all the images were saved in a multi-dimensional array, and then converted into a two-dimensional data matrix of "pixel-metabolite ion", which collected 1817 metabolite ion intensity data from a total of 12764 pixels from 6 different groups of biological samples.

[0038] 4. Division of homogeneous regions

[0039] With all 1817 ions as input variables, unsupervised clustering analysis of all pixel points in each group of samples was performed by t-SNE algorithm, under the guidance of the overall animal and tumor sample optical images, and the class labels of each pixel point were preliminarily marked, the pixel point data with obvious labeling errors was removed, and only the pixel point data with high labeling credibility was retained as the training data in the establishment of the machine learning model. The class labels are as follows: heart (label = 1), lung (label = 2), spleen (label = 3), kidney (label = 4), brain (label = 5), liver (label = 6), stomach (label = 7), intestine (label = 8), muscle (label = 9), thymus (label = 10), skin (label = 11), salivary gland (label = 12), mandible (label = 13), intestinal contents (label = 14), bone (label = 15), tumor parenchymal region (label = 16), necrotic region (label = 17), base (label = 18).

[0040] First, all pixel points were randomly divided into training set, validation set and test set according to the proportion of 70%:15%:15%, and the training and prediction effects of K-means clustering (KMeans), K-nearest neighbor algorithm (KNN), random forest (RF), support vector machine (SVM), linear discriminant analysis (LDA) and shallow neural network (NN) model were investigated in MATLAB2021a classification learning application (classification learner APP). Area under curve (AUC), true positive rate (TPR), false negative rate (FNR), positive prediction value (PPV), false discovery rate (FDR) and other indicators were used as indicators of the prediction performance of the training model for homogeneous regions. Finally, KNN was selected as the best machine learning model, and the prediction results are shown in FIG. 6. Figure 4 After machine learning, the pixel points in different regions of the 6 groups of whole animals and tumor tissues were accurately labeled. According to the region class labels of the pixel points, the two-dimensional data matrix of each biological tissue sample was divided into several sub-matrices, and the sub-matrices with the same class label were recombined into a region matrix. Finally, a total of 18 homogeneous region matrices were generated in the whole animal and tumor tissue.

[0041] Extraction of high-order feature information and measurement of metabolic disturbance change

[0042] In the primary stage of predicting and evaluating the in vivo effects of candidate drugs, all 1817 metabolite ions were selected to represent the overall metabolic profile. When evaluating specific biological functions, metabolic pathways, metabolite types, or biological transformation effects, the corresponding primary feature metabolite groups were extracted according to the 54 types recorded in the metabolite ion information library of mass spectrometry imaging (see Table 1). The 12764 pixel point data in the above 18 homogeneous region matrices were first reduced to a two-dimensional or three-dimensional high-order feature space using the t-SNE method. The cluster centers of the pixel point clusters in the control group were calculated using the Gaussian Mixed Model (GMM) method. The Euclidean distance was used as the measure to calculate the distance between each pixel point in the homogeneous region matrix and the cluster center, which was used as the metabolic disturbance score of the corresponding spatial position of each pixel point under the influence of drug intervention.

[0043] Table 1

[0044]

[0045] Visualization of metabolic disturbance changes

[0046] After obtaining the metabolic disturbance scores of the 12764 pixel points in the 18 homogeneous region matrices, the corresponding physical space positions were refilled according to the 6 groups of samples and the pixel position indexes within the groups to which each pixel point belonged, and the metabolic disturbance images of each biological tissue sample under the influence of drugs were obtained. The overall metabolic profile disturbance images at the overall animal level under the influence of tumor metabolism and drug intervention are shown in FIG. 2. Figure 5 From this figure, it can be judged that the test drug PTR has a significant effect on the metabolism of the spleen, and as the drug dose increases, it suggests that the test drug should pay special attention to the toxic and side effects of the spleen region. The overall metabolic profile disturbance images at the tumor level under the influence of tumor metabolism and drug intervention are shown in FIG. 3. Figure 6 From this figure, it can be judged that the positive control drug PTX and the test drug PTR can effectively inhibit tumor metabolism, and the two drugs have a large impact on the metabolic pathways, metabolite types, metabolic functions, and biological transformation pathways of tumor metabolism, which are shown in FIG. 4. Figure 7 ​

Claims

1. A method for extracting high-dimensional feature information of spatial metabolomics, measuring and visualizing mass spectrometry imaging, mainly comprising the following steps: (1) Mass spectrometry imaging data collection of biological tissue samples: desorbing and ionizing a plurality of groups of biological tissue samples using a vacuum or atmospheric open ion source, and collecting mass spectrometry data, and reconstructing metabolite ion images from the collected mass spectrometry data; (2) Preprocessing of mass spectrometry imaging data structure: (i) Data structure conversion: converting the collected set of metabolite ion images into a two-dimensional data matrix of "biological tissue pixel point-metabolite ion"; (ii) Pixel index editing: editing the row and column indices of the spatial positions of the biological tissue sample pixels in the image; (iii) Homogeneous region division: taking the metabolite ions collected by each pixel point of the biological tissue sample as input features, and performing cluster analysis on all pixel points of the biological tissue sample by machine learning method, and marking the region class label; (iv) Splitting and reorganizing of biological tissue sample matrix: according to the region class label of each pixel point in the biological tissue sample, the matrix of each biological tissue sample is split into a plurality of sub-matrices, and the sub-matrices of the same class label of each biological tissue sample are reorganized into a region matrix; (3) Establishment of metabolite ion information library of mass spectrometry imaging: constructing chemical and biological information data of all metabolites detected by mass spectrometry imaging, and editing search indexes for these information entries according to categories, which are used for subsequent extraction of a certain type of feature metabolite; (4) Selection of primary feature information and extraction of high-order feature information: taking all ions in the metabolite ion information library of mass spectrometry imaging as selected features to represent the overall metabolic profile, and forming a region overall metabolic profile matrix; or finding out the search indexes related to a certain specific metabolic pathway, biological function, biological transformation type or chemical structure type in the metabolite ion information library of mass spectrometry imaging, forming a search vector, and using the search vector to extract a group of primary feature metabolites corresponding to each pixel in the region matrix, forming a region feature matrix; reducing the multi-dimensional feature vector composed of feature metabolites in the region overall metabolic profile matrix or the region feature matrix to a two-dimensional or three-dimensional high-order feature information space; (5) Measuring the metabolic disturbance change in the high-order feature information space: clustering each group of biological tissue sample pixel points in the two-dimensional or three-dimensional high-order feature information space, calculating the cluster center of the pixel point cluster of the control group, and the distance between each pixel point of the other group of biological tissue samples and the cluster center as the metabolic disturbance score of the corresponding spatial position of each pixel point under the action of drugs or disease pathology; (6) Visualization of metabolic disturbance change: filling the metabolic disturbance scores of the pixel points in each region matrix into the corresponding spatial positions according to the biological tissue sample and pixel index to obtain the metabolic disturbance images of each biological tissue sample under the action of drugs.

2. The analytical method according to claim 1, characterized in that, In (1) the mass spectrometry imaging data acquisition of the biological tissue sample, the biological tissue sample includes solid organs or hollow organs from preclinical test animals, allogeneic transplant solid tumors, whole animal organ sections; the solid organs include heart, brain, kidney, liver, spleen, lung, thymus, the hollow organs include esophagus, stomach, small intestine, large intestine; the biological tissue sample includes postoperative tissues or endoscopic sampled tissues, including postoperative tumor tissues, paracancer tissues, normal tissues or other pathological tissues.

3. The analytical method according to claim 1, characterized in that, In (1) the mass spectrometry imaging data acquisition of the biological tissue sample, the vacuum or atmospheric open ion source includes air flow assisted ionization (AFAI), desorption electrospray ionization (DESI), matrix-assisted laser desorption ionization (MALDI), secondary ionization (SI), and laser ablation electrospray ionization (LAESI).

4. The analytical method according to claim 1, characterized in that, In (1) the mass spectrometry imaging data acquisition of the biological tissue sample, the mass spectrometry data acquisition is used to acquire information of metabolite ions, including targeted selected ion detection, multiple reaction monitoring, full scan acquisition, and targeted selected ion detection and positive and negative ion full scan alternating acquisition.

5. The analytical method according to claim 1, characterized in that, In (1) the mass spectrometry imaging data acquisition of the biological tissue sample, the metabolites include amino acids, oligopeptides, carbohydrates, purines, pyrimidines, acyl carnitines, fatty acids, glycerides, phospholipids, sphingolipids, cholesterol esters, vitamins, organic acids, polyamines, and ketone bodies.

6. The analytical method according to claim 1, characterized in that, In (2) the preprocessing of the mass spectrometry imaging data structure, the homogeneous region includes a region of the same type in the whole animal tissue sample, or a substructure region in an organ, or a micro region in a tissue, the region of the same type includes heart, liver, kidney, spleen, brain, lung, intestinal tract, stomach, skin, gonads, thymus, skeletal muscle, esophagus; the substructure region in an organ includes the cortical and medullary regions of the brain or kidney; the micro region in a tissue includes the neoplastic or necrotic regions of tumor tissues.

7. The analytical method according to claim 1, characterized in that, In (2) the preprocessing of the mass spectrometry imaging data structure, the machine learning method includes supervised learning algorithms, unsupervised learning algorithms, and deep learning models, wherein the supervised learning algorithms include K-nearest neighbor algorithm, random forest, support vector machine, discriminant analysis, and shallow neural network; the unsupervised learning algorithms include principal component analysis, K-means clustering analysis, hierarchical clustering analysis, self-organizing mapping, and Gaussian mixture model; the deep learning models include deep neural network and convolutional neural network.

8. The analytical method according to claim 1, characterized in that, In (3), the mass spectrometry imaging metabolite ion information library, the entry of the information library includes ion accurate mass-to-charge ratio, adduct ion type, chemical structure, functional group, metabolic pathway, biological function and biological transformation type.

9. The analytical method according to claim 1, characterized in that, In (4), the selection of primary feature information and the extraction of high-order feature information, the selection of primary feature information refers to the process of selecting all entries in the mass spectrometry imaging metabolite ion information library without target or selecting several metabolite ions associated in function, pathway, structure type or biological transformation with target to form a multi-dimensional primary feature vector.

10. The analytical method according to claim 1, characterized in that, In (4), the selection of primary feature information and the extraction of high-order feature information, the extraction of high-order feature information refers to the process of linear or nonlinear combination of the abundance information of all metabolites or several metabolites with certain association to generate new variables that can reflect the overall metabolic profile or specific metabolic feature difference.

11. The analytical method according to claim 1, characterized in that, In (4), the selection of primary feature information and the extraction of high-order feature information, the primary feature metabolite group refers to several metabolites participating in a metabolic pathway, bearing a certain biological function, having the same structure nucleus, functional group, or biological transformation.

12. The analytical method according to claim 1, characterized in that, In (5), the measurement of metabolic disturbance change in high-order feature space, the dimension reduction includes principal component extraction, non-negative matrix factorization, and t-distributed neighborhood embedding method.

13. The analytical method according to claim 1, characterized in that, In (5), the measurement of metabolic disturbance change in high-order feature space, the high-order feature space refers to the reduction of multi-dimensional information composed of several metabolites to two-dimensional or three-dimensional space through the high-order feature information extraction step in (4); 14. The analytical method according to claim 1, characterized in that, In (5), the measurement of metabolic disturbance change in high-order feature space, the control group refers to a healthy control group without modeling process, or a model control group without drug treatment after disease model construction; 15. The analytical method according to claim 1, characterized in that, In (5), the measurement of metabolic disturbance change in high-order feature space, the cluster center of the pixel point cluster refers to the geometric center or barycenter of the pixel point cluster from the same biological tissue source in the high-order feature space; 16. The analytical method according to claim 1, characterized in that, In (5), the measurement of metabolic disturbance change in high-order feature space, the distance refers to Euclidean distance, Mahalanobis distance, Manhattan distance, and the cosine of the included angle, and the measurement result of the metabolic disturbance score, which measures the degree of change in the characteristics of specific metabolite groups or overall metabolic profile under the action of drug intervention or disease pathology.

Images