Metabolic marker for auxiliary diagnosis of schizophrenia, construction method and application

Abstract

The invention provides a metabolic marker for auxiliary diagnosis of schizophrenia, a construction method and application. Wherein the metabolic marker comprises lysophosphatidylcholine (LPC) (17: 0 / 0: 0), oleamide, lysophosphatidyl ethanolamine LysoPE (18: 0 / 0: 0), palmitamide, lysophosphatidyl ethanolamine LysoPE (0: 0 / 20: 4), lysophosphatidyl ethanolamine LysoPE (22: 6 / 0: 0), p-methylphenol sulfate, lactic acid, 3-hydroxybutyric acid, lysophosphatidyl ethanolamine LysoPE (18: 2 / 0: 0), lysophosphatidyl ethanolamine LysoPE (16: 0 / 0: 0) and sphingosine 1-phosphate. By utilizing the metabolic marker, auxiliary diagnosis of schizophrenia can be rapidly and accurately completed in the early stage of disease attack.

Description

Technical Field

[0001] The present invention relates to the field of biomedical technology, and to the application of metabolic markers in disease diagnosis, and in particular to metabolic markers for auxiliary diagnosis of schizophrenia, and methods for constructing and applying them. Background Art

[0002] Schizophrenia is a chronic and severe mental illness with a lifetime risk of about 1%. Schizophrenia usually appears in adolescence and early adulthood, and is characterized by hallucinations, delusions, thought disorders, and cognitive impairments. Schizophrenia patients are prone to self-harm and suicide, and the risk of suicide death in their lifetime is 5% to 10%. In addition, due to poor diet and lifestyle habits, schizophrenia patients have a greatly increased risk of cardiovascular disease, type 2 diabetes, and metabolic syndrome, resulting in an average life expectancy that is about 15 years shorter than the general population.

[0003] The onset of schizophrenia is affected by genetic and environmental factors. Due to the heterogeneity of the etiology, the physiological and pathological mechanisms of the disease are still unclear. At present, the diagnostic criteria for schizophrenia are still based on clinical symptoms and rely on the judgment made by psychiatrists through interviews and observations of patients' behaviors. The lack of objective and accurate diagnostic indicators often leads to missed diagnoses and misdiagnoses, which is not conducive to early screening.

[0004] The metabolome is a dynamic whole of endogenous metabolites in organisms. By detecting changes in endogenous metabolites in the human body, the dynamic multi-parameter metabolic response of diseases can be monitored. Metabolomics uses technologies such as nuclear magnetic resonance and mass spectrometry to predict major diseases through biomarkers or conduct in-depth research on the occurrence and development mechanisms of major diseases, in order to provide a scientific basis for the prevention and treatment of diseases. However, this field still faces challenges in terms of metabolite sensitivity, data complexity, and omics data integration.

[0005] Machine learning algorithms have unique advantages in processing high-dimensional data sets. They can effectively mine biological information and truly reflect the biochemical processes under physiological and pathological conditions of the body. They are a new technical means to efficiently process high-throughput big data. In recent years, machine learning has been applied to metabolomics data analysis and disease classification, effectively improving the accuracy of metabolite identification, the efficiency of data analysis, and the accuracy of diagnostic methods.

[0006] Therefore, how to use metabolomics combined with machine learning to screen metabolic markers, effectively improve the accuracy and efficiency of diagnostic methods, and achieve auxiliary diagnosis of schizophrenia is a technical problem that needs to be urgently solved by technical personnel in this field. Summary of the invention

[0007] In order to solve the above problems, the present invention provides a group of metabolic markers, which are used for auxiliary diagnosis of schizophrenia.

[0008] The first object of the present invention is to provide a group of metabolic markers for auxiliary diagnosis of schizophrenia, wherein the metabolic markers include lysophosphatidylcholine LPC (17:0 / 0:0), oleamide, lysophosphatidylethanolamine LysoPE (18:0 / 0:0), palmitamide, lysophosphatidylethanolamine LysoPE (0:0 / 20:4), lysophosphatidylethanolamine LysoPE (22:6 / 0:0), p-methylphenol sulfate, lactic acid, 3-hydroxybutyric acid, lysophosphatidylethanolamine LysoPE (18:2 / 0:0), lysophosphatidylethanolamine LysoPE (16:0 / 0:0) and 1-phosphate sphingosine.

[0009] The second aspect of the present invention also provides a prediction model for assisting the diagnosis of schizophrenia, which includes the above-mentioned metabolic markers.

[0010] In certain embodiments, in the prediction model for assisting in the diagnosis of schizophrenia, the area under the curve of the prediction model is 0.976923.

[0011] In a third aspect, the present invention also provides a kit for auxiliary diagnosis of schizophrenia, comprising the above-mentioned metabolic markers.

[0012] In certain embodiments, the kit further comprises a standard substance of ketoprofen and L-2-chlorophenylalanine.

[0013] The fourth aspect of the present invention also provides a method for screening metabolic markers for auxiliary diagnosis of schizophrenia, comprising:

[0014] S1. Collect and process plasma samples to obtain a data set;

[0015] S2. performing ultra-high performance liquid chromatography-quadrupole / electrostatic field orbital trap tandem high-resolution mass spectrometry analysis on the plasma sample;

[0016] S3, using Thermo Xcalibur software for data acquisition and processing, using Compound Discovery software to retain time information, extract molecular formula, mass-to-charge ratio, and secondary ion fragment information;

[0017] S4. The information extracted in S3 is imported into SIMCA for multivariate statistical analysis. By establishing an orthogonal partial least squares discriminant analysis model, the differential metabolites are screened using the variable importance projection index as the screening index. The variable importance projection index includes P value, vip value and differential multiple FC; the screened differential metabolites with VIP>1, P<0.05 and differential multiple FC>1.25 or <0.8 are used as candidate data sets for selecting metabolic markers;

[0018] S5. Using an integrated machine learning method to screen biomarkers, and screening differential metabolic markers from the candidate data set;

[0019] S6. Splitting the data set into a training set and a test set, training 6 machine learning models one by one using a grid search with cross-validation, and selecting a final prediction model with the best prediction ability from the differential metabolic markers.

[0020] In certain embodiments, in the method for screening metabolic markers for auxiliary diagnosis of schizophrenia, the machine learning method includes extreme gradient boosting algorithm, random forest, minimum absolute value convergence and selection operator algorithm, ridge regression and linear support vector machine.

[0021] In certain embodiments, in the method for screening metabolic markers for auxiliary diagnosis of schizophrenia, the specific steps of screening biomarkers using an integrated machine learning method are:

[0022] Each algorithm conducts 10,000 independent random grouping experiments, and the top ten feature factors ranked in importance in each experiment are counted; the algorithms include extreme gradient boosting algorithm, random forest, minimum absolute value convergence, selection operator algorithm and ridge regression;

[0023] The importance of each feature factor is ranked according to its contribution to the prediction accuracy. A voting strategy is adopted to vote for the top ten feature factors in each algorithm. The voting results of 10,000 experiments are accumulated to select the top ten feature factors of each algorithm.

[0024] Combined with the linear support vector machine algorithm, the coefficients of irrelevant feature factors are reduced to 0 through regularization;

[0025] Delete the characteristic factors with coefficients of 0, and count the characteristic factors obtained from each experimental screening;

[0026] Vote for the characteristic factors screened in each experiment, accumulate the voting results of 10,000 experiments, and select the top ten characteristic factors;

[0027] The intersection of the voting results of the above algorithms was selected as the metabolic marker for the diagnosis of schizophrenia.

[0028] In certain embodiments, in the method for screening metabolic markers for auxiliary diagnosis of schizophrenia, the six machine learning models include decision tree, random forest, support vector machine, k-nearest neighbor, logistic regression and Gaussian naive Bayes.

[0029] In certain embodiments, in the method for screening metabolic markers for auxiliary diagnosis of schizophrenia, the machine learning model compares the predictive ability of each model based on accuracy, sensitivity, specificity, precision, recall rate, Matthews correlation coefficient and area under the ROC curve AUC, and selects the model with the best predictive ability as the final prediction model.

[0030] Beneficial effects:

[0031] The metabolic markers, construction methods and applications for auxiliary diagnosis of schizophrenia described in the present invention provide a group of metabolic markers. The above-mentioned metabolic markers can be used to quickly and accurately complete the auxiliary diagnosis of schizophrenia in the early stage of the disease. The accuracy of the auxiliary diagnosis reaches 0.934783, and the AUC value reaches 0.976923. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is the principal component analysis (PCA) score diagram of the positive ion mode of the schizophrenia patient group (Schizophrenia, SP) vs the healthy control group (Control, C) in the embodiment;

[0033] Figure 2 It is the principal component analysis (PCA) score diagram of the negative ion pattern of the schizophrenia patient group (Schizophrenia, SP) vs the healthy control group (Control, C) in the embodiment;

[0034] Figure 3 It is the orthogonal partial least squares discriminant analysis (OPLS-DA) score diagram of the positive ion mode of the schizophrenia patient group (Schizophrenia, SP) vs the healthy control group (Control, C) in the embodiment;

[0035] Figure 4 It is the orthogonal partial least squares discriminant analysis (OPLS-DA) score diagram of the negative ion mode of the schizophrenia patient group (Schizophrenia, SP) vs the healthy control group (Control, C) in the embodiment;

[0036] Figure 5It is a permutation test diagram of the OPLS-DA model of the positive ion mode of the schizophrenia patient group (Schizophrenia, SP) vs the healthy control group (Control, C) in the embodiment;

[0037] Figure 6 It is a permutation test diagram of the OPLS-DA model of the negative ion pattern of the schizophrenia patient group (Schizophrenia, SP) vs the healthy control group (Control, C) in the embodiment;

[0038] Figure 7 The feature importance ranking of the Extreme Gradient Boosting (XGBoost) algorithm in the embodiment;

[0039] Figure 8 Ranking the importance of features of the random forest algorithm (RF) in the embodiment;

[0040] Fig. 9 The feature importance ranking of the least absolute shrinkage and selection operator (LASSO) algorithm in the embodiment;

[0041] Fig.10 It is the feature importance ranking of the Ridge Regression algorithm in the embodiment;

[0042] Fig.11 In the embodiment, six machine learning models, including decision tree (DT), random forest (RF), support vector machine (SVM), k-nearest neighbor (kNN), logistic regression (LR) and Gaussian Naive Bayesian (GNB), were used to perform receiver operating characteristic curve (ROC) analysis on the diagnosis of schizophrenia in the test set. DETAILED DESCRIPTION

[0043] Various exemplary embodiments of the present invention will now be described in detail. This detailed description should not be considered as limiting the present invention, but should be understood as a more detailed description of certain aspects, features, and embodiments of the present invention.

[0044] It should be understood that the terms described in the present invention are only for describing special embodiments and are not intended to limit the present invention. In addition, for the numerical range in the present invention, it should be understood that the upper and lower limits of the scope and each intermediate value therebetween are specifically disclosed. Each smaller range between the intermediate value in any stated value or stated range and any other stated value or intermediate value in the described range is also included in the present invention. The upper and lower limits of these smaller ranges can be independently included or excluded in the scope.

[0045] Unless otherwise indicated, all technical and scientific terms used herein have the same meanings as those of ordinary skill in the art described in the present invention are generally understood. Although the present invention only describes preferred methods and materials, any methods and materials similar or equivalent to those described herein may also be used in the implementation or verification of the present invention. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods and / or materials related to the documents. In the event of a conflict with any incorporated document, the content of this specification shall prevail.

[0046] As used herein, "include", "including", "have", "contain", etc. are all open terms, which means including but not limited to. As used herein, "and / or" includes any or all combinations of the things mentioned. Unless otherwise specified, % refers to mass volume percentage.

[0047] like Figures 1 to 11 As shown, the first aspect of the present invention provides a group of metabolite markers for auxiliary diagnosis of schizophrenia, wherein the metabolite markers include lysophosphatidylcholine LPC (17:0 / 0:0), oleamide, lysophosphatidylethanolamine LysoPE (18:0 / 0:0), palmitamide, lysophosphatidylethanolamine LysoPE (0:0 / 20:4), lysophosphatidylethanolamine LysoPE (22:6 / 0:0), p-methylphenol sulfate, lactic acid, 3-hydroxybutyric acid, lysophosphatidylethanolamine LysoPE (18:2 / 0:0), lysophosphatidylethanolamine LysoPE (16:0 / 0:0) and 1-sphingosine phosphate.

[0048] A second aspect of the present invention provides a prediction model for assisting the diagnosis of schizophrenia, which includes the above-mentioned metabolic markers.

[0049] In the above scheme, the prediction model for assisting the diagnosis of schizophrenia has an area under the curve of 0.976923.

[0050] The third aspect of the present invention also provides a kit for auxiliary diagnosis of schizophrenia, which comprises the above-mentioned metabolic marker.

[0051] In the above scheme, the kit also includes ketoprofen and L-2-chlorophenylalanine standards.

[0052] The fourth aspect of the present invention also provides a method for screening metabolic markers for auxiliary diagnosis of schizophrenia, comprising:

[0053] S1. Collect and process plasma samples to obtain a dataset;

[0054] S2. performing ultra-high performance liquid chromatography-quadrupole / electrostatic field orbital trap tandem high-resolution mass spectrometry analysis on the plasma sample;

[0055] S3, using Thermo Xcalibur software for data acquisition and processing, using Compound Discovery software to retain time information, extract molecular formula, mass-to-charge ratio, and secondary ion fragment information;

[0056] S4. The information extracted in S3 is imported into SIMCA for multivariate statistical analysis. By establishing an orthogonal partial least squares discriminant analysis model, the differential metabolites are screened using the variable importance projection index as the screening index. The variable importance projection index includes P value, vip value and differential multiple FC; the screened differential metabolites with VIP>1, P<0.05 and differential multiple FC>1.25 or <0.8 are used as candidate data sets for selecting metabolic markers;

[0057] S5. Using an integrated machine learning method to screen biomarkers, and screening differential metabolic markers from the candidate data set;

[0058] S6. Splitting the data set into a training set and a test set, training 6 machine learning models one by one using a grid search with cross-validation, and selecting a final prediction model with the best prediction ability from the differential metabolic markers.

[0059] In the above scheme, the screening method of metabolic markers for auxiliary diagnosis of schizophrenia is characterized in that the machine learning method includes extreme gradient boosting algorithm, random forest, minimum absolute value convergence and selection operator algorithm, ridge regression and linear support vector machine.

[0060] In the above scheme, the screening method of metabolic markers for auxiliary diagnosis of schizophrenia is characterized in that the specific steps of screening biomarkers using an integrated machine learning method are:

[0061] Each algorithm conducts 10,000 independent random grouping experiments, and the top ten feature factors ranked in importance in each experiment are counted; the algorithms include extreme gradient boosting algorithm, random forest, minimum absolute value convergence, selection operator algorithm and ridge regression;

[0062] The importance of each feature factor is ranked according to its contribution to the prediction accuracy. A voting strategy is adopted to vote for the top ten feature factors in each algorithm. The voting results of 10,000 experiments are accumulated to select the top ten feature factors of each algorithm.

[0063] Combined with the linear support vector machine algorithm, the coefficients of irrelevant feature factors are reduced to 0 through regularization;

[0064] Delete the characteristic factors with coefficients of 0, and count the characteristic factors obtained from each experimental screening;

[0065] Vote for the characteristic factors screened in each experiment, accumulate the voting results of 10,000 experiments, and select the top ten characteristic factors;

[0066] The intersection of the voting results of the above algorithms was selected as the metabolic marker for the diagnosis of schizophrenia.

[0067] In the above scheme, the screening method of metabolic markers for auxiliary diagnosis of schizophrenia is characterized in that the six machine learning models include decision tree, random forest, support vector machine, k-nearest neighbor, logistic regression and Gaussian naive Bayes.

[0068] In the above scheme, the method for screening metabolic markers for auxiliary diagnosis of schizophrenia is characterized in that the machine learning model compares the predictive ability of each model with accuracy, sensitivity, specificity, precision, recall rate, Matthews correlation coefficient and area under the ROC curve AUC, and selects the model with the best predictive ability as the final prediction model.

[0069] Example

[0070] In order to verify that the metabolic markers described in the present invention can quickly and accurately assist in the diagnosis of schizophrenia in the early stages of schizophrenia, the following experiment was conducted.

[0071] 1. Experimental instruments and reagents

[0072] The U1timate 3000 ultra-high performance liquid chromatography system (Dionex, USA) was connected in series with a Q-Exactive quadrupole-electrostatic field orbitrap high-resolution mass spectrometer (Thermo Fisher Scientific, USA). The chromatographic column used was ACQUITY HSS T3 1.8 μm (2.1×100 mm) (Waters, Milford, MA, USA); Heraeus Fresco 17 centrifuge (Thermo Fisher Scientific).

[0073] HPLC grade methanol, acetonitrile, and formic acid were purchased from Thermo Fisher, USA; L-2-chlorophenylalanine and ketoprofen standards were purchased from J&K Chemical (Beijing, China); Wahaha purified water was used in the experiments;

[0074] 2. Experimental Methods

[0075] 1. Experimental samples

[0076] After obtaining the patient's consent, the patients were included in the Diagnostic and Statistical Manual of Mental Disorders (5th Edition).

[0077] A total of 116 patients with schizophrenia admitted to the inpatient department of Zhumadian Second People's Hospital from May 2023 to August 2023 were collected from hospitalized patients who met the diagnostic criteria of schizophrenia according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, DSM-5. In addition, 110 healthy volunteers were collected from the physical examination department as a normal control group. All blood samples were collected from patients in the early morning when they were fasting. 3 mL of venous blood was collected from each subject and placed in an EDTA anticoagulant blood collection tube. After collection, the blood was centrifuged at 3000 r / min for 10 minutes at 4°C, and the supernatant (plasma) was aspirated and immediately frozen in a -80°C refrigerator after aliquoting.

[0078] 2. Preparation of plasma samples

[0079] After the plasma samples were taken out, they were placed on ice to thaw. After vortex mixing, 50 μL of plasma samples were drawn into a 1.5 mL centrifuge tube, and 150 μL of methanol solution containing internal standards (containing ketoprofen 500 ng / mL, L-2-chlorophenylalanine 50 ng / mL) was added. Vortex mixing was performed for 30 seconds, and centrifugation was performed at 13000 rpm for 10 min (4 ° C). The supernatant was drawn into the injection vial. 10 μL was drawn from each plasma sample and vortexed and mixed. The QC sample pretreatment was performed in the same way as above. In order to ensure the reliability of the data, QC sample analysis was interspersed in the process of all sample metabolomics data collection. Before sample analysis, 5 QC samples were tested continuously. After the instrument was stable, the sample was analyzed and the QC solution was tested every 5 samples.

[0080] 3. Detection methods of metabolites in plasma

[0081] Chromatographic conditions Chromatographic separation was performed by ultra-high performance liquid chromatography (UPLC, Waters Ultimate 3000, USA), the chromatographic column was ACQUITY UPLC BEH C18 (50×2.1 mm, 1.7 μm), and the column temperature was 40°C. The mobile phase composition was acetonitrile in phase A and 0.1% formic acid in aqueous solution in phase B. Gradient elution conditions: 0-0.5 min, 5% A; 0.5-1.0 min, 5%-60% A; 0.5-1.0 min, 5%-60% A; 1.0-7.0 min, 60%-80% A; 7.0-9.0 min, 80%-100% A; 9.0-11.0 min, 100% A; 11.0-13.0 min, 5% A; flow rate 0.2 mL / min, the post-column effluent was directly detected by mass spectrometry without splitting.

[0082] Mass spectrometry conditions Mass spectrometry analysis was performed using a Q Exactive quadrupole-electrostatic field orbital trap high-resolution mass spectrometer, and the ion source was a heatable electrospray ionization source (HESI). The auxiliary gas temperature was 300℃, the ion source temperature was 350℃, the capillary temperature was 320℃, the auxiliary gas flow rate was 10uL / min, the mass spectrometry analysis scanning mode was full scan / ddms2, the primary mass spectrometry resolution was 70000, the secondary mass spectrometry resolution was 17500, and the mass-to-charge ratio window width was set to 8; the collision energy gradient was 20, 40 and 60eV; the scanning range in positive and negative ion modes was 80.00~1200.00m / z, and the spray voltage and sheath gas flow rate were 3.50kV, 40μL / min and 2.80kV, 38μL / min, respectively. All samples were randomly injected.

[0083] 4. Data processing and analysis

[0084] All data were collected and processed by Thermo Xcalibur software (version 3.0, Thermo Scientific, USA), and then Compound Discovery software (version 3.0, Thermo Scientific, USA) was used for peak correction, peak matching and peak alignment information extraction. The specific parameters are as follows: the retention time (RT) width is set to 0.1min, the mass width is 5ppm, and the intensity threshold for filtering the ion peak is 1,000,000 in both positive and negative ion modes. The data set was imported into the multivariate statistical analysis software SIMCA (version 14.1, Umetrics, Sweden) for orthogonal partial least squares discriminant analysis (OPLS-DA), as well as 200 permutation tests to assess the risk of model overfitting. By establishing an OPLS-DA model, the variable weight importance ranking (VIP value) was obtained, and the metabolites with VIP>1 were subjected to an independent sample t-test to screen out statistically significant (P<0.05) differential metabolites, and the primary accurate m / z, secondary fragment ion map, retention time (RT) and other information collected by mass spectrometry were compared with the database for identification. The databases include ChemSpider and MassList with primary mass spectrometry information; mzCloud with primary and secondary mass spectrometry information, the laboratory-built database mzVault, the Human Metabolome database (HMBD, http: / / hmdb.ca / ), MassBank (https: / / massbank.eu / MassBank / ), and the PubChem compound database (https: / / pubchem.ncbi.nlm.nih.gov / ). Finally, the compound structure was confirmed by comparison with the standard.

[0085] Differential metabolites with VIP>1, P<0.05 and FC>1.25 or <0.8 were used as the data set for selecting metabolic markers. The ensemble method was used for biomarker screening, making full use of the advantages of individual techniques, overcoming their limitations and improving reliability. Five algorithms, including Extreme Gradient Boosting (XGBoost), random forest (RF), least absolute value convergence and selection operator (LASSO), Ridge Regression and Linear Support Vector Classification (LinearSVC), were combined for feature selection and dimensionality reduction. Then, based on the feature factors screened by each algorithm, a voting strategy was applied for final prediction. The voting process relied on the combined performance of the five algorithms, thus avoiding large errors or misclassifications that may be caused by a single model. Finally, based on the voting results, metabolic markers for the diagnosis of schizophrenia were screened.

[0086] The diagnostic metabolic markers of schizophrenia in the previous step were used as features, and the disease conditions were used as labels. The data set was randomly split into a training set and a test set at a ratio of 80% and 20%, and six machine learning models were trained one by one using a grid search with cross-validation: decision tree (DT), random forest (RF), support vector machine (SVM), k-nearest neighbor (kNN), logit regression (LR) and Gaussian Naive Bayesian (GNB). The optimal parameter combination of each model was obtained using accuracy as an indicator. The prediction ability of each model was compared using accuracy, sensitivity, specificity, precision, recall, Matthews coefficient (MCC) and AUC, and the model with the best prediction ability was selected as the final prediction model.

[0087] 3. Experimental Results

[0088] Through principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) (such as Figures 1 to 6 , Figures 1 to 6A represents positive ion mode, and B represents negative ion mode). It can be seen that the schizophrenia group (Schizophrenia, SP) and the normal control group (Control, C) composed of healthy people can be clearly separated and clustered, indicating that the endogenous metabolites of schizophrenia patients and healthy people are significantly different. The model evaluation parameters of OPLS-DA show that the model has good explanatory and predictive abilities. The response ranking test (RPT) method was used to perform permutation verification after 200 modelings, which confirmed that the model was stable and reliable without overfitting.

[0089] The differential metabolites with VIP>1, P<0.05 and FC>1.25 or <0.8 were used as the data set for selecting metabolic markers to screen metabolic markers. Four machine learning methods, XGBoost, RF, LASSO and Ridge Regression, screened the top ten predictors according to the feature importance scores (see Figures 7-10 ). At the same time, combined with the LinearSVC algorithm, different feature screening strategies are adopted. The coefficients of irrelevant feature factors are reduced to the minimum or to 0 by using the regularization method, and irrelevant feature factors are deleted. Then, based on the feature factors screened by these different algorithms, a voting strategy is applied for final prediction. The voting process relies on the combined performance of these five algorithms, thereby avoiding large errors or misclassifications that may be caused by a single model. Finally, based on the voting results, metabolic markers for the diagnosis of schizophrenia were screened, and the results are shown in Table 1. Compared with the normal control group, lysophosphatidylcholine LPC (17:0 / 0:0), oleic acid amide, lysophosphatidylethanolamine LysoPE (18:0 / 0:0), palmitamide, lysophosphatidylethanolamine LysoPE (0:0 / 20:4), lysophosphatidylethanolamine LysoPE (22:6 / 0:0), p-methylphenol sulfate, lactic acid, 3-hydroxybutyric acid, lysophosphatidylethanolamine LysoPE (18:2 / 0:0), lysophosphatidylethanolamine LysoPE (16:0 / 0:0) and 1-phosphate sphingosine can be used as potential diagnostic markers for schizophrenia.

[0090] Table 1 Plasma metabolic markers in patients with schizophrenia

[0091]

[0092] The diagnostic model was established with the metabolic marker group in the previous step as the feature and the disease condition as the label, and the performance indicators of the six machine learning models were calculated (see Table 2). By comparing the accuracy, sensitivity, specificity, precision, recall rate, Matthews coefficient (MCC) and AUC, it can be found that the logistic regression (LR) achieved the best performance in predicting schizophrenia, with an accuracy of 0.934783 and an AUC value of 0.976923 (see Fig.11 ).

[0093] Table 2 Performance indicators of six machine learning models: decision tree (DT), random forest (RF), support vector machine (SVM), k-nearest neighbor (kNN), logistic regression (LR) and Gaussian Naive Bayesian (GNB)

[0094] Table 2 Performance indicators of six machine learning models

[0095]

[0096] In summary, lysophosphatidylcholine LPC (17:0 / 0:0), oleic acid amide, lysophosphatidylethanolamine LysoPE (18:0 / 0:0), palmitamide, lysophosphatidylethanolamine LysoPE (0:0 / 20:4), lysophosphatidylethanolamine LysoPE (22:6 / 0:0), p-methylphenol sulfate, lactic acid, 3-hydroxybutyric acid, lysophosphatidylethanolamine LysoPE (18:2 / 0:0), lysophosphatidylethanolamine LysoPE (16:0 / 0:0) and 1-phosphate sphingosine can be used in combination for the diagnosis of schizophrenia with high diagnostic accuracy. A person skilled in the art can develop it into a diagnostic kit for schizophrenia, which contains lysophosphatidylcholine LPC (17:0 / 0:0), oleic acid amide, lysophosphatidylethanolamine LysoPE (18:0 / 0:0), palmitamide, lysophosphatidylethanolamine LysoPE (0:0 / 20:4), lysophosphatidylethanolamine LysoPE (22:6 / 0:0), p-methylphenol sulfate, lactic acid, 3-hydroxybutyric acid, lysophosphatidylethanolamine LysoPE (18:2 / 0:0), lysophosphatidylethanolamine LysoPE (16:0 / 0:0) and 1- The standards of 12 compounds, including sphingosine phosphate, are used to detect lysophosphatidylcholine LPC (17:0 / 0:0), oleamide, lysophosphatidylethanolamine LysoPE (18:0 / 0:0), palmitamide, lysophosphatidylethanolamine LysoPE (0:0 / 20:4), lysophosphatidylethanolamine LysoPE (22:6 / 0:0), p-methylphenol sulfate, lactic acid, 3-hydroxybutyric acid, lysophosphatidylethanolamine LysoPE (18:2 / 0:0), lysophosphatidylethanolamine LysoPE (16:0 / 0:0) and 1-sphingosine phosphate in the sample to be tested. The kit can also use the standards of ketoprofen and L-2-chlorophenylalanine as internal standards for detection.

[0097] It will be apparent to those skilled in the art that various modifications and variations may be made to the specific embodiments of the present invention description without departing from the scope or spirit of the present invention. Other embodiments derived from the present invention description will be apparent to those skilled in the art. The present application description and examples are exemplary only.

Claims

1. A group of metabolic markers for use in the auxiliary diagnosis of schizophrenia, characterized in that: The metabolic markers include lysophosphatidylcholine LPC (17:0 / 0:0), oleamide, lysophosphatidylethanolamine LysoPE (18:0 / 0:0), palmitamide, lysophosphatidylethanolamine LysoPE (0:0 / 20:4), lysophosphatidylethanolamine LysoPE (22:6 / 0:0), p-methylphenol sulfate, lactic acid, 3-hydroxybutyric acid, lysophosphatidylethanolamine LysoPE (18:2 / 0:0), lysophosphatidylethanolamine LysoPE (16:0 / 0:0) and 1-phosphate sphingosine.

2. A prediction model for assisting the diagnosis of schizophrenia, characterized in that: Comprising the metabolic marker as claimed in claim 1.

3. A prediction model for assisting diagnosis of schizophrenia as claimed in claim 2, characterized in that: The area under the curve of this prediction model is 0.976923.

4. A kit for auxiliary diagnosis of schizophrenia, characterized in that: Comprising the metabolic marker according to claim 1.

5. The kit according to claim 3, characterized in that The kit also includes standards of ketoprofen and L-2-chlorophenylalanine.

6. A method for screening metabolic markers for auxiliary diagnosis of schizophrenia, characterized in that: include: S1. Collect and process plasma samples to obtain a dataset; S2. performing ultra-high performance liquid chromatography-quadrupole / electrostatic field orbital trap tandem high-resolution mass spectrometry analysis on the plasma sample; S3, using Thermo Xcalibur software for data acquisition and processing, using Compound Discovery software to retain time information, extract molecular formula, mass-to-charge ratio, and secondary ion fragment information; S4. The information extracted in S3 is imported into SIMCA for multivariate statistical analysis. By establishing an orthogonal partial least squares discriminant analysis model, the differential metabolites are screened using the variable importance projection index as the screening index. The variable importance projection index includes P value, vip value and differential multiple FC; the screened differential metabolites with VIP>1, P<0.05 and differential multiple FC>1.25 or <0.8 are used as candidate data sets for selecting metabolic markers; S5. Using an integrated machine learning method to screen biomarkers, and screening differential metabolic markers from the candidate data set; S6. Splitting the data set into a training set and a test set, training 6 machine learning models one by one using a grid search with cross-validation, and selecting a final prediction model with the best prediction ability from the differential metabolic markers.

7. The method for screening metabolic markers for auxiliary diagnosis of schizophrenia according to claim 6, characterized in that: The machine learning methods include extreme gradient boosting algorithm, random forest, least absolute value convergence and selection operator algorithm, ridge regression and linear support vector machine.

8. The method for screening metabolic markers for auxiliary diagnosis of schizophrenia according to claim 7, characterized in that: The specific steps for biomarker screening using integrated machine learning methods are: Each algorithm conducts 10,000 independent random grouping experiments, and the top ten feature factors ranked in importance in each experiment are counted; the algorithms include extreme gradient boosting algorithm, random forest, minimum absolute value convergence, selection operator algorithm and ridge regression; Sort the importance of each feature factor according to its contribution to the prediction accuracy; A voting strategy is used to vote for the top ten feature factors in each algorithm; The voting results of 10,000 experiments are accumulated, and the top ten feature factors of each algorithm are selected; Combined with the linear support vector machine algorithm, the coefficients of irrelevant feature factors are reduced to 0 through regularization; Delete the characteristic factors with coefficients of 0, and count the characteristic factors obtained from each experimental screening; Vote for the characteristic factors screened in each experiment, accumulate the voting results of 10,000 experiments, and select the top ten characteristic factors; The intersection of the voting results of the above algorithms was selected as the metabolic marker for the diagnosis of schizophrenia.

9. The method for screening metabolic markers for auxiliary diagnosis of schizophrenia according to claim 6, characterized in that: The six machine learning models include decision tree, random forest, support vector machine, k-nearest neighbor, logistic regression and Gaussian naive Bayes.

10. The method for screening metabolic markers for auxiliary diagnosis of schizophrenia according to claim 8, characterized in that: The machine learning model compares the prediction capabilities of each model using accuracy, sensitivity, specificity, precision, recall, Matthews correlation coefficient, and area under the ROC curve AUC, and selects the model with the best prediction capability as the final prediction model.

Images