Lipid metabolite combination for early diagnosis marker of vkh and application thereof

Abstract

The present application relates to the technical field of biological medicine, in particular to a lipid metabolite combination for early diagnosis of VKH and application thereof, by obtaining plasma samples of patients with initial acute stage (early stage) VKH syndrome and healthy controls, and constructing sample-full lipid quantitative expression matrix, screening differential lipids through PCA unsupervised analysis and OPLS-DA supervised model, further screening lipids with high diagnostic performance by using elastic net logistic regression model, finally obtaining a diagnostic marker combination composed of 25 lipid metabolites, solving the technical problems of existing VKH diagnosis technology, such as invasiveness, insufficient sensitivity and stability of protein marker detection, limited diagnostic efficiency, lack of systematicness and standardized process in lipidomics research, achieving the effect of non-invasive, efficient, high sensitivity and high specificity of early auxiliary diagnosis of VKH syndrome, and providing objective and reliable technical means for precise identification of atypical cases and large-scale clinical screening.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, specifically to a lipid metabolite combination for early diagnostic markers of VKH and its application. Background Technology

[0002] Vogt-Koyanagi-Harada syndrome (VKH) is a blinding autoimmune disease that commonly affects people of Asian, Hispanic, Native American, and Middle Eastern descent. The onset age is typically between 20 and 50 years old. The disease is characterized by bilateral granulomatous panuveitis, often accompanied by involvement of multiple systems including the meninges, hearing, and skin. VKH syndrome has a clear early treatment window; timely intervention can effectively control inflammation, prevent irreversible visual impairment, and even achieve clinical cure. Delayed diagnosis, however, easily leads to recurrent episodes and ultimately blindness.

[0003] Traditional clinical diagnosis of VKH syndrome relies primarily on a comprehensive assessment of cerebrospinal fluid examination, ocular imaging, and clinical signs. While cerebrospinal fluid examination is valuable for early diagnosis, lumbar puncture is an invasive procedure with potential risks of infection and bleeding, making it unsuitable as a routine clinical screening method. Furthermore, some atypical cases present with subtle clinical manifestations, easily confused with ocular diseases such as posterior scleritis, Behcet's disease uveitis, and ankylosing spondylitis-associated uveitis, further complicating early and accurate diagnosis.

[0004] In recent years, plasma biomarkers have become a core research direction in the diagnosis of Vogt-Koyanagi-Harada syndrome due to their non-invasive sampling and ease of dynamic monitoring. Existing research largely focuses on plasma proteomics, and protein biomarkers such as APOH and STXBP1 have been screened for auxiliary diagnosis of Vogt-Koyanagi-Harada syndrome, as illustrated in publication CN117590006B, "Application of Biomarkers in the Preparation of Products for Diagnosing Vogt-Koyanagi-Harada Syndrome." However, those skilled in the art have discovered numerous technical shortcomings in the long-term use of these protein biomarkers:

[0005] (1) The high abundance of proteins in plasma can interfere with low abundance differential proteins, making it difficult for mass spectrometry techniques such as SWATH-MS to accurately identify and quantify key differential proteins related to diseases, thus limiting the sensitivity of biomarker screening.

[0006] (2) Existing protein markers such as APOH and STXBP1 have limited diagnostic efficacy when used alone, and the AUC value of combined detection is only 0.765, which is difficult to meet the needs of accurate diagnosis of atypical cases;

[0007] (3) The plasma proteomics detection process is complicated, requiring multiple steps such as protease digestion, mass spectrometry library construction, and ELISA verification. The detection cycle is long and the equipment and reagent costs are high, which limits its large-scale clinical application.

[0008] Therefore, developing a VKH-assisted diagnostic technology that is stable, highly sensitive, has excellent diagnostic efficacy, and is suitable for clinical application has become an urgent technical problem to be solved in this field. Summary of the Invention

[0009] The purpose of this invention is to address the shortcomings of existing technologies by providing a set of lipid metabolite biomarkers for early diagnosis of VKH and a lipidomics detection method for auxiliary diagnosis of VKH syndrome. This involves obtaining plasma samples from patients with newly diagnosed acute (early) VKH syndrome and healthy controls, constructing a sample-to-whole lipid quantitative expression matrix, screening differentially expressed lipids using unsupervised PCA analysis and a supervised OPLS-DA model, and further screening for lipids with high diagnostic performance using an elastic network logistic regression model. This results in a biomarker set consisting of 25 lipid metabolites, which solves the technical problems of invasiveness, insufficient sensitivity and stability of protein biomarker detection, limited diagnostic efficacy, and lack of systematic and standardized procedures in lipidomics research in existing VKH diagnostic techniques. This achieves a non-invasive, efficient, highly sensitive, and highly specific early auxiliary diagnostic effect for VKH syndrome, establishing a standardized VKH plasma lipidomics detection system, and providing an objective and reliable technical means for accurate identification of atypical cases and large-scale clinical screening.

[0010] The objective of this invention is achieved through the following approach:

[0011] A combination of lipid metabolites for early diagnostic biomarkers of VKH, comprising:

[0012] Cardiolipin CL (80:14) - Fatty acid FA (16:0), Phosphatidylserine PS (18:1-22:0), Phosphatidylserine PS (20:1-20:4), Phosphatidylcholine PC (17:0-18:0), Phosphatidylcholine PC (20:2-20:3), Phosphatidylethanolamine PE (22:4-22:4), Phosphatidylcholine PC (20:0-18:1), Phosphatidylcholine PC (18:2-22:4), Phosphatidylcholine PC (18:3-20:5), Cardiolipin CL (80:15) - Fatty acid FA (16:0), Phosphatidylcholine PC (18:1-20:2), Phosphatidylinositol PI (18:1-20:4), Phosphatidylcholine PC (18:3_18:3), phosphatidylcholine PC (20:2_20:2), phosphatidylcholine PC (18:2_20:3), phosphatidylethanolamine PE (18:2_20:0), phosphatidylcholine PC (20:3_20:5), phosphatidylserine PS (18:1_20:4), phosphatidylcholine PC (20:3_20:3), phosphatidylserine PS (20:0_20:4), phosphatidylserine PS (18:0_18:0), phosphatidylinositol PI (20:0_20:4), bisphosphatidylglycerol BMP (16:0_20:0), phosphatidylcholine PC (14:0_16:0), phosphatidylcholine PC (16:0_14:0).

[0013] Preferably, the lipid metabolites together constitute a plasma lipid biomarker, which is a lipid biomarker for early diagnosis of VKH.

[0014] Preferably, the plasma lipid marker is used in the preparation of a non-invasive screening or diagnostic kit for early VKH.

[0015] A kit for screening or diagnosing VKH, the kit containing reagents for detecting the content of the lipid metabolite combination, the kit being able to distinguish VKH syndrome patients from healthy individuals.

[0016] A lipidomics detection method for the auxiliary diagnosis of VKH syndrome includes the following steps:

[0017] 1) Collect blood samples from patients with VKH syndrome and healthy controls;

[0018] 2) Preprocess the blood samples collected in step 1), including plasma separation and lipid extraction, to obtain lipid samples;

[0019] 3) Combining ultra-high performance liquid chromatography and triple quadrupole mass spectrometry, quantitative lipid analysis was performed on lipid samples from VKH syndrome patients and healthy controls, respectively, to obtain several lipid peaks. The lipid peaks were then used to construct a sample-to-whole lipid quantitative expression matrix.

[0020] 4) By performing unsupervised and supervised analysis on the sample-to-whole lipid quantitative expression matrix, several lipids that can effectively distinguish the VKH syndrome patient group from the healthy control group and have biological associations were screened from the lipid samples as differential lipids;

[0021] 5) Based on the differentially expressed lipids obtained in step 4), an elastic network logistic regression diagnostic model was constructed, and several combinations of candidate diagnostic biomarkers for VKH syndrome were screened from the differentially expressed lipids using the elastic network logistic regression diagnostic model.

[0022] 6) By constructing a simple logistic regression model, the diagnostic efficiency and accuracy of the combination of candidate diagnostic biomarkers for VKH syndrome were verified, and the verified combination of candidate diagnostic biomarkers for VKH syndrome was used for early diagnosis of VKH.

[0023] Preferably, in step 3), the sample-to-whole lipid quantitative expression matrix is ​​constructed in the following manner:

[0024] 3-1) Two separation modes are set up: reversed-phase separation and hydrophilic interaction chromatography. The two separation schemes use corresponding dedicated chromatographic columns and mobile phase systems, which can cover lipids with different polarity ranges.

[0025] 3-2) The lipid samples were separated by ultra-high performance liquid chromatography using reversed-phase separation and hydrophilic interaction chromatography, respectively, to obtain two types of lipid elution fractions separated according to retention time;

[0026] 3-3) The lipid elution fraction obtained in step 3-2) is introduced into a triple quadrupole mass spectrometer, matching parameters are set, and various lipid characteristic peak signals are collected in multiple reaction monitoring quantification mode;

[0027] 3-4) Preprocess the collected lipid peak signals, remove lipid peaks with a missing rate of more than 50% in any group, and fill the remaining missing values ​​with 1 / 2 of the minimum value of the corresponding peak to obtain several lipid peaks that can be used for subsequent analysis.

[0028] 3-5) Construct a sample-to-whole lipid quantitative expression matrix by using lipid samples as rows and lipid peaks as columns, and configuring corresponding VKH syndrome patient or healthy control labels.

[0029] Preferably, in step 4), the differentially expressed lipids are obtained by screening in the following manner:

[0030] 4-1) Perform unsupervised principal component analysis (PCA) on the total lipid expression matrix of the VKH syndrome patient group and the healthy control group to generate PCA scatter plots. Determine whether the PCA scatter plots meet the following conditions. If they do, proceed to step 4-2). If they do not, the process is terminated and backtracked for rectification:

[0031] ① In the PCA scatter plot, the VKH syndrome patient group and the healthy control group showed a clear inter-group separation trend in the principal component space, proving that there was an overall statistical difference in the lipid profiles of the two groups.

[0032] ② In the PCA scatter plot, the samples within both the VKH syndrome patient group and the healthy control group showed good clustering with no extreme outliers, proving that the sample inclusion, data preprocessing and detection data quality were up to standard.

[0033] 4-2) Construct an OPLS-DA model for screening differential lipids, and use 7-fold cross-validation to calculate the R² and Q² values ​​of the OPLS-DA model to evaluate the fitting effect of the OPLS-DA model. Verify the robustness of the OPLS-DA model through several permutation tests to eliminate the risk of overfitting.

[0034] 4-3) Using the validated OPLS-DA model, calculate the variable projection importance (VIP) and p-value for each lipid in the sample-to-whole lipid quantitative expression matrix. Perform FDR correction on the p-value, and identify the lipids corresponding to the sample-to-whole lipid quantitative expression matrix as differentially expressed lipids when the VIP and p-values ​​meet the following conditions:

[0035] Condition 1: Variable projection importance (VIP) > 1;

[0036] Condition 2: P-value after FDR correction < 0.05;

[0037] 4-4) Perform metabolic pathway enrichment analysis on the differentially identified lipids in step 4-3) to identify key biological pathways associated with differentially identified metabolites and lipids, elucidate pathological mechanisms, and discover potential therapeutic targets.

[0038] Preferably, in step 5), the combination of candidate diagnostic biomarkers for VKH syndrome is obtained by screening in the following manner:

[0039] 5-1) Using binary logistic regression as the underlying framework of the model and elastic network regularization as the core optimization and feature selection kernel of the model, construct an elastic network regularized logistic regression model;

[0040] 5-2) Using the differentially expressed lipids obtained in step 4) as the training set, and combining 10-fold cross-validation, the elastic network regularized logistic regression model is trained to determine the optimal hyperparameters of the elastic network regularized logistic regression model, including the regularization strength and L1 / L2 mixed parameters.

[0041] 5-3) The diagnostic efficacy and overfitting risk of the elastic network regularized logistic regression model are verified by using ROC curve area analysis, two-tailed DeLong test and permutation test. If the model meets the preset diagnostic efficacy and robustness requirements, it is solidified as the optimal elastic network regularized logistic regression model. If the requirements are not met, the model is iteratively optimized by hyperparameter reconstruction.

[0042] 5-4) Extract the regression coefficient matrix of the optimal elastic network regularized logistic regression model, verify the regression coefficients of each differential lipid one by one, and screen out several differential lipids with non-zero regression coefficients as candidate diagnostic biomarkers for VKH syndrome.

[0043] Preferably, in step 5-2), the optimization principle of the hyperparameters is:

[0044] If the cross-validation AUC values ​​of multiple hyperparameters are similar, prioritize choosing a higher regularization strength λ to reduce model complexity;

[0045] If the regularization strength λ is the same, a higher L1 ratio should be preferred to balance prediction performance and model simplicity.

[0046] Preferably, in step 6), the diagnostic efficiency and accuracy of the combination of candidate diagnostic biomarkers for VKH syndrome are verified, specifically as follows:

[0047] 6-1) Based on the combinations of several candidate diagnostic biomarkers for VKH syndrome selected in step 5), a logistic regression model is constructed;

[0048] 6-2) Using the ROC curve and the corresponding area under the curve (AUC) as evaluation indicators, the diagnostic efficacy of the logistic regression model is evaluated through 10-fold cross-validation.

[0049] 6-3) The two-tailed DeLong test was used to calculate the P-value of the AUC of the logistic regression model. When P < 0.05, the logistic regression model was considered to be statistically significant, which was used to verify the accuracy of the logistic regression model in distinguishing VKH patients from healthy controls.

[0050] The beneficial effects of this invention include the following:

[0051] ① This invention uses lipids, a biomacromolecule that directly participates in the core links of VKH pathogenesis (immune cell activation, inflammatory response regulation, and maintenance of autoimmune tolerance), as the detection target. It uses abnormal plasma lipid metabolites, a core molecular feature of VKH disease, as the diagnostic basis, fundamentally avoiding the inherent defects of plasma proteomics detection, such as strong signal interference from high-abundance proteins and difficulty in accurately quantifying low-abundance disease-related biomarkers.

[0052] ② This invention uses a combination of unsupervised principal component analysis (PCA) and a supervised OPLS-DA model to screen differentially expressed lipids. First, unsupervised PCA analysis is used to control the quality of sample data and verify the overall differences between groups. Then, the supervised OPLS-DA model, combined with VIP value and FDR-corrected P value, is used to accurately screen disease-related differentially expressed lipids. Finally, an elastic network logistic regression model with optimized hyperparameters is optimized through 10-fold cross-validation to identify several (25) core lipid metabolite combinations with stable diagnostic value from a large number of lipid molecules. The biomarker screening process is scientific and rigorous, and its correlation with the pathological process of VKH and its disease specificity are superior to existing protein biomarkers such as APOH and STXBP1.

[0053] ③ This invention establishes a standardized detection protocol covering the entire process from plasma sample pretreatment, lipid extraction, dual-mode chromatographic separation, mass spectrometry MRM quantification to standardized data analysis and diagnostic model computation. This avoids the problems of cumbersome procedures, long detection cycles, and poor result stability associated with plasma proteomics detection, which requires protease digestion and multiple rounds of validation. Furthermore, plasma lipid metabolite detection inherently possesses the advantages of simple operation and high reproducibility. The detection protocol of this invention has high standardization and reproducibility, significantly lowering the threshold for clinical application. It perfectly meets the application needs of large-scale routine clinical screening and early diagnosis of VKH, overcoming the core bottleneck of existing lipidomics technologies that hinders clinical translation.

[0054] ④ The 25 lipid metabolite combinations obtained by screening in this invention are directly associated with the core immune-inflammatory disorder pathological process of VKH. The logistic regression diagnostic model constructed based on this biomarker combination was validated for system efficacy through 10-fold cross-validation and two-tailed DeLong test. It has extremely high sensitivity and specificity, and its diagnostic efficacy is significantly better than existing protein biomarker combined detection schemes. It can effectively make up for the shortcomings of traditional invasive diagnostic methods and protein biomarker detection, and provide a reliable and objective non-invasive detection method for the early and accurate diagnosis of VKH and the clinical differentiation of atypical cases. It has outstanding clinical application value and industrial transformation prospects.

[0055] Definitions:

[0056] Variable Importance in Projection (VIP): Primarily used in multivariate statistical models such as Partial Least Squares Regression (PLS), VIP quantifies the contribution of a single independent variable to the construction of the model's projection space and the explanation of the variance in the dependent variable.

[0057] False Discovery Rate (FDR): A statistical correction metric for multiple hypothesis testing scenarios, referring to the proportion of false positive results (incorrect rejection of the null hypothesis) among all test results judged as significantly positive.

[0058] P-value: The core statistic for hypothesis testing, which indicates the probability of obtaining the current observed data or more extreme data given that the null hypothesis is true.

[0059] MetaboAnalyst is a free, web-based metabolomics data analysis and annotation platform.

[0060] Lipid peaks: are observed in nuclear magnetic resonance spectroscopy (NMR spectroscopy). 1 In H-MRS, a set of characteristic signal peaks are formed by the characteristic hydrogen nuclei resonance of lipid molecules (triglycerides, free fatty acids, phospholipids, etc.). Attached Figure Description

[0061] Figure 1 These are OPLS-DA score plots of patient and healthy samples;

[0062] Figure 2 This is a score chart of 25 lipid metabolites selected by a machine learning algorithm;

[0063] Figure 3 This is a ROC curve of the diagnostic performance of a combination of biomarkers composed of 25 lipid metabolites.

[0064] Figure 4 This is a bar chart showing the differences in the expression of 25 substances between patient and healthy samples;

[0065] Figure 5 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0066] like Figures 1 to 5 As shown, a lipidomics detection method for the auxiliary diagnosis of VKH syndrome includes the following steps:

[0067] 1) Collect blood samples from patients with VKH syndrome and healthy controls;

[0068] 2) Preprocess the blood samples collected in step 1), including plasma separation and lipid extraction, to obtain lipid samples;

[0069] 3) Combining ultra-high performance liquid chromatography and triple quadrupole mass spectrometry, quantitative lipid analysis was performed on lipid samples from VKH syndrome patients and healthy controls, respectively, to obtain several lipid peaks. The lipid peaks were then used to construct a sample-to-whole lipid quantitative expression matrix.

[0070] 4) By performing unsupervised and supervised analysis on the sample-to-whole lipid quantitative expression matrix, several lipids with statistical differences, which can effectively distinguish the VKH syndrome patient group from the healthy control group and have biological associations were screened from the lipid samples as differentially expressed lipids.

[0071] 5) Based on the differentially expressed lipids obtained in step 4), an elastic network logistic regression diagnostic model was constructed, and several combinations of candidate diagnostic biomarkers for VKH syndrome were screened from the differentially expressed lipids using the elastic network logistic regression diagnostic model.

[0072] 6) By constructing a simple logistic regression model, the diagnostic efficiency and accuracy of the combination of candidate diagnostic biomarkers for VKH syndrome were verified, and the verified combination of candidate diagnostic biomarkers for VKH syndrome was used for early diagnosis of VKH.

[0073] The following is an example of implementing the above method:

[0074] 1) Blood samples were collected from 26 patients with VKH syndrome and 39 healthy controls using coagulation-promoting tubes. After tightening the tube caps, the tubes were gently inverted and mixed 3-5 times to ensure that the blood and coagulation-promoting agent were in full contact.

[0075] This study strictly followed the Declaration of Helsinki, recruiting 26 patients with newly diagnosed acute VKH syndrome and 39 healthy controls at the First Affiliated Hospital of Chongqing Medical University from February 2022 to December 2025. All participants were recruited voluntarily, and all signed written informed consent forms after fully understanding the study content, risks, and rights. Prior to blood collection, none of the patients had received systemic glucocorticoids, immunosuppressants, or biologics. Patient diagnosis strictly adhered to the revised VKH diagnostic criteria of the International Committee, combined with the clinical diagnostic criteria established by our research group. All patients presented with early ocular manifestations, including diffuse choroiditis and exudative retinal detachment, accompanied by neurological symptoms and signs. Exclusion criteria included hypertension, cardiovascular disease, diabetes, hepatitis, a history of tuberculosis, or other autoimmune diseases. There were no significant differences in age, sex, or body mass index (BMI) among the participants.

[0076] 2) Blood samples from 26 VKH syndrome patients and 39 healthy controls collected in step 1) using procoagulant tubes were preprocessed, including plasma separation and lipid extraction, to obtain lipid samples, as follows:

[0077] 2-1) Plasma separation:

[0078] 2-1-1) Let the coagulation tube containing the blood sample stand for 30 minutes at room temperature.

[0079] 2-1-2) Place the settled coagulation tube into a high-speed centrifuge, adjust the speed to 3000 rpm, and centrifuge for 10 minutes to fully separate the plasma from the blood cells.

[0080] 2-1-3) After centrifugation, slowly aspirate the clear plasma from the supernatant using a pipette and transfer it to a sterile 1.5mL centrifuge tube. Label the sample accordingly.

[0081] 2-1-4) Quickly place the centrifuge tubes containing plasma into an ultra-low temperature freezer at -80°C and avoid repeated freeze-thaw cycles of the samples throughout the process.

[0082] 2-2) Lipid Extraction

[0083] This embodiment uses the methyl tert-butyl ether (MTBE) method to extract lipids from plasma in centrifuge tubes. The specific process is as follows:

[0084] 2-2-1) Remove the centrifuge tube containing plasma from the -80℃ ultra-low temperature freezer, place it on ice to thaw slowly, and gently vortex to mix after thawing, avoiding violent shaking that could lead to lipid degradation.

[0085] 2-2-2) Use a pipette to draw 50 μL of thawed plasma and add it to a new 1.5 mL sterile centrifuge tube. Add 200 μL of methanol, place the centrifuge tube on a vortex mixer, and shake for 30 seconds to mix thoroughly and achieve plasma protein membrane rupture.

[0086] 2-2-3) Add 750 μL of MTBE solution containing internal standard to the centrifuge tube and vortex for 1 minute to fully dissolve the lipids in the MTBE organic phase.

[0087] 2-2-4) Add an appropriate amount of ultrapure water to the centrifuge tube, vortex for 30 seconds to mix, and let stand at room temperature for 5 minutes to induce the separation of the aqueous and organic phases. Then place the centrifuge tube in a centrifuge, adjust the speed to a suitable speed (usually 12000 rpm), and centrifuge for 5 minutes to completely separate the two phases.

[0088] 2-2-5) After centrifugation, slowly aspirate the upper transparent organic phase (the lipid phase) with a pipette and transfer it to a new sterile 1.5 mL centrifuge tube, avoiding aspirating the lower aqueous phase and the intermediate impurity layer.

[0089] 2-2-6) Place the centrifuge tube containing the organic phase into a vacuum concentrator, set appropriate conditions (temperature ≤30℃, vacuum degree according to instrument calibration), evaporate and dry until the organic phase is completely volatilized to obtain lipid powder.

[0090] 2-2-7) Add an appropriate amount of dichloromethane / methanol mixed solution (volume ratio 1:1, v / v) to the dried centrifuge tube, vortex for 1 minute to completely reconstitute the lipid powder. The reconstituted solution is the lipid extract, which is used for subsequent quantitative lipidomics analysis.

[0091] 3) Combining ultra-high performance liquid chromatography and triple quadrupole mass spectrometry, quantitative lipid analysis was performed on lipid samples from VKH syndrome patients and healthy controls, respectively, yielding several lipid peaks. These lipid peaks were then used to construct a sample-to-whole lipid quantitative expression matrix, as follows:

[0092] In this embodiment, an Agilent 1290 Infinity II ultra-high performance liquid chromatography system was selected, along with the SCIEXTripleQuad. TM A 7500+ triple quadrupole mass spectrometer was used in conjunction with a liquid chromatography-tandem mass spectrometry (LC-MS / MS) analysis platform for targeted lipid quantification analysis. The specific workflow is as follows:

[0093] 3-1) Instrument Inspection and Power-On

[0094] 3-1-1) Agilent 1290 Infinity II Ultra-High Performance Liquid Chromatography System: Turn on the main unit power, start the pump, injector and column oven in sequence, and wait for the system self-test to complete (about 5-10 minutes) to ensure that there are no errors in each module and that it is running normally.

[0095] 3-1-2) SCIEX TripleQuad TM 7500+ Triple Quadrupole Mass Spectrometer: Turn on the power to the mass spectrometer main unit and the workstation computer, launch the SCIEX OS software (version 3.3.1), establish a connection with the mass spectrometer, and wait for the mass spectrometer vacuum level to reach the detection requirements (vacuum level ≤ 1×10⁻⁶). -5 (Torr), preheat the ion source for 30 minutes to ensure the instrument is in a stable state.

[0096] 3-2) Instrument parameter settings

[0097] 3-2-1) Liquid Chromatography Parameter Settings (Two separation modes, operate separately):

[0098] 3-2-1-1) Mode 1: Reversed-phase separation, using a Kinetex C18 column

[0099] 3-2-1-1-1) Open Agilent OpenLab software, connect the UHPLC system, create a new method file, and install a Kinetex C18 column (2.1×100mm, particle size 2.6μm).

[0100] 3-2-1-1-2) Adjust the column temperature according to experimental requirements (normally 40℃), start the column oven to preheat, and equilibrate the column for 30 minutes.

[0101] 3-2-1-1-3) Injection volume 1-5μL (adjust according to sample lipid concentration), sample holder temperature 4℃, injection needle cleaned with methanol 3 times to avoid cross-contamination.

[0102] 3-2-1-1-4) Mobile phase setup:

[0103] Mobile phase A: A mixed solution of water / acetonitrile / methanol containing 7 mmol / L ammonium acetate, in a volume ratio of 1:1:1 (v / v / v).

[0104] Mobile phase B: Isopropanol solution containing 7 mmol / L ammonium acetate;

[0105] Flow rate: 0.3 mL / min (can be finely adjusted according to the separation of chromatographic peaks).

[0106] 3-2-1-1-5) Set the gradient according to the target lipid separation requirements to ensure good lipid peak separation and no overlap.

[0107] 3-2-1-1-6) Save the reverse phase separation method file, start the pump, run the gradient program, and balance the system until the baseline is stable (baseline fluctuation ≤ 0.01AU).

[0108] 3-2-1-2) Mode 2, hydrophilic interaction chromatography (HILIC) separation, using a Luna NH2 column.

[0109] 3-2-1-2-1) Replace the column: Remove the Kinetex C18 column and install a Luna NH2 column (2.0×100mm, particle size 3μm), ensuring that it is securely installed and leak-free.

[0110] 3-2-1-2-2) Equilibrate the column for 30 minutes, consistent with reversed-phase separation or adjusted according to experimental requirements.

[0111] (3-2-1-2-3) Consistent with reverse phase separation, maintain consistent injection volume, sample holder temperature, and injection needle cleaning procedures.

[0112] 3-2-1-2-4) Mobile phase setup:

[0113] Mobile phase A: A water / acetonitrile mixed solution containing 2 mmol / L ammonium acetate, volume ratio 1:1 (v / v);

[0114] Mobile phase B: a mixed solution of dichloromethane and acetonitrile containing 2 mmol / L ammonium acetate, at a volume ratio of 1:13 (v / v).

[0115] Flow rate: 0.25 mL / min (adapted to Luna NH2 column specifications, fine-tunable).

[0116] 3-2-1-2-5) Gradient elution program: Based on the chromatographic separation characteristics of hydrophilic interactions, a gradient is set in combination with the retention characteristics of the target lipids (the specific gradient is not given in the example; after optimization, the lipid peaks are ensured to be clearly separated).

[0117] 3-2-1-2-6) Save the HILIC separation method file, start the pump, run the gradient program, and balance the system until the baseline is stable.

[0118] 3-2-2) Mass spectrometer parameter settings (strictly follow the parameters given in the example)

[0119] 3-2-2-1) Open the SCIEX OS software (version 3.3.1), enter the "Method Editing" interface, and ensure that it is properly connected to the mass spectrometer.

[0120] 3-2-2-2) Scanning mode: Select the multiple reaction monitoring (MRM) mode for accurate quantitative detection of target lipids.

[0121] 3-2-2-3) Ion source parameters:

[0122] Ion spray voltage: ±3000V (applicable to both positive and negative ion modes);

[0123] Curtain air pressure: 35 psi;

[0124] Ion source gas 1 pressure: 40 psi;

[0125] Ion source gas 2 pressure: 70 psi;

[0126] Ion source temperature: 325℃ in positive mode and 550℃ in negative mode (adjust according to the detection mode).

[0127] 3-2-2-4) MRM channel settings: Based on the precursor ions and characteristic fragment ions of the target lipid, establish corresponding MRM monitoring channels, optimize parameters such as collision energy and declustering voltage, and ensure stable and highly specific lipid response values.

[0128] 3-2-2-5) Linkage settings: Link the liquid chromatography method and the mass spectrometry method for reversed phase separation and HILIC separation respectively to ensure that liquid chromatography elution and mass spectrometry acquisition are synchronized and to save the complete instrument method for both separation modes.

[0129] 3-3) Lipid analysis

[0130] 3-3-1) Open the SCIEX OS software (version 3.3.1) "Sequence Editing" interface, create a new sequence, arrange it in the order of "Blank Sample → Sample to be Analyzed → QC Sample" (QC sample can be prepared by mixing all samples to be analyzed), set the run time for each sample (determined according to the gradient elution program), and check "Automatically save raw data".

[0131] 3-3-2) Sample Analysis:

[0132] 3-3-2-1) First, use reversed-phase separation mode (Kinetex C18 column) to start the sequence and analyze all lipid samples. Monitor the total ion current (TIC) and MRM chromatogram in real time to ensure symmetrical peak shape and no interference from extraneous peaks.

[0133] 3-3-2-2) After the reversed-phase separation analysis is completed, switch to HILIC separation mode (Luna NH2 column), equilibrate the system until the baseline is stable, start the sequence, complete the secondary analysis of all samples, and save the original data of both modes.

[0134] In this embodiment, if problems such as abnormal peak shape, baseline drift, or abnormal response value occur during the analysis, the sequence must be stopped immediately, the status of the chromatographic column, mobile phase, and ion source must be checked, and the analysis must be repeated after troubleshooting.

[0135] 3-3-2-3) After all sample analyses are completed, turn off the mobile phase channel of the pump, rinse both columns with the corresponding organic solvents for 30 minutes (rinse the reversed-phase column with methanol and the HILIC column with acetonitrile), then turn off the power to the liquid chromatography and mass spectrometry instruments in sequence, and tidy up the lab bench.

[0136] 3-3-3) Data Processing

[0137] 3-3-3-1) Start the SCIEX OS software (version 3.3.1) and BIOTREE BioBud software (version 2.0.4) and import the raw data of the two separation modes acquired by liquid chromatography-tandem mass spectrometry into the software.

[0138] 3-3-3-2) Data Preprocessing:

[0139] Peak identification and integration: The software automatically identifies and integrates lipid peaks, and manually corrects abnormal peaks (such as baseline shifts and peak splits) to ensure accurate integration.

[0140] Lipid peak screening: Remove lipid peaks with a missing rate of more than 50% in any group of samples, and retain lipid peaks with good stability.

[0141] Missing value imputation: For missing values ​​in the remaining lipid peaks, imputation is performed using half the minimum value of the corresponding lipid peak to ensure data integrity.

[0142] 3-3-3-3) The preprocessed data were quantitatively calculated using BIOTREE BioBud software (version 2.0.4), and the concentration values ​​of each lipid in each sample were obtained by combining internal standard (if applicable) correction.

[0143] 3-3-3-4) Finally, 1186 lipid peaks were retained for subsequent data analysis (such as differential lipid screening, statistical analysis, etc.). The quantitative results were exported to an Excel spreadsheet for easy processing.

[0144] 3-3-3-5) Using lipid samples as rows and lipid peaks as columns, and configuring corresponding VKH syndrome patient or healthy control labels, a sample-to-total lipid quantitative expression matrix is ​​constructed.

[0145] 4) Using unsupervised and supervised analysis of the sample-to-whole lipid quantitative expression matrix, several lipids with statistically significant differences, capable of effectively distinguishing the VKH syndrome patient group from the healthy control group, and biologically associated were screened from the lipid samples as differentially expressed lipids. The specific method is as follows:

[0146] In this embodiment, SIMCA version 18.0.1 software is used for principal component analysis (PCA) and orthogonal partial least squares discriminant analysis.

[0147] Unsupervised principal component analysis (PCA): Eliminates redundancy in high-dimensional data through unsupervised dimensionality reduction, visualizes the overall distribution characteristics of samples, preliminarily assesses the inherent differences in lipid profiles between groups and the clustering degree of samples within groups, and identifies anomalous outliers.

[0148] Supervised orthogonal partial least squares discriminant analysis (OPLS-DA): Eliminates systemic noise irrelevant to grouping, maximizes the highlighting of lipid profile differences between groups, provides a reliable model basis for differential lipid screening, and evaluates the separation effect between groups.

[0149] 4-1) Unsupervised principal component analysis (PCA) was used.

[0150] 4-1-1) Import the preprocessed sample-to-lipid quantitative expression matrix into SIMCA 18.0.1 software, and label the samples according to VKH syndrome patient group and healthy control group to ensure that the grouping information corresponds one-to-one with the sample.

[0151] 4-1-2) Perform unit variance scaling (UV Scaling) on ​​the input matrix to eliminate the interference of the magnitude difference of different lipid variables on the dimensionality reduction results and ensure that the contribution weight of each variable to the model is balanced.

[0152] 4-1-3) Based on the standardized dataset, a PCA model is constructed, the core principal components are extracted, and the PCA scatter plot is generated to realize the two-dimensional / three-dimensional visualization of high-dimensional lipidomics data.

[0153] 4-1-4) Determine whether the scatter plot obtained from PCA meets the following conditions. If it does, proceed to step 4-2). If it does not, the process terminates and backtracks for rectification:

[0154] ① In the PCA scatter plot, the VKH syndrome patient group and the healthy control group showed a clear inter-group separation trend in the principal component space, proving that there was an overall statistical difference in the lipid profiles of the two groups.

[0155] ② In the PCA scatter plot, the samples within both the VKH syndrome patient group and the healthy control group showed good clustering with no extreme outliers, proving that the sample inclusion, data preprocessing and detection data quality were up to standard.

[0156] In this embodiment, if the two groups in the PCA scatter plot are completely mixed or there are a large number of outliers, the subsequent OPLS-DA between groups is likely to be overfitting, and the analysis results are completely unreliable.

[0157] Depend on Figure 1 It was found that the samples from the VKH syndrome patient group and the healthy control group showed a significant separation trend in the principal component space, with high aggregation of samples within each group and significant differences in lipid profiles. The OPLS-DA model's 1000-permutation test results showed that the model's true R... 2 With Q 2 The value is significantly higher than the values ​​after all random permutations, indicating no overfitting.

[0158] 4-2) Supervised orthogonal partial least squares discriminant analysis (OPLS-DA) was used to evaluate the separation effect between groups and screen differentially expressed lipid metabolites:

[0159] 4-2-1) Use a standardized lipid quantification matrix and grouping labels that are completely consistent with PCA analysis to ensure the consistency of the analytical process and the comparability of results.

[0160] 4-2-2) In SIMCA 18.0.1 software, an OPLS-DA model is constructed based on the grouping information, and UV standardization preprocessing is performed simultaneously to match the preprocessing rules of PCA analysis.

[0161] 4-2-3) Using the 7-fold cross-validation method, calculate the core fitting parameters of the OPLS-DA model:

[0162] R²Y: The model's explanatory power for the sample grouping variables, reflecting the model's good fit;

[0163] Q²Y: The predictive power of the model, reflecting its generalization performance.

[0164] 4-2-4) Perform 1000 permutation tests, randomly shuffle the sample group labels, and repeatedly fit the OPLS-DA model. Compare the R-squared of the true model and the random permutation model. 2 Y, Q 2 Y value.

[0165] If the R of the real model 2 Y, Q 2 The Y value is significantly higher than the corresponding values ​​of all random permutation models, and the intercept of the Q²Y regression line is <0.05, indicating that the model has no risk of overfitting and its robustness meets the analysis requirements.

[0166] 4-3) Screening for lipid metabolites with significant differences

[0167] 4-3-1) Based on the validated robust OPLS-DA model, calculate the variable projection importance (VIP) score for each lipid variable to quantify the contribution of a single variable to the differences between groups.

[0168] 4-3-2) Perform two-tailed statistical tests between groups on all lipid variables, and use the false discovery rate (FDR) method to perform multiple tests to correct the original P values, strictly controlling the false positive rate.

[0169] 4-3-3) Lipid molecules that simultaneously meet the following two conditions are defined as differentially expressed lipids with significant intergroup differences:

[0170] Condition 1: VIP > 1;

[0171] Condition 2: P < 0.05 after FDR correction.

[0172] In this embodiment, both VIP (variable contribution) and corrected P-value (statistical significance) thresholds must be met simultaneously. Single-indicator screening can lead to false positives / false negatives, and only lipids that meet both thresholds can be included in the final difference list.

[0173] 4-3-4) Based on the above thresholds, differential lipid screening was completed, and the standardized name, quantitative difference information, VIP value, corrected P value and other core parameters of the differential lipids were output. Finally, a total of 262 differentially significant lipids were identified.

[0174] 4-4) Metabolic pathway enrichment analysis of differentially expressed lipids

[0175] 4-4-1) Data import: Import the standardized annotation IDs (such as HMDB ID, Lipid Maps ID) and intergroup quantitative difference information of the 262 differentially significant lipids obtained from the screening into the MetaboAnalyst online analysis platform.

[0176] 4-4-2) Parameter settings: Select human species database, use hypergeometric test for pathway enrichment statistical analysis, and combine topological analysis weights to evaluate the core importance of pathways.

[0177] 4-4-3) Result determination: Pathways with P < 0.05 after FDR correction were identified as significantly enriched pathways, clarifying the core biological pathways related to VKH syndrome associated with differential lipids.

[0178] 5) Based on the differentially expressed lipids obtained in step 4), an elastic network logistic regression diagnostic model was constructed. This model was then used to screen several candidate diagnostic biomarker combinations for VKH syndrome from the differentially expressed lipids. The specific process is as follows:

[0179] 5-1) Dataset and Model Framework Preparation

[0180] 5-1-1) Runtime environment configuration: Based on the Python 3.x programming language, the elastic network logistic regression diagnostic model is built by calling the scikit-learn 3.14.2 algorithm library, and the fixed random seed ensures that the results are repeatable.

[0181] 5-1-2) Dataset preprocessing: Standardized preprocessing was performed on differentially expressed lipids. Stratified sampling was used to ensure a balanced distribution of samples between the VKH patient group and the healthy control group. The dataset was partitioned based on a 10-fold cross-validation framework to avoid data leakage.

[0182] 5-2) Construction and Hyperparameter Optimization of Elastic Network Logistic Regression Model

[0183] 5-2-1) Classifier selection: The logistic regression model with Elastic Net regularization constraint is used as the diagnostic classifier, and L1 (Lasso) and L2 (Ridge) penalty terms are combined simultaneously to achieve feature selection and model complexity control.

[0184] 5-2-2) Hyperparameter Grid Optimization: 10-fold cross-validation is used to perform global grid optimization of the model's hyperparameters. Optimized parameters include:

[0185] Regularization strength λ: controls the overall penalty intensity of the model and reduces the risk of overfitting;

[0186] L1 / L2 mixing parameter α: value range [0,1], α=1 is pure L1 regularization, α=0 is pure L2 regularization.

[0187] In this embodiment, the optimal model selection priority is as follows:

[0188] Prioritize models with the highest area under the receiver operating characteristic (AUC) curve in 10-fold cross-validation;

[0189] When the cross-validation AUC of different models is similar, the model with higher regularization strength λ should be selected first to reduce model complexity;

[0190] When the λ value is the same, the model with a higher L1 ratio α is preferred to balance the model's predictive performance and simplicity.

[0191] 5-3) Using ROC curve area under analysis, two-tailed DeLong test, and permutation test techniques, the diagnostic efficacy and overfitting risk of the elastic network regularized logistic regression model are verified. If the model meets the preset diagnostic efficacy and robustness requirements, it is solidified as the optimal elastic network regularized logistic regression model. If the requirements are not met, the model is iteratively optimized through hyperparameter reconstruction.

[0192] In this embodiment, the specific expression of the optimal elastic network regularized logistic regression model is as follows:

[0193]

[0194] In the formula, n is the total sample size (i.e., the number of differentially expressed lipids). To predict probabilities for the model, λ is the feature regression coefficient vector, λ>0 is the regularization intensity coefficient, and ρ∈[0,1] is the mixed proportion coefficient of L1 and L2 regularization terms.

[0195] In this embodiment, the optimal model was determined as follows: after optimization using a 10-fold cross-validation grid, the optimal hyperparameter combination was determined to be α=0.3 and λ=31.6228. Based on this, the final elastic grid logistic regression diagnostic model was constructed. The AUC value of this model reached 0.972, P<0.001, which shows high diagnostic sensitivity and specificity.

[0196] 5-4) Screening of candidate diagnostic biomarker combinations

[0197] 5-4-1) Feature selection rules: such as Figure 2 As shown, the regression coefficient matrix of the optimal elastic network regularized logistic regression model was extracted, and the regression coefficients of each differential lipid were verified one by one. Several differential lipids with non-zero regression coefficients were selected as candidate diagnostic biomarkers for VKH syndrome. The biological nature of the candidate diagnostic biomarkers for VKH syndrome is small glycerophospholipids.

[0198] In this embodiment, lipid variables with non-zero regression coefficients were identified as core features with independent diagnostic contributions to VKH syndrome. Following the above screening, 25 candidate diagnostic biomarker combinations for VKH syndrome (i.e., lipid metabolite combinations) were finally obtained, specifically including:

[0199] Cardiolipin CL (80:14) - Fatty acid FA (16:0), Phosphatidylserine PS (18:1-22:0), Phosphatidylserine PS (20:1-20:4), Phosphatidylcholine PC (17:0-18:0), Phosphatidylcholine PC (20:2-20:3), Phosphatidylethanolamine PE (22:4-22:4), Phosphatidylcholine PC (20:0-18:1), Phosphatidylcholine PC (18:2-22:4), Phosphatidylcholine PC (18:3-20:5), Cardiolipin CL (80:15) - Fatty acid FA (16:0), Phosphatidylcholine PC (18:1-20:2), Phosphatidylinositol PI (18:1-20:4), Phosphatidylcholine PC (18:3_18:3), phosphatidylcholine PC (20:2_20:2), phosphatidylcholine PC (18:2_20:3), phosphatidylethanolamine PE (18:2_20:0), phosphatidylcholine PC (20:3_20:5), phosphatidylserine PS (18:1_20:4), phosphatidylcholine PC (20:3_20:3), phosphatidylserine PS (20:0_20:4), phosphatidylserine PS (18:0_18:0), phosphatidylinositol PI (20:0_20:4), bisphosphatidylglycerol BMP (16:0_20:0), phosphatidylcholine PC (14:0_16:0), phosphatidylcholine PC (16:0_14:0).

[0200] like Figure 4 As shown, the expression of 25 lipid molecules screened based on the optimal model differed significantly between the VKH group and the healthy control group, suggesting that they can serve as an important indicator for distinguishing patients from normal individuals.

[0201] It is worth noting that in this embodiment, any one of the 25 lipid molecules exhibits weak distinguishing efficacy between VKH syndrome patients and healthy individuals, with an area under the curve (AUC) ranging from 0.756 to 0.900 and a mean of only 0.843 ± 0.035. This indicates insufficient sensitivity and specificity, making them prone to misdiagnosis and missed diagnosis when used alone, and thus unsuitable as a standalone diagnostic biomarker. However, by constructing a multi-molecule combined biomarker panel from the 25 selected lipid metabolites, the various lipid components can complement each other in terms of diagnostic features, comprehensively integrating the overall abnormal expression characteristics of the body's lipid profile. Compared to a single biomarker, this combined panel significantly improves the distinguishing ability between the case group and the healthy control group, significantly increasing the diagnostic AUC to 0.953, demonstrating good clinical auxiliary diagnostic application value.

[0202] 6) By constructing a simple logistic regression model, the diagnostic efficiency and accuracy of candidate diagnostic biomarker combinations for VKH syndrome were validated. The validated candidate diagnostic biomarker combinations for VKH syndrome were then used for early diagnosis of VKH, as detailed below:

[0203] 6-1) Sum normalization: For each enrolled sample's raw lipidomics data, calculate the sum of the relative contents of all lipids in the sample, convert the relative contents of a single glycerophospholipid into "the original relative content of the lipid / the total lipid content of the corresponding sample", complete the sum normalization within the sample, and eliminate the systematic bias caused by fluctuations in sample loading amount and mass spectrometry detection efficiency.

[0204] 6-2) Z-score standardization: For the 25 glycerophospholipid variables that have been summed and normalized, Z-score standardization was performed on all enrolled samples;

[0205] 6-3) In this embodiment, based on the aforementioned optimal elastic network regularized logistic regression model, the 25 candidate lipid biomarkers for VKH syndrome obtained by screening are used as inputs and substituted into the above binary logistic regression framework to obtain the following linear prediction equation for the auxiliary diagnosis of VKH syndrome.

[0206] The specific formula for binary logistic regression probability prediction is as follows:

[0207] ;

[0208] Where P is the positive predictive probability of the disease, Z is the linear combination term, the independent variable X is the relative lipid content after summation normalization and Z-score standardization, and e is the natural constant.

[0209] 6-4) Use 10-fold cross-validation to plot receiver operating characteristic (ROC) curves, calculate the area under the curve (AUC) value, and simultaneously output the model's core performance parameters such as diagnostic sensitivity, specificity, accuracy, positive predictive value, and negative predictive value.

[0210] 6-5) The two-tailed DeLong test was used to perform statistical difference analysis on the AUC values ​​of the ROC curve to verify whether the AUC value was significantly higher than the random classification level (AUC=0.5). P<0.05 was used as the threshold for determining that the difference was statistically significant.

[0211] like Figure 3 As shown, the logistic regression model constructed above underwent 10-fold cross-validation, with an average AUC of 0.953±0.095 and a DeLong test P<0.001. This model, combined with the corresponding 25 glycerophospholipid biomarkers, has a very high ability to distinguish between VKH syndrome patients and healthy controls, with good diagnostic sensitivity and specificity, and has the potential for clinical auxiliary diagnosis.

[0212] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications made to the present invention by those skilled in the art without departing from the spirit of the present invention shall fall within the protection scope of the present invention.

Claims

1. A combination of lipid metabolites for early diagnostic biomarkers of VKH, characterized in that, This group of lipid metabolites includes: Cardiolipin CL (80:14) - Fatty acid FA (16:0), Phosphatidylserine PS (18:1-22:0), Phosphatidylserine PS (20:1-20:4), Phosphatidylcholine PC (17:0-18:0), Phosphatidylcholine PC (20:2-20:3), Phosphatidylethanolamine PE (22:4-22:4), Phosphatidylcholine PC (20:0-18:1), Phosphatidylcholine PC (18:2-22:4), Phosphatidylcholine PC (18:3-20:5), Cardiolipin CL (80:15) - Fatty acid FA (16:0), Phosphatidylcholine PC (18:1-20:2), Phosphatidylinositol PI (18:1-20:4), Phosphatidylcholine PC (18:3_18:3), phosphatidylcholine PC (20:2_20:2), phosphatidylcholine PC (18:2_20:3), phosphatidylethanolamine PE (18:2_20:0), phosphatidylcholine PC (20:3_20:5), phosphatidylserine PS (18:1_20:4), phosphatidylcholine PC (20:3_20:3), phosphatidylserine PS (20:0_20:4), phosphatidylserine PS (18:0_18:0), phosphatidylinositol PI (20:0_20:4), bisphosphatidylglycerol BMP (16:0_20:0), phosphatidylcholine PC (14:0_16:0), phosphatidylcholine PC (16:0_14:0).

2. A combination of lipid metabolites for early diagnostic biomarkers of VKH, characterized in that, The lipid metabolites together constitute a plasma lipid biomarker, which is used for the early diagnosis of VKH.

3. A combination of lipid metabolites for early diagnostic biomarkers of VKH, characterized in that, The application of the plasma lipid biomarkers in the preparation of VKH early non-invasive screening or diagnostic kits.

4. A kit for screening or diagnosing VKH, characterized in that, The kit contains reagents for detecting the content of the lipid metabolite combination as described in claim 1, and the kit is capable of distinguishing between patients with VKH syndrome and healthy individuals.

5. A lipidomics detection method for the auxiliary diagnosis of VKH syndrome, characterized in that, Includes the following steps: 1) Collect blood samples from patients with VKH syndrome and healthy controls; 2) Preprocess the blood samples collected in step 1), including plasma separation and lipid extraction, to obtain lipid samples; 3) Combining ultra-high performance liquid chromatography and triple quadrupole mass spectrometry, quantitative lipid analysis was performed on lipid samples from VKH syndrome patients and healthy controls, respectively, to obtain several lipid peaks. The lipid peaks were then used to construct a sample-to-whole lipid quantitative expression matrix. 4) Unsupervised and supervised analysis were performed on the sample-to-whole lipid quantitative expression matrix to screen several lipids that could effectively distinguish between the VKH syndrome patient group and the healthy control group as differential lipids; 5) Based on the differentially expressed lipids obtained in step 4), an elastic network logistic regression diagnostic model was constructed, and several combinations of candidate diagnostic biomarkers for VKH syndrome were screened from the differentially expressed lipids using the elastic network logistic regression diagnostic model. 6) By constructing a simple logistic regression model, the diagnostic efficiency and accuracy of the combination of candidate diagnostic biomarkers for VKH syndrome were verified, and the verified combination of candidate diagnostic biomarkers for VKH syndrome was used for early diagnosis of VKH.

6. The lipidomics detection method according to claim 5, characterized in that, In step 3), the sample-to-whole lipid quantitative expression matrix is ​​constructed as follows: 3-1) Two separation modes are set up: reversed-phase separation and hydrophilic interaction chromatography. The two separation schemes use corresponding dedicated chromatographic columns and mobile phase systems, which can cover lipids with different polarity ranges. 3-2) The lipid samples were separated by ultra-high performance liquid chromatography using reversed-phase separation and hydrophilic interaction chromatography, respectively, to obtain two types of lipid elution fractions separated according to retention time; 3-3) The lipid elution fraction obtained in step 3-2) is introduced into a triple quadrupole mass spectrometer, matching parameters are set, and various lipid characteristic peak signals are collected in multiple reaction monitoring quantification mode; 3-4) Preprocess the collected lipid peak signals, remove lipid peaks with a missing rate of more than 50% in any group, and fill the remaining missing values ​​with 1 / 2 of the minimum value of the corresponding peak to obtain several lipid peaks that can be used for subsequent analysis. 3-5) Construct a sample-to-whole lipid quantitative expression matrix by using lipid samples as rows and lipid peaks as columns, and configuring corresponding VKH syndrome patient or healthy control labels.

7. The lipidomics detection method according to claim 5, characterized in that, In step 4), the differentially expressed lipids are obtained by screening in the following manner: 4-1) Perform unsupervised principal component analysis (PCA) on the total lipid expression matrix of the VKH syndrome patient group and the healthy control group to generate PCA scatter plots. Determine whether the PCA scatter plots meet the following conditions. If they do, proceed to step 4-2). If they do not, the process is terminated and backtracked for rectification: ① In the PCA scatter plot, the VKH syndrome patient group and the healthy control group showed a clear inter-group separation trend in the principal component space, proving that there was an overall statistical difference in the lipid profiles of the two groups. ② In the PCA scatter plot, the samples within both the VKH syndrome patient group and the healthy control group showed good clustering with no extreme outliers, proving that the sample inclusion, data preprocessing and detection data quality were up to standard. 4-2) Construct an OPLS-DA model for screening differential lipids, and use 7-fold cross-validation to calculate the R² and Q² values ​​of the OPLS-DA model to evaluate the fitting effect of the OPLS-DA model. Verify the robustness of the OPLS-DA model through several permutation tests to eliminate the risk of overfitting. 4-3) Using the validated OPLS-DA model, calculate the variable projection importance (VIP) and p-value for each lipid in the sample-to-whole lipid quantitative expression matrix. Perform FDR correction on the p-value, and identify the lipids corresponding to the sample-to-whole lipid quantitative expression matrix as differentially expressed lipids when the VIP and p-values ​​meet the following conditions: ① Variable projection importance (VIP) > 1; ② The P-value after FDR correction is <0.05; 4-4) Perform metabolic pathway enrichment analysis on the differentially identified lipids in step 4-3) to identify key biological pathways associated with differentially identified metabolites and lipids, elucidate pathological mechanisms, and discover potential therapeutic targets.

8. The lipidomics detection method according to claim 5, characterized in that, In step 5), the combination of candidate diagnostic biomarkers for VKH syndrome was obtained by screening in the following manner: 5-1) Using binary logistic regression as the underlying framework of the model and elastic network regularization as the core optimization and feature selection kernel of the model, construct an elastic network regularized logistic regression model; 5-2) Using the differentially expressed lipids obtained in step 4) as the training set, and combining 10-fold cross-validation, the elastic network regularized logistic regression model is trained to determine the optimal hyperparameters of the elastic network regularized logistic regression model, including the regularization strength and L1 / L2 mixed parameters. 5-3) The diagnostic efficacy and overfitting risk of the elastic network regularized logistic regression model are verified by using ROC curve area analysis, two-tailed DeLong test and permutation test. If the model meets the preset diagnostic efficacy and robustness requirements, it is solidified as the optimal elastic network regularized logistic regression model. If the requirements are not met, the model is iteratively optimized by hyperparameter reconstruction. 5-4) Extract the regression coefficient matrix of the optimal elastic network regularized logistic regression model, verify the regression coefficients of each differential lipid one by one, and screen out several differential lipids with non-zero regression coefficients as a combination of candidate diagnostic biomarkers for VKH syndrome.

9. The lipidomics detection method according to claim 5, characterized in that, In step 5-2), the optimization principle of the hyperparameters is as follows: If the cross-validation AUC values ​​of multiple hyperparameters are similar, prioritize choosing a higher regularization strength λ to reduce model complexity; If the regularization strength λ is the same, a higher L1 ratio should be preferred to balance prediction performance and model simplicity.

10. The lipidomics detection method according to claim 5, characterized in that, In step 6), the diagnostic efficiency and accuracy of the combination of candidate diagnostic biomarkers for VKH syndrome are verified, as follows: 6-1) Based on the combinations of several candidate diagnostic biomarkers for VKH syndrome selected in step 5), a logistic regression model is constructed; 6-2) Using the ROC curve and the corresponding area under the curve (AUC) as evaluation indicators, the diagnostic efficacy of the combination of candidate diagnostic biomarkers for VKH syndrome was evaluated through 10-fold cross-validation. 6-3) The two-tailed DeLong test was used to calculate the P-value of the AUC of the logistic regression model. When P < 0.05, the logistic regression model was considered to be statistically significant, which was used to verify the accuracy of the logistic regression model in distinguishing VKH patients from healthy controls.

Images