Biomarker for detecting stable coronary artery disease and application thereof

Abstract

The invention discloses a biomarker for detecting stable coronary artery disease (SCAD) and application thereof, and particularly aims at patients who have serious coronary artery stenosis but have not suffered from major adverse cardiovascular events (MACE). Wherein the biomarkers comprise a CDO homologous protein (BOC), a galectin-9 (LGALS9), an adrenomedullin (ADM), an angiotensin converting enzyme 2 (ACE2) and a brain natriuretic peptide (NPPB). A detection model with high sensitivity and high specificity is constructed by analyzing clinical risk factors of a to-be-detected patient and the relative expression level of the biomarker in a plasma sample of the to-be-detected patient. According to the detection model, the high-risk stable coronary artery disease patient can be accurately identified in a non-invasive manner before MACE occurs, a clinician is helped to make a personalized treatment scheme more effectively, and therefore the occurrence risk of MACE is reduced.

Description

Technical Field

[0001] The present invention relates to the field of cardiovascular disease detection, and in particular to biomarkers for detecting stable coronary artery disease and applications thereof. Background Art

[0002] Coronary artery disease (CAD) is one of the leading causes of morbidity and mortality worldwide, encompassing both acute coronary syndrome (ACS) and stable coronary artery disease (SCAD). Although ACS has attracted a great deal of clinical attention due to its acute and life-threatening presentation, SCAD is often overlooked due to its relatively asymptomatic nature and the complexity of assessing its progression. However, high-risk SCAD patients with severe coronary artery stenosis remain at high risk for major adverse cardiovascular events (MACE), such as myocardial infarction and heart failure, despite following guideline-recommended treatment regimens. Inadequate identification and management of this patient population often leads to disease progression and irreversible damage, and more effective methods are urgently needed to identify and intervene in these high-risk patients in advance.

[0003] Existing diagnostic tools for coronary artery stenosis, such as coronary CT angiography (CTA), invasive coronary angiography, and intravascular imaging techniques (such as IVUS and OCT), play a vital role in guiding percutaneous coronary intervention (PCI), but these methods have inherent disadvantages such as patient discomfort, high operational risks, and high costs, which limit their widespread application in early screening. In addition, non-invasive assessments based on clinical symptoms and traditional risk factors are often inaccurate, leading to delayed identification of high-risk SCAD cases. Studies have shown that approximately 20% of SCAD cases may be misdiagnosed at the time of initial diagnosis. These limitations indicate that relying solely on existing technologies cannot fully and accurately screen high-risk patients, especially when there are no obvious symptoms, which often delays the timing of intervention.

[0004] In terms of biomarkers, cardiac troponin (cTn), as a mature marker of myocardial injury, has limited effectiveness in identifying early high-risk SCAD patients, and cTn levels will only increase after myocardial injury reaches a significant degree. Other markers such as NT-proBNP and C-reactive protein (CRP) have also been explored for risk stratification, but they are insufficient in specificity and it is difficult to effectively distinguish early high-risk patients. Therefore, existing biomarkers and diagnostic tools have limitations in the early identification of high-risk SCAD patients, and existing technologies still need to be improved and developed. Summary of the invention

[0005] In view of the above-mentioned deficiencies of the prior art, the purpose of the present invention is to provide biomarkers for detecting stable coronary artery disease and their applications, aiming to solve the problem that the existing biomarkers for stable coronary artery disease have low specificity and lack of suitable detection models.

[0006] The technical solution of the present invention is as follows:

[0007] In a first aspect, a group of biomarkers for detecting stable coronary artery disease is provided, wherein the biomarkers include: CDO homologous protein, galectin-9, adrenomedullin, angiotensin converting enzyme 2 and brain natriuretic peptide.

[0008] Optionally, the biomarkers further include: serine protease 8 and V-SET immunoglobulin domain-containing protein 2.

[0009] In a second aspect, there is provided a use of the biomarker as described in the first aspect in the preparation of a stable coronary artery disease detection product.

[0010] In a third aspect, a reagent for detecting individual information is provided for use in the preparation of a stable coronary artery disease detection product, wherein the individual information includes: clinical risk factors of the patient to be tested and the relative expression levels of biomarkers in plasma samples; wherein the biomarkers include: CDO homologous protein, galectin-9, adrenomedullin, angiotensin-converting enzyme 2 and brain natriuretic peptide.

[0011] In a fourth aspect, a reagent for detecting individual information is provided for use in the preparation of a stable coronary artery disease detection product, wherein the individual information includes: the relative expression level of biomarkers in the plasma sample of the patient to be tested; wherein the biomarkers include: CDO homologous protein, galectin-9, adrenomedullin, angiotensin converting enzyme 2, brain natriuretic peptide, serine protease 8 and V-SET immunoglobulin domain-containing protein 2.

[0012] Optionally, the stable coronary artery disease detection product includes a detection reagent, a detection kit, a detection chip and a detection model.

[0013] In the fifth aspect, a detection model for identifying patients with stable coronary artery disease is provided, wherein the detection model is characterized by the clinical risk factors of the patients to be tested and the relative expression levels of biomarkers in plasma samples, and is constructed using a machine learning algorithm; wherein the biomarkers include: CDO homologous protein, galectin-9, adrenomedullin, angiotensin-converting enzyme 2 and brain natriuretic peptide.

[0014] In the sixth aspect, a detection model for identifying patients with stable coronary artery disease is provided, wherein the detection model is characterized by biomarkers in the plasma samples of the patients to be tested and is constructed using a machine learning algorithm; wherein the biomarkers include: CDO homologous protein, galectin-9, adrenomedullin, angiotensin converting enzyme 2, brain natriuretic peptide, serine protease 8 and V-SET immunoglobulin domain-containing protein 2.

[0015] Optionally, the machine learning algorithm is logistic regression and support vector machine.

[0016] Optionally, the detection model uses a nomogram to predict the probability of stable coronary artery disease in the patient to be tested.

[0017] Beneficial effects: Compared with the prior art, the advantage of the present invention is that it provides a set of new biomarkers for detecting stable coronary artery disease (SCAD), especially for patients with severe coronary artery stenosis but who have not yet experienced major adverse cardiovascular events (MACE). The present invention constructs a detection model with high sensitivity and high specificity by analyzing the clinical risk factors of the patients to be tested and the relative expression levels of the biomarkers as described above in their plasma samples. The detection model can accurately identify high-risk stable coronary artery disease patients non-invasively before MACE occurs, helping clinicians to more effectively formulate personalized treatment plans, thereby reducing the risk of MACE. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a principal component analysis diagram of differential protein analysis performed by OLINK proteomics in Example 1.

[0019] Figure 2 It is a volcano plot of differential protein analysis by OLINK proteomics in Example 1.

[0020] Figure 3 It is a heat map of differential protein analysis by OLINK proteomics in Example 1.

[0021] Figure 4 This is a PPI network analysis diagram of differential protein analysis performed by OLINK proteomics in Example 1.

[0022] Figure 5 This is a GO enrichment analysis diagram of differential protein analysis performed by OLINK proteomics in Example 1.

[0023] Figure 6Flow chart of the potential biomarker identification of high-risk SCAD combining proteomics and machine learning methods in Example 1.

[0024] Figure 7 This is a diagram showing the construction and evaluation results of the protein model, clinical model and comprehensive model in Example 1.

[0025] Figure 8 This is a graph showing the prediction of SCAD incidence and model performance evaluation using the comprehensive model in Example 1. DETAILED DESCRIPTION

[0026] The present invention provides biomarkers for detecting stable coronary artery disease and applications thereof. To make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail below.

[0027] The embodiment of the present invention provides a group of biomarkers for detecting stable coronary artery disease, wherein the biomarkers include: CDO homologous protein, galectin-9, adrenomedullin, angiotensin converting enzyme 2 and brain natriuretic peptide.

[0028] In one embodiment, the stable coronary artery disease is high-risk stable coronary artery disease; the high-risk stable coronary artery disease refers to a coronary artery diameter reduction of 70% or more as shown by coronary angiography.

[0029] In one embodiment, the biomarkers further include: serine protease 8 and V-SET immunoglobulin domain-containing protein 2.

[0030] The embodiments of the present invention provide the use of the above-mentioned biomarkers in the preparation of a stable coronary artery disease detection product.

[0031] An embodiment of the present invention provides the use of a reagent for detecting individual information in the preparation of a stable coronary artery disease detection product, wherein the individual information includes: clinical risk factors of the patient to be tested and the relative expression levels of biomarkers in the plasma sample; wherein the biomarkers include: CDO homologous protein, galectin-9, adrenomedullin, angiotensin converting enzyme 2 and brain natriuretic peptide.

[0032] In one embodiment, the clinical risk factor is gender. Specifically, studies have shown that stable coronary artery disease has gender differences, and being male is one of the clinical risk factors for stable coronary artery disease.

[0033] An embodiment of the present invention provides the use of a reagent for detecting individual information in the preparation of a stable coronary artery disease detection product, wherein the individual information includes: the relative expression level of biomarkers in a plasma sample of a patient to be tested; wherein the biomarkers include: CDO homologous protein, galectin-9, adrenomedullin, angiotensin converting enzyme 2, brain natriuretic peptide, serine protease 8 and V-SET immunoglobulin domain-containing protein 2.

[0034] In one embodiment, the stable coronary artery disease detection product includes a detection reagent, a detection kit, a detection chip and a detection model.

[0035] An embodiment of the present invention provides a detection model for identifying patients with stable coronary artery disease. The detection model is characterized by the clinical risk factors of the patients to be tested and the relative expression levels of biomarkers in plasma samples, and is constructed using a machine learning algorithm; wherein the biomarkers include: CDO homologous protein, galectin-9, adrenomedullin, angiotensin-converting enzyme 2 and brain natriuretic peptide.

[0036] In one embodiment, the clinical risk factor is gender.

[0037] An embodiment of the present invention provides a detection model for identifying patients with stable coronary artery disease. The detection model is characterized by biomarkers in plasma samples of patients to be tested and is constructed using a machine learning algorithm; wherein the biomarkers include: CDO homologous protein, galectin-9, adrenomedullin, angiotensin converting enzyme 2, brain natriuretic peptide, serine protease 8 and V-SET immunoglobulin domain-containing protein 2.

[0038] In one embodiment, the machine learning algorithms are logistic regression and support vector machine.

[0039] Specifically, this embodiment uses logistic regression and support vector machine to build the detection model, but is not limited to this. Other machine learning algorithms such as random forest, gradient boosting tree (GBDT), deep neural network (DNN), etc. can also be used for model construction. These models have advantages in processing high-dimensional data, nonlinear relationships or complex feature interactions, and can replace existing methods to improve the prediction ability and robustness of the model.

[0040] In one embodiment, the detection model uses a nomogram to predict the probability of stable coronary artery disease in the patient to be tested.

[0041] Specifically, the nomogram is a way to convert a complex regression model into a visualization tool. This embodiment uses a nomogram to convert the features of the detection model (such as CDO homologous protein, galectin-9, adrenomedullin, angiotensin-converting enzyme 2, brain natriuretic peptide and gender) into a corresponding scoring system. Each feature has a corresponding scale. The user can find the corresponding points in the figure according to the actual value of the patient, and then add the points of all features to obtain the total score. The total score corresponds to a predicted risk probability in the figure. Further, the calibration curve can be used to evaluate the predictive performance of the nomogram model. In addition, it is also possible to consider using decision curve analysis (DCA) to further evaluate the benefits of the model in clinical applications, and using indicators such as confusion matrix and F1 score to refine the performance evaluation of the classification model to ensure that the model not only performs well in AUC, but also achieves the expected results in other evaluation indicators.

[0042] In one embodiment, the method for constructing the detection model includes:

[0043] The OLINK proteomics platform was used to perform high-throughput protein detection on patients with stable coronary artery disease and healthy controls, and proteins with significant differential expression were screened for further functional analysis.

[0044] After functional analysis, the data features are refined and the data set is divided into a training set and a test set in a ratio of 3:1; the training set is used for model training and parameter tuning, while the test set is used to evaluate the generalization performance of the model;

[0045] Recursive feature elimination (RFE) combined with cross-validation was used to screen out multiple feature combinations, and then a unified machine learning method was used to build models based on each feature combination, thereby constructing multiple models.

[0046] Using accuracy (AC) as the evaluation criterion, the optimal detection model is selected.

[0047] Specifically, this embodiment uses the OLINK proteomics platform for high-throughput detection, which is a technical means with high sensitivity and wide coverage, but is not limited to this. Other similar proteomics detection platforms, such as SomaLogic or Mass Spectrometry (mass spectrometry technology), can also be considered to detect protein markers. These methods can provide accurate protein level measurements, thereby providing reliable input data for machine learning models. In addition, metabolomics or genomics data can also be used to replace proteomics to build similar detection models. Metabolomics can provide information on changes in metabolite levels, and genomics can reveal gene expression or variation. Through these omics data, combined with similar machine learning models, it is also possible to achieve effective identification of high-risk SCAD patients. Further, as an improvement to the proteomics detection method, it is possible to consider the use of multi-omics integrated analysis, including the comprehensive use of proteomics, metabolomics, transcriptomics and other data. Through the fusion of multi-omics data, biological information at different levels can be captured, which helps to build a more comprehensive detection model, thereby improving the accuracy and predictive power of the model.

[0048] The data set is divided into training set and test set in a ratio of 3:1 to ensure the training and verification effect of the model. In addition to recursive feature elimination (RFE), other feature selection methods can also be used, such as LASSO regression and tree-based feature importance analysis. These methods can also effectively screen out the features that contribute most to model prediction, thereby simplifying the model and improving its performance.

[0049] In one embodiment, the cross validation is 10-fold cross validation, but is not limited thereto. In order to further enhance the generalization ability of the model, different data partitioning methods may also be considered, such as 5-fold cross validation, to divide the data set into more subsets to ensure that each subset can participate in the training and validation of the model.

[0050] In one embodiment, the candidate proteins in the feature combination screened by recursive feature elimination combined with 10-fold cross validation can be further verified by independent experimental methods. For example, enzyme-linked immunosorbent assay (ELISA) or Western Blotting can be used to confirm whether the expression levels of these proteins in high-risk SCAD patients are consistent with the predicted results, thereby increasing the reliability and clinical translation potential of the detection scheme.

[0051] The present invention will be further described below by means of specific examples.

[0052] Example 1

[0053] The main steps of this embodiment are summarized as follows:

[0054] 1. Proteomic detection and differential protein screening: First, high-throughput protein detection was performed on high-risk SCAD patients and healthy controls through the OLINK proteomics platform to screen out proteins with significant differential expression, and further functional analysis was performed. Among them, high-risk SCAD patients refer to patients with severe coronary artery stenosis but no major adverse cardiovascular events (MACE); severe coronary artery stenosis refers to coronary angiography showing a reduction in the inner diameter of the coronary artery of 70% or more, and MACE includes cardiovascular death, non-fatal myocardial infarction, stroke, events requiring revascularization (such as percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG)), unstable angina, and worsening of heart failure.

[0055] 2. Unified machine learning modeling process: In order to ensure the scientificity and consistency of the model, a unified machine learning method is applied, including data feature refinement (DFR), candidate feature selection (CFS), final model construction (FMC) and other steps to build different types of detection models.

[0056] 3. Model construction and performance evaluation: Three detection models (protein model, clinical risk factor model, and comprehensive model) were constructed and compared, the performance of each model was evaluated, and the optimal feature combination was determined. In addition, nomograms and calibration curves were used to further evaluate the predictive performance of the optimal model.

[0057] The following sections explain each step in detail.

[0058] 1. Proteomic detection and differential protein analysis

[0059] 1.1 Proteomics platform and detection methods

[0060] In this example, 92 cardiovascular-related markers were detected by high-throughput detection using the OLINK proteomics platform to discover biomarkers associated with severe coronary artery stenosis. The OlinkCardiovascular II Panel, which contains 92 markers, was used to analyze plasma samples from 64 SCAD patients and 36 healthy controls using high-throughput proteomics technology based on relative quantitative intensity values. The clinical information of the patients is shown in Table 1.

[0061] (1) Definition of Intensity Value: Intensity value is a quantitative indicator of relative protein expression level obtained through the OLINK proteomics platform. These values ​​reflect the expression abundance of each target protein in the sample and are mainly used to compare the relative differences in protein expression between different samples or groups, rather than measuring their absolute concentrations, thereby helping to identify disease-related biomarkers.

[0062] (2) Method for obtaining intensity values: The process for obtaining intensity values ​​includes sample preparation, proximity extension analysis (PEA), PCR amplification and fluorescence detection, and data normalization and standardization. First, plasma samples from 64 high-risk SCAD patients and 36 healthy controls were collected and preprocessed to ensure their suitability for subsequent testing. Then, DNA probes linked to specific antibodies were used to identify and amplify the DNA sequence of the target protein using PEA technology. Next, the amplified DNA was amplified with high sensitivity using polymerase chain reaction (PCR), and the corresponding fluorescence intensity value was generated by fluorescence signal detection. Finally, the fluorescence signal was converted into Normalized Protein eXpression (NPX) values, which are usually expressed on a log2 scale and undergo strict quality control and standardization to eliminate technical variation and batch effects.

[0063] Table 1. Comparison of clinical characteristics between the CTR group and the high-risk SCAD group

[0064]

[0065]

[0066] Note: Values ​​are mean ± standard deviation or n (%).

[0067] 1.2 Differential protein analysis

[0068] (1) Principal component analysis (PCA): PCA results ( Figure 1 ) showed that the healthy control group and the high-risk SCAD patient group were significantly distinguished in terms of protein expression levels. This indicates that the plasma protein profiles of high-risk SCAD patients are quite different from those of healthy people, and these differences can be used to identify potential biomarkers for high-risk SCAD patients.

[0069] (2) Screening of significantly differentially expressed proteins: Differential protein analysis further screened out 32 proteins with significantly differentially expressed proteins between the healthy control group and the high-risk SCAD patient group based on relative quantitative intensity values ​​( Figure 2 and Figure 3 ), of which 26 proteins were upregulated in high-risk SCAD patients and 6 proteins were downregulated in high-risk SCAD patients. These differentially expressed proteins may be closely related to the pathological process of coronary artery stenosis and have potential clinical application value.

[0070] (3) Protein-protein interaction (PPI) analysis: In order to further understand the interactions of these differentially expressed proteins, this example performed a protein-protein interaction (PPI) network analysis ( Figure 4), the results showed that there was a significant interaction network between these differentially expressed proteins, suggesting that they may be involved in some common signaling pathways in the pathological process of SCAD.

[0071] (4) Gene Ontology (GO) enrichment analysis: Gene Ontology (GO) enrichment analysis ( Figure 5 ) revealed that these differentially expressed proteins were mainly involved in a number of key biological processes, such as the regulation of the ERK1 / 2 cascade, the regulation of body fluid levels, the MAPK cascade, and the regulation of systemic arterial blood pressure. These biological processes are closely related to the pathogenesis of SCAD, further supporting the potential of these proteins as biomarkers of SCAD.

[0072] 1.3 Differential proteins for machine learning model construction

[0073] Finally, these 32 proteins with significantly differential expression based on relative intensity values ​​were used to construct subsequent machine learning models to improve the ability to early identify high-risk SCAD patients.

[0074] 2. Unified Machine Learning Modeling Approach

[0075] 2.1 Overview of Machine Learning Process

[0076] In order to ensure the consistency of each model, this embodiment adopts a unified machine learning method to build the model, including the processes of data feature refinement (DFR), candidate feature selection (CFS) and final model construction (FMC). Figure 6 A detailed flow chart of the entire modeling process is presented.

[0077] 2.2 Data Feature Refinement (DFR)

[0078] (1) Removal of zero-variance features: First, remove features that do not change in all samples, i.e., zero-variance features. These features cannot provide effective information to distinguish high-risk SCAD patients from healthy controls.

[0079] (2) Collinearity feature refinement: Features with high collinearity are processed and the most representative features are retained, thereby reducing the complexity of the model and reducing the risk of overfitting.

[0080] (3) Statistical significance screening: A statistical test was performed on each feature to screen out features with a p-value less than 0.05 to ensure that each feature had a significant correlation with the target variable (the presence or absence of SCAD).

[0081] 2.3 Dataset Division

[0082] After feature refinement, this embodiment divides the data set into a training set and a test set in a ratio of 3:1 to ensure that the model can be trained and verified on different data, prevent overfitting, and evaluate the generalization ability of the model.

[0083] 2.4 Candidate Feature Selection (CFS)

[0084] In the candidate feature selection stage, this embodiment uses recursive feature elimination (RFE) combined with 10-fold cross validation (CV, repeated 10 times) to select the optimal feature combination. Recursive feature elimination is a model-based feature selection method that finds the best feature subset by gradually removing unimportant features. Combined with cross validation, the robustness of feature selection can be ensured and the model's dependence on specific data can be reduced.

[0085] 2.5 Final Model Construction (FMC)

[0086] After selecting the best features, this embodiment uses these features to build the final model. The model construction uses machine learning algorithms such as logistic regression and support vector machine, and uses accuracy (AC) as the evaluation criterion to select the optimal model architecture. The training set is used for model training and parameter tuning, while the test set is used to evaluate the generalization performance of the model.

[0087] 3. Detection model construction and performance evaluation

[0088] 3.1 Construction and judgment principles of three detection models

[0089] (1) Input data: For unknown samples, the intensity values ​​of related proteins are measured using the OLINK proteomics platform, and data on clinical risk factors are collected at the same time.

[0090] (2) Data preprocessing: The input intensity values ​​and clinical data are standardized and feature refined in the same way as during training to ensure that the data format of the new samples is consistent with the training model.

[0091] (3) Model prediction:

[0092] ① Protein model ( Figure 7AC): Using 32 differentially expressed proteins screened based on intensity values, 7 candidate proteins were further screened out through machine learning methods (recursive feature elimination and cross-validation), including galectin-9 (LGALS9), adrenomedullin (ADM), brain natriuretic peptide (NPPB), angiotensin-converting enzyme 2 (ACE2), serine protease 8 (PRSS8), V-SET immunoglobulin domain-containing protein 2 (VSIG2), and CDO homologous protein (BOC). The model had an AUC of 0.967 in the training set and an AUC of 1.000 in the test set, showing a high classification accuracy.

[0093] ②Clinical risk factor model ( Figure 7 DF): Through machine learning methods (recursive feature elimination and cross-validation), the following eight key risk factor combinations were finally screened out from clinical variables: diabetes, high-density lipoprotein cholesterol (HDL), low-density lipoprotein cholesterol (LDL), gender, age, total cholesterol (CHOL), creatinine (Cr) and alanine aminotransferase (ALT). Using these risk factors to build a model, the AUC of the training set was 0.969, while the AUC of the test set was 0.790, which was slightly inferior to the protein model.

[0094] ③ Comprehensive model ( Figure 7 Medium GI): The 32 differentially expressed proteins screened based on the intensity value were fused with all clinical risk factors, and feature selection was performed using machine learning methods. Finally, 5 key protein markers (BOC, LGALS9, ADM, ACE2, NPPB) and 1 clinical risk factor (sex, male was recorded as 1) were screened out to construct a comprehensive model. The AUC of the comprehensive model in the training set and the test set were 0.994 and 1.000, respectively, showing the best classification effect, and it is the optimal model of the present invention.

[0095] (4) Judgment formula and standard: Each model learns a judgment formula through a machine learning algorithm (logistic regression and support vector machine are used in this embodiment), which is expressed as:

[0096]

[0097] Among them, β 0 is the intercept, β 1 ,β 2 ,…,β n is the model parameter, X 1 ,X 2 ,…,X n are input features (relative expression levels of proteins and clinical risk factors).

[0098] (5) Risk judgment: The predicted probability value output by the model is compared with the pre-set threshold to determine whether the sample is a high-risk SCAD patient. The pre-set threshold is determined by maximizing the Youden index in the ROC curve analysis.

[0099] 3.2 Model performance evaluation method

[0100] (1) Acquisition and application of ROC curve: ROC curve (Receiver Operating Characteristic Curve) is used to evaluate the classification performance of the model under different thresholds. Its generation steps are: ① Use the training set to build the model and evaluate it on the test set. ② For each sample in the test set, the model outputs a predicted probability value. ③ Draw the ROC curve: The horizontal axis is the false positive rate (False Positive Rate, FPR), and the vertical axis is the true positive rate (True Positive Rate, TPR). By changing the judgment threshold, calculate the FPR and TPR under different thresholds, and draw the ROC curve. ④ Calculate the AUC value: The area under the ROC curve (Area Under the Curve, AUC) is used to quantify the overall performance of the model. The closer the AUC value is to 1, the stronger the classification ability of the model.

[0101] (2) Threshold setting: The threshold is used to convert the predicted probability output by the model into a binary classification result (high risk or low risk). The optimal threshold is determined by ROC curve analysis using the method of maximizing Youden's J Statistic, that is, the threshold that maximizes (TPR-FPR) is selected as the classification standard. For example, if the optimal threshold is 0.5, samples with a predicted probability ≥ 0.5 are judged as high-risk SCAD patients, otherwise they are low-risk.

[0102] 3.3 Application of Nomogram and Calibration Curve

[0103] (1) Nomogram Figure 8 A): A nomogram is a way to convert a complex regression model into a visualization tool for predicting the probability of a single high-risk SCAD patient. In this embodiment, a nomogram is used to convert the features of the comprehensive model (including 5 key proteins and 1 clinical feature gender) into a corresponding scoring system. Each feature has a corresponding scale, and the user can find the corresponding number of points in the graph according to the actual intensity value of the key protein in the patient's plasma and the gender (male is marked as 1), and then add up the points of all features to get the total score, which corresponds to a predicted risk probability in the graph.

[0104] (2) Calibration curve Figure 8Middle B): The calibration curve is used to evaluate the predictive performance of the nomogram model, showing the degree of fit between the model's predicted values ​​and the actual observed values. In this embodiment, the calibration curve evaluates the predictive performance of the comprehensive model for 100 samples (64 high-risk SCAD patients and 36 healthy controls). The horizontal axis in the calibration curve represents the probability of SCAD predicted by the model, and the vertical axis represents the actual observed probability of SCAD. Ideally, the predicted values ​​of the model should be completely consistent with the actual values, so the calibration curve should be distributed along the diagonal.

[0105] By combining the nomogram with the calibration curve, the comprehensive model of the present invention demonstrates its intuitiveness and accuracy in risk prediction, providing clinicians with an effective and practical decision support tool to help better screen and manage high-risk SCAD patients in clinical practice.

[0106] 3.4 Effects of the technical solution of this embodiment

[0107] The technical solution of this embodiment provides an efficient and accurate means to identify high-risk SCAD patients by combining proteomics detection and machine learning methods, significantly improving the defects of the existing technology. The following is a detailed description of the overall effect of this embodiment and the technical effects directly caused by certain technical features:

[0108] (1) Overall effect

[0109] Early identification of high-risk SCAD patients: Through proteomic differential analysis based on intensity values ​​between high-risk SCAD patients and healthy controls, this example can screen out significantly differentially expressed proteins associated with severe coronary artery stenosis. These proteins are used as biomarkers in the construction of machine learning models, making the detection model highly sensitive and specific, and can effectively identify high-risk SCAD patients at an early stage, helping clinicians take timely intervention measures.

[0110] Non-invasive detection: Compared with existing invasive diagnostic methods, the proteomics-based biomarker detection in this embodiment can be analyzed through plasma samples, achieving non-invasive detection. This not only reduces patient discomfort and medical risks, but is also more suitable for large-scale screening and early warning.

[0111] Significant improvement in model performance: This example constructs different types of detection models (protein model, clinical risk factor model, comprehensive model), and further evaluates their performance through nomograms and calibration curves, verifying the superior performance of the comprehensive model on the training set and test set, and can accurately predict the risk probability of high-risk SCAD patients.

[0112] (2) Specific effects brought by key technical features

[0113] Combination of proteomics and machine learning: Proteomics testing provides rich differentially expressed protein data based on relative intensity values, while machine learning methods ensure that the model can accurately distinguish high-risk SCAD patients from healthy controls through data feature refinement, candidate feature selection and model construction. This combination enables the detection model to have higher recognition ability, especially the AUC of the comprehensive model performs well in both the training set and the test set, proving the reliability and effectiveness of this method.

[0114] Recursive feature elimination (RFE) and cross-validation: This example uses recursive feature elimination combined with 10-fold cross-validation to select the optimal features. This method can reduce the model's dependence on a specific data set, enhance the model's generalization ability, and make the selected feature subset more robust.

[0115] Application of nomograms and calibration curves: The nomogram is used to simplify the complex prediction process of the model into a visualization tool, allowing clinicians to obtain individualized risk prediction probabilities based on the specific characteristics of the patient. The calibration curve evaluates the fit between the model prediction value and the actual observed value, further verifying the reliability and prediction accuracy of the model in practical applications.

[0116] In summary, the present invention discovers specific biomarkers associated with severe coronary artery stenosis through proteomics technology, and applies these biomarkers to construct a detection model for identifying high-risk SCAD patients. The model has high sensitivity and high specificity, and can accurately identify high-risk SCAD patients non-invasively before major adverse cardiovascular events (MACE) occur, guide clinical interventions, such as timely implementation of treatment measures such as PCI, and avoid irreversible myocardial damage. The detection scheme of the present invention overcomes the shortcomings of traditional diagnostic methods in the early identification of high-risk SCAD patients, not only provides higher detection accuracy, but also has the advantages of simple operation and wide adaptability in clinical applications. By applying this scheme in early screening, high-risk patients can be found in time, helping clinicians to more effectively formulate personalized treatment plans, thereby reducing the risk of MACE.

[0117] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A set of biomarkers for detecting stable coronary artery disease, characterized in that: The biomarkers include: CDO homologous protein, galectin-9, adrenomedullin, angiotensin converting enzyme 2 and brain natriuretic peptide.

2. The biomarker according to claim 1, characterized in that The biomarkers also include: serine protease 8 and V-SET immunoglobulin domain-containing protein 2.

3. Use of the biomarker according to claim 1 or 2 in the preparation of a stable coronary artery disease detection product.

4. Use of a reagent for detecting individual information in the preparation of a stable coronary artery disease detection product, characterized in that: The individual information includes: clinical risk factors of the patient to be tested and the relative expression level of the biomarker as claimed in claim 1 in the plasma sample.

5. Use of a reagent for detecting individual information in the preparation of a stable coronary artery disease detection product, characterized in that: The individual information includes: the relative expression level of the biomarker as claimed in claim 2 in the plasma sample of the patient to be tested.

6. The use according to claim 4 or 5, characterized in that: The stable coronary artery disease detection product includes a detection reagent, a detection kit, a detection chip and a detection model.

7. A detection model for identifying patients with stable coronary artery disease, characterized in that: The detection model is characterized by the clinical risk factors of the patient to be tested and the relative expression levels of the biomarkers as described in claim 1 in the plasma sample, and is constructed using a machine learning algorithm.

8. A detection model for identifying patients with stable coronary artery disease, characterized in that: The detection model is characterized by the relative expression level of the biomarker as described in claim 2 in the plasma sample of the patient to be tested and is constructed using a machine learning algorithm.

9. The detection model according to claim 7 or 8, characterized in that: The machine learning algorithms are logistic regression and support vector machine.

10. The detection model according to claim 7 or 8, characterized in that: The detection model uses a nomogram to predict the probability of stable coronary artery disease in the patient to be tested.

Images