Coronary heart disease hierarchical diagnosis system based on multiple omics and application thereof

Abstract

The invention discloses a multi-omics-based coronary heart disease hierarchical diagnosis system and application thereof, and relates to application of personal information in a system for diagnosing whether a subject has no coronary heart disease (NCA) or suffers from the coronary heart disease, and the coronary heart disease comprises stable coronary heart disease (sCAD) and acute coronary syndrome (ACS). The multi-dimensional feature set is constructed by integrating clinical, microbial and metabolic features, potential information provided by data of different dimensions is fully utilized, and the diagnostic performance of the model is remarkably improved; through a Boruta algorithm and an SHAP framework, a series of new metabolite and microbial biomarker information is found, and the method can be used for accurately distinguishing the coronary heart disease and the stable coronary heart disease acute coronary syndrome.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, and in particular to a multi-omics-based stratified diagnostic system for coronary heart disease and its application. Background Technology

[0002] Coronary artery disease (CAD) is an ischemic disease of the myocardium caused by coronary atherosclerosis and is one of the leading causes of death and disability worldwide. Traditional diagnosis of CAD relies primarily on clinical characteristics such as age, sex, blood pressure, blood lipid levels, blood glucose levels, smoking history, and family history. While these clinical characteristics play an important role in initial screening and risk assessment of CAD, they have limitations in differentiating between different stages of CAD, especially between non-coronary artery disease (NCA) and acute coronary syndrome (ACS).

[0003] In recent years, with the rapid development of biomedical technology, especially the rise of microbiome and metabolomics, an increasing number of studies have shown that changes in gut microbiota and metabolites are closely related to the occurrence and development of coronary heart disease. As the largest microbial community in the human body, the gut microbiota participates in various physiological and pathological processes, including immune regulation, metabolic function, and inflammatory responses. Metabolomics studies have revealed changes in metabolites in patients with coronary heart disease, and these changes may be closely related to disease progression and clinical manifestations. However, existing studies mostly focus on single-dimensional analysis, such as relying solely on clinical characteristics, microbiome, or metabolome data, lacking multidimensional comprehensive analysis. This results in limited model performance and an inability to fully utilize the potential information provided by these multidimensional data.

[0004] The main limitation of existing technologies lies in their failure to effectively integrate clinical, microbiological, and metabolic features, resulting in insufficient model performance in distinguishing different stages of coronary artery disease (CAD). Furthermore, existing technologies fail to identify and utilize key biomarkers in the microbiome and metabolites, and cannot capture dynamic changes during disease progression. These limitations restrict the application of existing technologies in the early diagnosis and precise stratification of CAD.

[0005] Currently, most coronary artery disease (CAD) diagnostic models rely solely on traditional clinical characteristics. These models collect information such as patient age, sex, blood pressure, and lipid levels, and then use statistical methods or simple machine learning algorithms for risk assessment. For example, some studies use logistic regression models, taking traditional clinical characteristics as input variables, to predict the probability of a patient developing CAD. These models, relying solely on traditional clinical characteristics, perform poorly in distinguishing different stages of CAD, demonstrating limited discriminative ability. This is mainly because traditional clinical characteristics cannot capture the subtle changes in CAD at different stages, especially the dynamic changes between stable and acute states. Some studies are attempting to incorporate metabolomics data into CAD diagnostic models. These models improve diagnostic accuracy by detecting specific metabolites in a patient's blood or urine, combined with traditional clinical characteristics. For example, some studies use gas chromatography-mass spectrometry (GC-MS) to detect metabolites and then utilize machine learning algorithms to build diagnostic models.

[0006] While these models have improved diagnostic accuracy to some extent, they still have limitations. First, these models primarily focus on changes in metabolites, neglecting the important role of the gut microbiota in the occurrence and development of coronary artery disease. Second, when distinguishing different stages of coronary artery disease, especially between non-coronary artery disease (NCA) and acute coronary syndrome (ACS), these models still have AUC values ​​below 0.8, demonstrating limited discriminatory power. Summary of the Invention

[0007] To address the aforementioned problems in the prior art, this invention provides a multi-omics-based stratified diagnostic system for coronary heart disease and its application.

[0008] One object of the present invention is to provide the application of personal information in a system for diagnosing whether a subject has coronary heart disease or not.

[0009] Another objective of this invention is to provide a method for constructing a diagnostic model to distinguish between subjects who have coronary heart disease and those who do not.

[0010] Another object of the present invention is to provide a diagnostic system for distinguishing between subjects who do not have coronary heart disease and those who have coronary heart disease.

[0011] Specifically, on the one hand, the present invention provides the application of personal information in a system for diagnosing whether a subject has no coronary artery disease (NCA) or has coronary artery disease, wherein the coronary artery disease includes stable coronary artery disease (sCAD) and acute coronary syndrome (ACS);

[0012] (1) When distinguishing between non-coronary artery disease and stable coronary artery disease, the personal information mentioned includes:

[0013] Paraprevotella, Phocaeicola, Bacteroides, Mediterranean bacillus, Clostridium, Gemmiger, Odoribacter, salicylic acid, phenylacetylglutamine, carnitine, 2-hydroxybutyric acid, 3-hydroxybutyric acid, phenylalanine, betaine;

[0014] (2) When distinguishing between those without coronary artery disease and acute coronary syndrome, the personal information mentioned includes:

[0015] Alistipes, Paraprevotella, Phocaeicola, Desulfovibrio, Clostridium, Lachnospiraceae, Haemophilus, Bilophila, Dysosmobacter, Faecalibacillus, Clostridium, 2-hydroxybutyric acid, salicylic acid, lysine;

[0016] (3) When differentiating between stable coronary artery disease and acute coronary syndrome, the personal information mentioned includes:

[0017] Bacteroides, Lachnospira, Pseudoflavonifractor, Blautia, Megamonas, Prevotella, Anaerostipes, Pyramidobacter, Romboutsia, Faecalibacillus, 3-hydroxybutyric acid, 2-hydroxybutyric acid, phenylacetylglutamine, succinic acid, indoleacrylic acid, tryptophan, salicylic acid, tyrosine.

[0018] In the above applications, preferably, (2) when distinguishing between non-coronary heart disease and acute coronary syndrome, the personal information also includes: GGB3490 SGB4664 (Firmwallis), GGB9713 SGB15249.

[0019] In the above applications, preferably, (3) when distinguishing between stable coronary heart disease and acute coronary syndrome, the personal information also includes: GGB9758 SGB15368 (Bacteroidetes).

[0020] In the above applications, preferably, the personal information includes clinical characteristic information, gut microbiota species information, and metabolite information, wherein:

[0021] (1) When distinguishing between non-coronary artery disease and stable coronary artery disease, the personal information mentioned includes:

[0022] Clinical characteristics: hypertension, age, total cholesterol, LDL cholesterol;

[0023] Gut microbiota species information: Pararaprevotella clara, Prevotella marseillensis, Phocaeicola plebeius, Phocaeicola massiliensis, Bacteroides nordii, Mediterraneibacter butyricigenes, Clostridium symbiosum, Gemmiger formicilis, and Odoribacter splanchnicus;

[0024] Metabolite information: salicylic acid, phenylacetylglutamine, carnitine, 2-hydroxybutyric acid, 3-hydroxybutyric acid, phenylalanine, betaine;

[0025] (2) When distinguishing between those without coronary artery disease and acute coronary syndrome, the personal information mentioned includes:

[0026] Clinical characteristics: total cholesterol, age;

[0027] Intestinal flora species information: Alistipes onderdonkii, GGB3490SGB4664, Pararaprevotella clara, Phocaeicola plebeius, Desulfovibrio SGB5077, Clostridium sp. AF20_17LB, Lachnospiraceae bacterium AM48_27BH, Haemophilus parainfluenzae, and Bilophila wadsworthia; GGB9713 SGB15249, Dysosmobacter sp.NSJ_60, Faecalibacillus intestinalis, Clostridium sp.AF27_2AA;

[0028] Metabolite information: 2-hydroxybutyric acid, salicylic acid, lysine;

[0029] (3) When differentiating between stable coronary artery disease and acute coronary syndrome, the personal information mentioned includes:

[0030] Gut microbiota species information: Bacteroides eggerthii, Lachnospira pectinoschiza, Pseudoflavonifractorgallinarum, Blautia glucerasea, Megamonas funiformis, Prevotella, Prevotella SGB1687, GGB9758 SGB15368, Anaerostipes hadrus, Pyramidobacter piscolens [Pyramidobacter], Romboutsiatimonensis [Romboutsia], Faecalibacillus intestinalis [Faecalibacillus].

[0031] In the above application, preferably, the SHAP values ​​(impact on model output) of the personal information are as follows:

[0032] Salicylic acid, Pararaprevotella clara, LDL cholesterol, phenylacetylglutamine, Prevotella marseillensis, carnitine, Phocaeicola plebeius, Phocaeicola massiliensis, hypertension, Bacteroides nordii, age, 2-hydroxybutyric acid, 3-hydroxybutyric acid, total cholesterol, phenylalanine, Mediturneibacter butyricigenes, Clostridium symbolum, betaine, Gemmigerformicilis, Odoribacter splanchnicus; or

[0033] 2-Hydroxybutyric acid, salicylic acid, Alistipes onderdonkii, total cholesterol, lysine, GGB3490 SGB4664, Pararaprevotella clara, age, Phocaeicola plebeius, Desulfovibrio SGB5077, Clostridium sp. AF20_17LB, Lachnospiraceae bacterium AM48_27BH, Haemophilus parainfluenzae, Bilophila wadsworthia; or

[0034] 3-Hydroxybutyric acid, 2-Hydroxybutyric acid, Phenylacetylglutamine, Succinic acid, Indoleacrylic acid, Bacteroideseggerthii, Tryptophan, Lachnospira pectinoschiza, Pseudoflavonifractor gallinarum, Blautia glucerasea, Megamonas funiformis, Salicylic acid, Prevotella stercorea, Prevotella SGB1687, GGB9758 SGB15368, Tyrosine, Anaerostipes hadrus, Pyramidobacterpiscolens, Romboutsia timonensis, Faecalibacillus intestinalis.

[0035] The gut microbiota species information described in this invention is the result of metaphlan4 annotation, and the corresponding databases all contain genus and species information, making them traceable and verifiable. Nomenclature (with prefixes such as CFGB, OFGB, FGB, GGB, etc.) is commonly used in metagenomics research to identify microorganisms discovered through sequencing but not yet successfully isolated, cultured, or formally taxonomized in the laboratory. Existing microbial databases (such as NCBI and Greengenes) have insufficient genomic coverage of rare or novel species, resulting in the inability to accurately match known species. Although unidentified species lack precise annotation, their potential roles in the gut microbiota can still be revealed through phylogenetic inference, functional gene analysis, ecological network localization, and abundance-phenotype associations.

[0036] On the other hand, the present invention provides a method for constructing a diagnostic model to distinguish between subjects with and without coronary heart disease, wherein the construction method includes:

[0037] Data collection: Clinical characteristics, gut microbiota species information, and metabolite characteristics of the population sample were collected to construct a multidimensional feature set; the population sample included subjects without coronary heart disease and patients with coronary heart disease;

[0038] Feature selection: Calculate the correlation between each feature and subjects without coronary heart disease and patients with coronary heart disease, and select features using the Boruta algorithm to screen out key features for diagnosing whether a subject has coronary heart disease or not. The key features include the personal information mentioned above.

[0039] Model construction: The population samples are randomly divided into a discovery set and an independent validation set. The discovery set is input into a model built based on a machine learning algorithm, and the model is trained using a 10-fold cross-validation method to obtain the diagnostic model.

[0040] The coronary heart disease mentioned includes stable coronary heart disease and acute coronary syndrome.

[0041] According to some specific embodiments of the present invention, the machine learning algorithm includes one or more combinations of random forest, support vector machine (SVM), and neural network (NN).

[0042] According to some specific embodiments of the present invention, the machine learning algorithm is the LightGBM algorithm.

[0043] The algorithms used in this invention have certain advantages in processing high-dimensional data and complex feature relationships, and may further improve the performance of the model.

[0044] According to some specific embodiments of the present invention, preferably, the construction method further includes:

[0045] Data preprocessing: Data preprocessing was performed after data collection to remove missing and outlier values. Clinical characteristic information was normalized, and the relative abundance of gut microbiota species and metabolite characteristics were standardized.

[0046] According to some specific embodiments of the present invention, preferably, the construction method further includes:

[0047] Model building: During model training, the GridSearchCV method is used to adjust the model hyperparameters;

[0048] Model evaluation: The diagnostic model was tested using an independent validation set, and its performance was evaluated using the AUROC metric to obtain the model evaluation results.

[0049] According to some specific embodiments of the present invention, preferably, the construction method further includes: interpreting the diagnostic model using the SHapleyAdditive exPlanations (SHAP) framework to verify the importance of key features.

[0050] This invention provides a method for constructing a diagnostic model for diagnosing subjects with or without coronary heart disease, based on clinical characteristics, gut microbiota species, and metabolite information. In some specific embodiments of this invention, the method integrates clinical characteristic information, including basic information (such as age) and biochemical parameters (such as blood lipids, blood glucose, blood pressure, etc.), gut microbiota species information, including gut microbiota species composition and relative abundance information (obtained through metagenomic sequencing technology), and targeted quantitative metabolite information (obtained through UPLC-MS / MS technology), to construct a multidimensional feature set. Feature selection is performed using the Boruta algorithm to screen out key features that significantly contribute to the model's predictive ability. The LightGBM algorithm is used as the core machine learning algorithm, and validation is performed through 10-fold cross-validation and independent test datasets to ensure the model's robustness and generalization ability. Hyperparameter optimization is performed using the GridSearchCV method, which exhaustively searches a predefined parameter grid to find the optimal combination of hyperparameters, further improving model performance.

[0051] On the other hand, the present invention provides a diagnostic system for distinguishing between subjects who have and do not have coronary heart disease, wherein the diagnostic system comprises:

[0052] The data input module is used to input the aforementioned personal information of the subjects.

[0053] The data analysis module, connected to the data input module, inputs the subject's personal information into the diagnostic model obtained by the above construction method for analysis;

[0054] The results output module is connected to the data analysis module and outputs the diagnostic results of whether the subject has no coronary heart disease or has coronary heart disease;

[0055] The coronary heart disease mentioned includes stable coronary heart disease and acute coronary syndrome.

[0056] On the other hand, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements: a diagnosis result that distinguishes between a subject having no coronary heart disease or having coronary heart disease based on the subject's personal information;

[0057] The coronary heart disease mentioned includes stable coronary heart disease and acute coronary syndrome;

[0058] The combination of the markers is the same as the personal information mentioned above.

[0059] On the other hand, the present invention provides a computer-readable storage medium storing computer program instructions, which, when executed, achieve the following: a diagnosis result that distinguishes between a subject having no coronary heart disease and a subject having coronary heart disease based on the subject's personal information;

[0060] The coronary heart disease mentioned includes stable coronary heart disease and acute coronary syndrome;

[0061] The combination of the markers is the same as the personal information mentioned above.

[0062] This invention significantly improves the accuracy and reliability of distinguishing different stages of coronary artery disease (CAD) by integrating clinical, microbiological, and metabolic features to construct a machine learning model. Through multidimensional feature integration, this invention not only enhances the diagnostic performance of the diagnostic system but also provides new tools and methods for the early diagnosis and precise stratification of CAD. This invention aims to construct a diagnostic system that can more accurately distinguish between non-coronary artery disease (NCA), stable coronary artery disease (sCAD), and acute coronary syndrome (ACS) by integrating clinical features, gut microbiota species, and metabolites. This system not only improves diagnostic accuracy but also captures subtle changes in CAD at different stages, providing stronger support for clinical diagnosis and treatment. Attached Figure Description

[0063] Figure 1 This is a data graph showing the performance analysis and feature contribution analysis of the diagnostic model used to distinguish between non-coronary artery disease (NCA) and patients with coronary artery disease in Embodiment 1 of the present invention. Detailed Implementation

[0064] In order to provide a clearer understanding of the technical features, objectives and beneficial effects of the present invention, the technical solution of the present invention will now be described in detail below, but it should not be construed as limiting the scope of implementation of the present invention.

[0065] In some specific embodiments, the present invention provides a method for constructing a diagnostic model to distinguish between subjects without coronary heart disease and those with coronary heart disease, wherein the construction method includes:

[0066] 1. Data Collection

[0067] (1) Clinical characteristics: The participants’ age, gender, blood pressure, blood lipid level, blood glucose level, smoking history, family medical history and other clinical characteristics were collected.

[0068] (2) Gut microbiota species information: Gut microbiota species information was obtained through metagenomic sequencing technology, a process that encompassed the entire workflow from sample collection to bioinformatics analysis. Fresh fecal samples were collected from participants, DNA was extracted, high-throughput sequencing was performed, and data analysis was conducted to obtain information on the composition and relative abundance of gut microbiota species.

[0069] (3) Metabolite information: Metabolites and lipid molecules in patient plasma were detected using ultra-high performance liquid chromatography-mass spectrometry (UPLC-MS / MS). The specific steps included sample collection, metabolite extraction, instrument detection, and data analysis.

[0070] 2. Data Preprocessing

[0071] Data cleaning: Remove missing and outlier values ​​to ensure data integrity and consistency.

[0072] Normalization: Clinical features are normalized by scaling the data to the range of 0 to 1 to eliminate dimensional differences between different features.

[0073] Standardization processing: The information on gut microbiota species and metabolites is standardized to conform to a normal distribution, which facilitates subsequent statistical analysis and model construction.

[0074] 3. Feature Selection

[0075] In the feature selection stage, the Boruta algorithm is used to evaluate feature importance. The Boruta algorithm is an extension of the random forest algorithm that can eliminate unimportant features through rigorous statistical tests, while retaining features that significantly contribute to the model's predictive ability.

[0076] 4. Model Building

[0077] Machine Learning Algorithm Selection: Appropriate machine learning algorithms are selected, such as Random Forest, Support Vector Machine (SVM), and Neural Network (NN). This invention primarily employs the LightGBM algorithm due to its superior performance in handling high-dimensional data and complex feature relationships. LightGBM is an efficient algorithm based on Gradient Boosting Decision Tree (GBDT), with advantages in processing high-dimensional data and complex feature relationships. First, the dataset is divided into training and validation sets, and a 10-fold cross-validation method is used for model training to ensure model stability and generalization ability. During training, the GridSearchCV method is used to optimize the model's hyperparameters by exhaustively searching a predefined parameter grid to find the optimal combination of hyperparameters.

[0078] 5. Model Evaluation

[0079] Independent validation set: The constructed model is evaluated using an independent validation set that has never participated in the modeling process. The model's AUC value, sensitivity, specificity, and other metrics are calculated to verify the model's performance.

[0080] Performance metrics: Evaluation metrics such as AUC, sensitivity, and specificity of the model in distinguishing between non-coronary artery disease (NCA), stable coronary artery disease (sCAD), and acute coronary syndrome (ACS).

[0081] The following is a specific embodiment illustrating the present invention's multi-omics-based stratified acute coronary syndrome diagnostic system and its application.

[0082] Example 1

[0083] 1. Data Collection

[0084] Data collection for this study involved the consecutive recruitment of 285 participants at Fuwai Hospital, National Center for Cardiovascular Diseases, China, between March and December 2018. All participants had complete medical history, clinical and biochemical parameters. Based on clinical symptoms, signs, laboratory tests, electrocardiogram (ECG), coronary CT angiography (CTA), and coronary angiography results, participants were divided into three groups: non-coronary artery disease (NCA), stable coronary artery disease (sCAD), and acute coronary syndrome (ACS). Specific grouping criteria are as follows:

[0085] (1) Non-coronary artery disease group (NCA group, N=73): including healthy subjects or suspected patients whose coronary CTA or angiography did not show visible lesions or stenosis, and who had no symptoms and signs related to coronary artery disease.

[0086] (2) Stable coronary artery disease group (sCAD group, N=106): Patients needed to meet the criteria of coronary angiography showing stenosis ≥50% in the lumen diameter of at least one coronary artery and no history of acute coronary syndrome (ACS) prior to enrollment. In addition, a separate recovery sCAD group (N=52) was included in the validation analysis. These patients had experienced an ACS event at least 3 months prior to enrollment and recovered to a stable state through revascularization and drug therapy.

[0087] (3) Acute Coronary Syndrome Group (ACS group, N=106): Patients were clinically diagnosed with one of the following cardiovascular events: unstable angina (UAP), ST-segment elevation myocardial infarction (STEMI), or non-ST-segment elevation myocardial infarction (NSTEMI). Diagnostic criteria included elevated cardiac troponin I (cTnI), ischemic symptoms lasting more than 20 minutes, ST-T segment changes on electrocardiogram, or the appearance of new pathological Q waves.

[0088] Exclusion criteria include: other types of myocardial infarction (such as type 2, 3, 4a, 4b or 5), severe heart failure or cardiogenic shock, mechanical complications of myocardial infarction, cardiac arrest or cardiopulmonary resuscitation, use of antibiotics for more than 1 week in the past 3 months, history of ACS or coronary revascularization in the past 3 months, history of trauma or surgery in the past 3 months, history of cerebrovascular disease or gastrointestinal bleeding in the past 3 months, chronic gastrointestinal disease, tumors, autoimmune diseases, renal insufficiency (creatinine >3.0 mg / dL) or history of kidney transplantation, and intake of probiotics, prebiotics or yogurt in the past 7 days.

[0089] Data collection content

[0090] (1) Clinical features:

[0091] Basic Information: Basic information of participants was collected through face-to-face questionnaires, including demographic characteristics, health status, medical history, family history, medication history, work experience, gastrointestinal status, educational background, and physical activity level. Body mass index (BMI) was calculated, and waist circumference was measured (horizontally at 5 cm above the navel). Smoking was defined as smoking at least one cigarette per day for more than one year, and alcohol consumption was defined as at least 20g per day for men and at least 10g per day for women for more than one year.

[0092] Biochemical parameters:

[0093] Peripheral blood samples were collected after a 12-hour fast using EDTA anticoagulant tubes and tested as soon as possible. Plasma total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), apolipoprotein A (apoA), apolipoprotein B (apoB), and fasting blood glucose (FBG) concentrations were measured using a HITACHI 7150 automated analyzer. High-sensitivity C-reactive protein (hs-CRP) and plasma NT-proBNP levels were determined using immunoturbidimetry. Serum lipoprotein(a) (Lp(a)) concentration was measured using immunoturbidimetry, and creatinine (Scr) was measured using the Jaffe method.

[0094] (2) Gut Microbiota Species Information: Gut microbiota species information was obtained through metagenomic sequencing technology, a process encompassing the entire workflow from sample collection to bioinformatics analysis. First, fresh fecal samples were collected from patients. Immediately after collection, the samples were placed in ice packs and rapidly transported to the laboratory using insulated polystyrene foam containers to ensure sample integrity and microbial viability. Subsequently, EZNA was used... TMFecal DNA extraction kits (Omega Bio-Tek, VWR, Herlev, Denmark) are used to extract microbial DNA from samples. DNA suitable for sequencing should have high purity (OD ratio meets requirements), sufficient concentration, and no signs of degradation. The extracted DNA is eluted with 50 μL of elution buffer and stored at -80°C for subsequent experiments.

[0095] After DNA extraction, DNA libraries were constructed using the TruSeq Nano DNA LT Library Preparation Kit (FC-121-4001, Illumina, San Diego, CA, USA). First, the DNA was fragmented by incubation at 37°C for 30 minutes using dsDNA Fragmentation (M0348S, NEB, MA, USA). Subsequently, the fragments were end-trimmed using a filling reaction and exonuclease activity, and sequencing adapters were ligated to the fragment ends. The ligation products were amplified by PCR under the following conditions: initial denaturation at 95°C for 3 minutes; 8 cycles of denaturation (98°C for 15 seconds), annealing (60°C for 15 seconds), and extension (72°C for 30 seconds); and a final extension at 72°C for 5 minutes. The amplified library underwent quality control at LC-Bio Technology Co., Ltd. (Hangzhou, China). After passing the quality control, it was sequenced using an Illumina HiSeq 4000 sequencer in 150bp paired-end sequencing mode.

[0096] After sequencing, the large amount of raw data generated is processed through a bioinformatics analysis workflow. First, the raw sequences undergo quality control, including the removal of low-quality sequences and adapter contamination: professional bioinformatics tools, such as FastQC, are used to check sequence quality, and Trimmomatic or Cutadapt are used to remove low-quality sequences and adapter contamination. By setting quality thresholds and adapter sequence patterns, high-quality sequences are selected for subsequent analysis. Then, MetaPhlAn4 (https: / / huttenhower.sph.harvard.edu / metaphlan / ) is used to annotate the high-quality sequences and analyze the species composition and abundance information of the gut microbiota. The process of using MetaPhlAn4 to analyze the species composition and abundance information of the gut microbiota is as follows: First, ensure that MetaPhlAn4 is installed and that high-quality gut metagenomic sequencing data (e.g., in fastq format) is prepared. Next, run the command `metaphlan input.fastq.gz --input_type fastq --unknown_estimation --output_file output.txt`, where `input.fastq.gz` is the input file and `output.txt` is the output file. MetaPhlAn4 aligns sequences with a marker gene database, calculates the relative abundance of each taxonomic unit, and outputs the results. Finally, abundance information at the desired taxonomic level can be extracted from output.txt for subsequent analysis and visualization. Furthermore, functional gene prediction and metabolic pathway analysis can be performed to reveal the potential role of the gut microbiota in disease occurrence and development. This series of steps ensures high accuracy and reliability of gut microbiota species information, providing an important data foundation for subsequent multi-dimensional feature integration and machine learning modeling.

[0097] (3) Metabolite Information: Metabolites and lipid molecules in patient plasma were detected using ultra-high performance liquid chromatography-mass spectrometry (UPLC-MS / MS). Specific steps included sample collection, metabolite extraction, instrument detection, and data analysis. First, fasting plasma samples were collected from patients using heparin anticoagulant tubes. The samples were immediately cooled to 4°C and centrifuged (1,350 rpm, 12 minutes) within one hour of collection to separate the plasma, which was then aliquoted and stored at -80°C. For metabolite extraction, 100 μL of plasma sample was taken, and 20 μL of internal standard solution (including DL-valine-d8, L-tryptophan-d5, succinic acid-d4, etc.) was added. Then, 3 mL of methanol / methyl tert-butyl ether (1:1, v / v) mixed solvent was added, vortexed for 5 minutes (2,500 rpm), and centrifuged at 4°C for 10 minutes (4,200 rpm). Transfer the supernatant to a new tube, add 3.0 mL of methyl tert-butyl ether and 1.2 mL of purified water, vortex again for 15 minutes (2,500 rpm) and centrifuge for 10 minutes (4,200 rpm). Collect the organic and aqueous phases separately and dry them under a nitrogen stream. Resolve the aqueous phase residue in 100 μL of acetonitrile / water (2:98, v / v) for metabolomics analysis, and redissolve the organic phase residue in 400 μL of methanol / chloroform (1:1, v / v) for lipidomics analysis.

[0098] Instrumental analysis was performed using a two-dimensional ultra-high performance liquid chromatography system (ACQUITY UPLC I-Class, Waters) coupled with a high-resolution mass spectrometer (Q-Exactive, Thermo Fisher Scientific). Metabolite separation was performed on a Waters HSS T3 column (1.8 μm, 100 mm × 2.1 mm) at column temperatures of 35 °C (polar metabolites) and 45 °C (lipid metabolites), with a flow rate of 0.25 mL / min and injection volumes of 10 μL (polar metabolites) or 5 μL (lipid metabolites). For polar metabolites, mobile phase A was water containing 0.1% formic acid, and mobile phase B was acetonitrile; for lipid metabolites, mobile phase A was water containing 2 mM ammonium acetate and 0.1% formic acid, and mobile phase B was acetonitrile / isopropanol (50:50, v / v) containing 2 mM ammonium acetate and 0.1% formic acid. Gradient elution programs were optimized according to the metabolite type. Mass spectrometry data were acquired in positive and negative ion modes with the following parameters: sheath flow rate 40 / 45 arb, spray voltage 3.5 / -3.5 kV, capillary temperature 350 °C, resolution 70,000, automatic gain control target 3e6, maximum injection time 100 ms, and scan range m / z 70–1000 (polar metabolites) or 80–1200 (lipid metabolites).

[0099] Data analysis was performed using Progenesis QI software, including data alignment, normalization, matching, and identification. Metabolytic pathway mapping was performed using the MetaboAnalyst platform, and metabolite identification was conducted using the Metlin, HMDB, ALEX123, and LIPID MAPS databases. Quantitative analysis was performed using MultiQuant software (version 3.0.3, AB SCIEX) in scheduled multiple reaction monitoring (MRM) mode, optimizing declustering voltage, collision energy, and product ions to obtain the best signal. Through these steps, comprehensive detection and analysis of metabolites and lipid molecules in plasma were achieved, providing reliable data support for the discovery of disease biomarkers and mechanistic studies.

[0100] Through the above data collection and processing, this invention constructs a comprehensive dataset of population samples containing clinical characteristics, gut microbiota species information, and metabolite information, providing a solid foundation for subsequent machine learning model construction and staged diagnosis of coronary heart disease.

[0101] The clinical characteristics obtained from the above group sample statistics are shown in Table 1.

[0102] Table 1 Baseline characteristics of patients with NCA, sCAD, and ACS

[0103]

[0104] Note: #mean ± SD; %n (%); Among them, the non-coronary artery disease group (NCA group, N=73), the stable coronary artery disease group (sCAD group, N=106), and the acute coronary syndrome group (ACS group, N=106) are represented in Table 1. %n (%) indicates the data distribution in each group. For example, 25 (34.2) means that N=73*34.2%=25 in the NCA group, and 34.2% of the NCA group are female, which is 25 people. TG = triglycerides; TC = total cholesterol; HDL-C = high-density lipoprotein cholesterol; LDL-C = low-density lipoprotein cholesterol; In the case of normal distribution, continuous variables are presented in the form of mean ± standard deviation (SD), or in the case of variables that are not normally distributed, they are presented as the median with interquartile range.

[0105] The gut microbiota species and metabolite information datasets of the obtained population samples are shown in Table 2. In Table 2, A represents the group without coronary artery disease (NCA group, N=73); B represents the group with stable coronary artery disease (sCAD group, N=106); and C represents the group with acute coronary syndrome (ACS group, N=106). Due to the large amount of data on gut microbiota species and metabolite information, not all data is presented here. The portion shown represents the data on the individuals included in each group among the final selected key features, which is based on the statistical analysis of the original data and represents the average value of each key feature in groups A, B, and C. In Table 2, the gut microbiota species information is the average value of the relative abundance information (%) of each species in groups A, B, and C; the unit for metabolite information is ng / ml.

[0106] Table 2

[0107]

[0108]

[0109] 2. Data preprocessing and statistical analysis

[0110] In the data preprocessing stage, the raw data of the population sample were first cleaned to remove missing and outlier values ​​to ensure data integrity and consistency. For clinical characteristic data, normalization was used to scale the data to the range of 0 to 1 to eliminate dimensional differences between different characteristics. For gut microbiota species and metabolite information, standardization was performed to conform to a normal distribution, facilitating subsequent statistical analysis and model construction. Standardization typically uses the Z-score method, converting the data to a distribution with a mean of 0 and a standard deviation of 1. Continuous variables are expressed as mean ± standard deviation when normally distributed, and as median (interquartile range) when not normally distributed. Wilcoxon rank-sum test was used for comparisons between groups, and χ² test was used for comparisons of categorical variables. Statistical analysis was performed using R software (version 4.4.2).

[0111] 3. Feature Selection

[0112] In the feature selection phase, firstly, correlation analysis is used to calculate the correlation between each feature and coronary artery disease status (e.g., NCA and ACS, NCA and sCAD, sCAD and ACS), screening out features significantly related to coronary artery disease diagnosis. Subsequently, the Boruta algorithm (based on the BorutaPy package in Python) is used to evaluate feature importance. The specific operation process is as follows: First, the dataset is divided into feature variables and target variables. Then, the Boruta algorithm is used to analyze the feature variables. The Boruta algorithm is an extension of the random forest method that evaluates the importance of each feature by constructing shadow features (i.e., randomly arranged copies of features). The algorithm compares the importance scores of the original features and shadow features in the random forest, using rigorous statistical tests (such as multiple iterations of permutation tests) to determine which features are significantly more important than the random level. Finally, the Boruta algorithm removes unimportant features, retaining those that significantly contribute to the model's predictive ability. These features exhibit significant discriminative power in distinguishing different CAD stages, especially metabolite and microbial features, which have unique advantages in capturing dynamic changes in the disease.

[0113] 4. Model Building

[0114] In the model building phase, the LightGBM algorithm was chosen as the primary machine learning algorithm. LightGBM is an efficient algorithm based on Gradient Boosting Decision Tree (GBDT), which has advantages in handling high-dimensional data and complex feature relationships. First, the overall sample dataset was divided into a discovery set (80%) and an independent validation set (20%) using stratified random sampling (random_state = 2024) at an 8:2 ratio. The independent validation set was not involved in any model training or hyperparameter tuning and was only used for final evaluation. Then, within the discovery set, a 10-fold cross-validation method was used for model training and hyperparameter optimization: the discovery set was divided into 10 subsets, with 9 subsets used as the training set and the remaining subset as the validation set, and this process was repeated 10 times. The average performance of the 10 validation results was used as an estimate of the model's performance, thereby enhancing the model's stability and generalization ability. During training, the GridSearchCV method was used to optimize the model's hyperparameters by exhaustively searching a predefined parameter grid to find the optimal combination of hyperparameters. In addition, the model integrates clinical laboratory data, targeted quantitative metabolite data, and relative abundance data of gut microbiota species to construct a multidimensional feature set to improve the model's predictive performance.

[0115] The LightGBM diagnostic model is an ensemble machine learning algorithm based on gradient boosting decision trees. Its core design goal is to efficiently process large-scale data and achieve high-accuracy classification or regression prediction. It does not directly output results through a single mathematical formula, but rather makes predictions through a combined decision-making mechanism of multiple decision trees.

[0116] Model Structure: Tree Integration and Decision Mechanism

[0117] The LightGBM model consists of multiple decision trees (called "weak learners") stacked together, and each tree generates predictions through the following steps:

[0118] Feature splitting strategy

[0119] The leafwise growth method is adopted: each time, the node with the largest gain among all the current leaf nodes is selected for splitting, instead of splitting level-wise, thus reducing the loss function faster.

[0120] Gain calculation formula: Gain=21[∑i∈ILhi+λ(∑i∈ILgi)2+∑i∈IRhi+λ(∑i∈IRgi)2-∑i∈Ihi+λ(∑i∈Igi)2]-γ

[0121] in:

[0122] gi: First-order gradient of sample i (negative gradient residual)

[0123] hi: The second-order gradient (Hessian matrix) of sample i.

[0124] λ, γ: Regularization parameters (to control overfitting).

[0125] The code for constructing the prediction model in this invention is as follows:

[0126] from sklearn.metrics import auc,roc_auc_score,roc_curve

[0127] import pandas as pd

[0128] import lightgbm as lgb

[0129] dtrain=lgb.Dataset(data=X_train, label=y_train, free_raw_data=False)

[0130] clf=lgb.train(params=lgb_params,train_set=dtrain)

[0131] y_train_pred=clf.predict(X_train)

[0132] fpr_train,tpr_train,_=roc_curve(y_train,y_train_pred)

[0133] train_auc=auc(fpr_train,tpr_train)y_test_pred=clf.predict(X_test)

[0134] fpr_test,tpr_test,_=roc_curve(y_test,y_test_pred)

[0135] test_auc=auc(fpr_test,tpr_test)

[0136] y_external_pred=clf.predict(exter_train)

[0137] fpr_external,tpr_external,_=roc_curve(y_external,y_external_pred)

[0138] external_auc=auc(fpr_external,tpr_external)

[0139] For the analysis method, please refer to:

[0140] https: / / github.com / JingXu0701 / CAD_MultiOmics / blob / main / fig7 / Figure7_ML.ipynb

[0141] Taking model (2) as an example of distinguishing between those without coronary heart disease and acute coronary syndrome, after inputting the key features of the subject, the LightGBM diagnostic model first discretizes continuous features (such as age and biochemical indicators) into histogram intervals (bins) using a histogram algorithm to reduce computational complexity; categorical features are processed directly. The model uses multiple decision trees for collaborative analysis: each tree allocates paths to samples according to preset splitting rules and adopts a leaf-wise growth strategy, prioritizing the splitting of nodes with the largest information gain, and gradually refining the decision path. The outputs of all trees are weighted and summed to generate the original predicted value, which is finally converted into a probability value between 0 and 1 by the Sigmoid function, representing the subject's likelihood of having ACS.

[0142] The output results directly guide the differentiation of disease types using probability values:

[0143] ACS determination: If the probability is ≥0.5 (e.g., output 0.92), then it is determined as ACS.

[0144] NCA determination: If the probability is <0.5 (e.g., output 0.3), it is determined to be NCA (such as other diseases or normal conditions other than ACS).

[0145] The differentiation criteria not only rely on probability thresholds, but can also be analyzed through SHAP values ​​to determine the contribution of key features:

[0146] If the model identifies typical ACS features, the SHAP plot will show the positive impact of these features on the probability;

[0147] Conversely, if NCA-related features are present, the SHAP value will make a negative contribution, reducing the ACS probability and thus supporting NCA classification.

[0148] In this invention, the key features ultimately obtained include the following personal information:

[0149] When model (1) distinguishes between patients without coronary artery disease and those with stable coronary artery disease, the personal information includes:

[0150] Clinical characteristics: hypertension, age, total cholesterol, LDL cholesterol;

[0151] Intestinal flora species information: Paraprevotella clara, Prevotella marseillensis, Phocaeicola plebeius, Phocaeicola massiliensis, Bacteroides nordii, Mediterraneibacter butyricigenes, Clostridium symbolum, Gemmiger formimiminis and Odoribacter splanchnicus;

[0152] Metabolite information: salicylic acid, phenylacetylglutamine, carnitine, 2-hydroxybutyric acid, 3-hydroxybutyric acid, phenylalanine, betaine;

[0153] When model (2) distinguishes between patients without coronary artery disease and acute coronary syndrome, the personal information includes:

[0154] Clinical characteristics: total cholesterol, age;

[0155] Gut microbiota species information: Alistipes onderdonkii, GGB3490 SGB4664, Pararaprevotellaclara, Phocaeicola plebeius, Desulfovibrio SGB5077, Clostridium sp.AF20_17LB, Lachnospiraceae bacterium AM48_27BH, Haemophilus parainfluenzae, Bilophilawadsworthia GGB9713 SGB15249, Dysosmobacter sp.NSJ_60, Faecalibacillus intestinalis, and Clostridium sp.AF27_2AA;

[0156] Metabolite information: 2-hydroxybutyric acid, salicylic acid, lysine;

[0157] When model (3) distinguishes between stable coronary artery disease and acute coronary syndrome, the personal information includes:

[0158] Intestinal flora species information: Bacteroides eggerthii, Lachnospira pectinoschiza, Pseudobutyrivibrio gallinarum, Blautia glucerasea, Megamonas funiformis, Prevotella stercorea, Prevotella SGB1687, GGB9758 SGB15368, Anaerostipes hadrus, Pyramidobacter pisolens, Romboutsia timonensis, Faecalibacillus intestinals;

[0159] Metabolite information: 3-hydroxybutyric acid, 2-hydroxybutyric acid, phenylacetylglutamine, succinic acid, indoleacrylic acid, tryptophan, salicylic acid, tyrosine.

[0160] For unknown samples, to determine the risk of ACS and sCAD, the personal information data corresponding to the distinction between stable coronary heart disease and acute coronary syndrome in the above model (3) is substituted into the model to obtain the disease risk prediction for the sample and output the result; for example: when using the model that distinguishes between ACS and sCAD, the above personal information of ACS and sCAD is input, and the output result is: when the probability of ACS is greater than or equal to the probability of sCAD, the output prediction result is "has risk of ACS"; when the probability of having ACS is less than the probability of sCAD, the output prediction result is "does not have risk of ACS".

[0161] ACS risk.

[0162] When differentiating the risk of coronary heart disease in an unknown sample, the three sets of personal information data corresponding to the three models mentioned above can be substituted into the three models respectively to predict the disease risk of the sample and output the results; the result is judged based on the highest probability value of the disease output by the model prediction result.

[0163] 5. Model Evaluation

[0164] During the model evaluation phase, the constructed model is validated using an independent validation set that has never participated in modeling, in order to assess its robustness and generalization ability. The model's performance in distinguishing CAD is comprehensively evaluated by calculating metrics such as AUC, sensitivity, specificity, precision, and recall.

[0165] The key features of different models and their average values ​​in each group are shown in Tables 3-5. Gut microbiota species information is presented as relative abundance data (%), and metabolite information is in ng / ml.

[0166] Table 3 shows the key feature data of model (1) for distinguishing between non-coronary artery disease and stable coronary artery disease. In model (1), there are 16 groups without coronary artery disease and 20 groups with stable coronary artery disease. Table 4 shows the key feature data of model (2) for distinguishing between non-coronary artery disease and acute coronary syndrome. In model (2), there are 14 groups without coronary artery disease and 22 groups with stable coronary artery disease. Table 5 shows the key feature data of model (3) for distinguishing between stable coronary artery disease and acute coronary syndrome. In model (3), there are 25 groups without coronary artery disease and 18 groups with stable coronary artery disease. The given portion represents the data of personal information included in each group of the final selected key features, which is based on the statistics of the original data.

[0167] Table 3

[0168]

[0169]

[0170] Table 4

[0171]

[0172]

[0173] Table 5

[0174] label SCAD(mean) ACS(mean) 2-Hydroxybutyrate 18917.2 21630.67 3-Hydroxybutyrate 12747.16 50514.67 Anaerostipes_hadrus 0.056835 0.055484 Bacteroides_eggerthii 1.724457 0.891812 Blautia_glucerasea 0.021962 0.008354 Faecalibacillus_intestinalis 0.039896 0.047952 GGB9758_SGB15368 0.35765 0.444956 indoleacrylic acid 30201.6 29630 Lachnospira_pectinoschiza 0.370108 1.389827 Megamonas_funiformis 1.230475 0.702148 phenylacetylglutamine 305.542 365.7171 Prevotella_SGB1687 0.003731 0.002759 Prevotella_stercorea 0.186883 0.36526 Pseudoflavonifractor_gallinarum 0.001863 0.001664 Pyramidobacter_piscolens 0.120413 0.002483 Romboutsia_timonensis 0.006558 0.010379 Salicylic acid 27386.19 4021.24 Succinate 734.188 1216.156 Tryptophan 17967.32 16910 Tyrosine 9752.96 9268.389

[0175] Figure 1 This paper presents the performance analysis and feature contribution analysis of a multi-omics-based stratified diagnostic model for coronary artery disease. The distribution of the area under the receiver operating characteristic (AUC) curve of the diagnostic model constructed using cross-validation and incorporating features at different levels (microbiological features, metabolic features, and traditional risk factors) on the test dataset reveals the trajectory of its improved discriminative efficacy as features are fused. Figure 1 (A) in the model distinguishes between patients with no coronary artery disease and those with stable coronary artery disease (1). Figure 1 (B) in the model is a model (2) that distinguishes between stable coronary artery disease and acute coronary syndrome. Figure 1 (C) in the model (3) distinguishes between those without coronary artery disease and those with stable coronary artery disease. The statistical significance (p-value) of all model comparisons was calculated using the Wilcoxon rank-sum test. Figure 1 As shown, cross-validation analysis indicates that diagnostic performance is limited when using only traditional clinical biomarkers as features. By further integrating metabolic and microbial features, the model constructed in this embodiment demonstrates a significant improvement in accuracy. On an external, independent test dataset for validation, the joint model constructed in this embodiment achieves an AUC of 0.91 when distinguishing between patients without coronary artery disease and those with acute coronary syndrome, significantly outperforming the performance of traditional biomarkers. Furthermore, the SHapley AdditiveexPlanations (SHAP) framework is used to interpret the model, visualizing the contribution of each feature to the model's prediction results, further validating the importance of key features. Figure 1 As shown, the SHAP value represents the contribution of each feature to the prediction of a specific instance. Positive values ​​indicate that the feature increases the predicted value, while negative values ​​indicate that the feature decreases the predicted value. The SHAP plot shows the degree of contribution of key features to the combined model. Features are ranked according to their importance based on the absolute value of their SHAP values, and the color and intensity of each feature indicate the direction and magnitude of its contribution to the model output.

[0176] This invention integrates clinical features, gut microbiota species, and metabolites to construct a multi-omics-based stratified diagnostic system for coronary artery disease (CAD). It involves the application of personal information in diagnosing subjects with or without coronary artery disease (NCA) or with CAA, where CAA includes stable coronary artery disease (sCAD) and acute coronary syndrome (ACS). This system improves diagnostic accuracy and provides stronger support for clinical diagnosis and treatment. Compared with existing technologies, the model of this invention significantly improves the AUC value in distinguishing between NCA and ACS, demonstrating better discriminative ability. Existing technologies struggle to differentiate between sCAD and ACS, while the model of this invention can distinguish between both, achieving accurate diagnosis and stratification of different types of CAA.

[0177] The beneficial effects of this invention are as follows:

[0178] (1) Multidimensional feature integration: By integrating clinical, microbial and metabolic features, a multidimensional feature set is constructed, which makes full use of the potential information provided by data from different dimensions and significantly improves the diagnostic performance of the model.

[0179] (2) Key feature discovery: Through the Boruta algorithm and SHAP framework, a series of new metabolites and microbial biomarkers were discovered, which can be used to accurately distinguish between non-coronary artery disease (NCA) and acute coronary syndrome (ACS), non-coronary artery disease (NCA) and stable coronary artery disease (sCAD), and stable coronary artery disease (sCAD) and acute coronary syndrome (ACS), thus achieving accurate diagnosis and stratification of coronary artery disease.

[0180] (3) High-performance model construction: The LightGBM algorithm and 10-fold cross-validation method were used to construct a high-performance diagnostic model. The AUC value reached 0.91 when distinguishing between NCA and ACS, which is significantly better than the performance of traditional biomarkers.

[0181] (4) Model interpretability and generalizability: The model is interpreted through the SHAP framework, which enhances the interpretability of the model. At the same time, the robustness and generalizability of the model are ensured by verifying it through independent test datasets.

[0182] In summary, this invention provides an innovative solution and reliable tool for the accurate diagnosis and stratification of coronary heart disease by integrating multidimensional features, optimizing feature selection, and constructing a high-performance machine learning model, and has significant clinical application value.

[0183] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0184] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0185] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0186] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0187] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. Use of personal information in a system for diagnosing a subject as free of coronary heart disease or as having coronary heart disease, wherein, The coronary heart disease includes stable coronary heart disease, acute coronary syndrome; (1) When distinguishing between no coronary heart disease and stable coronary heart disease, the personal information includes: Paraprevotella, Phocaeicola, Bacteroides, Mediterraneibacter, Clostridium, Gemmiger, Odoribacter, salicylic acid, phenylacetyl glutamine, carnitine, 2-hydroxybutyric acid, 3-hydroxybutyric acid, phenylalanine, betaine; (2) When distinguishing between no coronary heart disease and acute coronary syndrome, the personal information includes: Alistipes, Paraprevotella, Phocaeicola, Desulfovibrio, Clostridium, Clostridiales, Haemophilus, Bilophila, Odoribacter, Faecalibacterium, Clostridium, 2-hydroxybutyric acid, salicylic acid, lysine; (3) When distinguishing between stable coronary heart disease and acute coronary syndrome, the personal information includes: Bacteroides, Ruminococcus, Pseudoflavonifractor, Blautia, Megasphaera, Prevotella, Anaeroglobus, Pyramidobacter, Dorea, Faecalibacterium, 3-hydroxybutyric acid, 2-hydroxybutyric acid, phenylacetyl glutamine, succinic acid, indoleacrylic acid, tryptophan, salicylic acid, tyrosine.

2. The use of claim 1, wherein the personal information includes clinical feature information, gut microbiota species information, and metabolite information, and wherein: (1) When distinguishing between no coronary heart disease and stable coronary heart disease, the personal information includes: Clinical feature information: hypertension, age, total cholesterol, low-density lipoprotein cholesterol; Gut microbiota species information: Paraprevotella clara, Prevotella marseillensis, Phocaeicola plebeius, Phocaeicola massiliensis, Bacteroides nordii, Mediterraneibacter butyricigenes, Clostridium symbiosum, Gemmiger formicilis, and Odoribacter splanchnicus; Metabolite information: salicylic acid, phenylacetyl glutamine, carnitine, 2-hydroxybutyric acid, 3-hydroxybutyric acid, phenylalanine, betaine; (2) When distinguishing between no coronary heart disease and acute coronary syndrome, the personal information includes: Clinical feature information: total cholesterol, age; Gut microbiota species information: Alistipes onderdonkii, GGB3490 SGB4664, Paraprevotella clara, Phocaeicola plebeius, Desulfovibrio SGB5077, Clostridium sp. AF20_17LB, Lachnospiraceae bacterium AM48_27BH, Haemophilus parainfluenzae, Bilophila wadsworthia, Metabolite information: 2-hydroxybutyric acid, salicylic acid, lysine; (3) when distinguishing stable coronary heart disease from acute coronary syndrome, the personal information includes: Intestinal flora species information: Bacteroides eggerthii, Lachnospira pectinoschiza, Pseudoflavonifractor gallinarum, Blautia glucerasea, Megamonas funiformis, Prevotella stercorea, Prevotella SGB1687, GGB9758 SGB15368, Anaerostipes hadrus, Pyramidobacter piscolens, Romboutsia timonensis, Faecalibacillus intestinalis; Metabolite information: 3-hydroxybutyric acid, 2-hydroxybutyric acid, phenylacetylglutamine, succinic acid, indoleacrylic acid, tryptophan, salicylic acid, tyrosine.

3. The use according to claim 2, wherein, The SHAP values (impact on model output) of the personal information in order are: Salicylic acid, Paraprevotella clara, low-density lipoprotein cholesterol, phenylacetylglutamine, Prevotella marseillensis, carnitine, Phocaeicola plebeius, Phocaeicola massiliensis, hypertension, Bacteroides nordii, age, 2-hydroxybutyric acid, 3-hydroxybutyric acid, total cholesterol, phenylalanine, Mediterraneibacter butyricigenes, Clostridium symbolum, betaine, Gemmiger formicilis, Odoribacter splanchnicus; or 2-hydroxybutyric acid, salicylic acid, Alistipes onderdonkii, total cholesterol, lysine, GGB3490 SGB4664, Paraprevotella clara, age, Phocaeicola plebeius, Desulfovibrio SGB5077, Clostridium sp. AF20_17LB, Lachnospiraceae bacterium AM48_27BH, Haemophilus parainfluenzae, Bilophila wadsworthia, GGB9713 SGB15249, Dysosmobacter sp. NSJ_60, Faecalibacillus intestinalis, Clostridium sp. AF27_2AA; or 3-hydroxybutyric acid, 2-hydroxybutyric acid, phenylacetylglutamine, succinic acid, indoleacrylic acid, Bacteroides eggerthii, tryptophan, Lachnospira pectinoschiza, Pseudoflavonifractor gallinarum, Blautia glucerasea, Megamonas funiformis, salicylic acid, Prevotella stercorea, Prevotella SGB1687, GGB9758 SGB15368, tyrosine, Anaerostipes hadrus, Pyramidobacter piscolens, Romboutsia timonensis, Faecalibacillus intestinalis.

4. A method for constructing a diagnostic model for differentiating a subject as free of coronary heart disease or as having coronary heart disease, wherein, The construction method comprises: Data collection: collect clinical feature information, intestinal flora species information and metabolite information of a population sample to construct a multi-dimensional feature set; the population sample comprises subjects without coronary heart disease and patients with coronary heart disease; Feature selection: calculate the correlation between each feature and subjects without coronary heart disease and patients with coronary heart disease, perform feature selection through a Boruta algorithm, and screen out key features for diagnosing whether a subject has coronary heart disease or not, wherein the key features comprise the personal information of claim 1 or 2; Model construction: randomly divide the population sample into a discovery set and an independent validation set, input the discovery set into a model constructed based on a machine learning algorithm, perform model training by using a 10-fold cross-validation method, and obtain the diagnosis model; The coronary heart disease comprises stable coronary heart disease and acute coronary syndrome.

5. The construction method of claim 4, wherein, The machine learning algorithm comprises one or a combination of two or more of a random forest, a support vector machine and a neural network. Preferably, the machine learning algorithm is a LightGBM algorithm.

6. The construction method of claim 4, wherein, The construction method further comprises: Data preprocessing: After data collection, data preprocessing is performed to remove missing values and outliers. Normalization is used for clinical feature information, and standardization is used for gut microbiota species relative abundance and metabolite feature information.

7. The construction method of claim 4, wherein, The construction method further comprises: Model construction: During model training, GridSearchCV method is used to adjust model hyperparameters; Model evaluation: The diagnostic model is tested using an independent validation set, and the performance of the diagnostic model is evaluated using the AUROC index to obtain the model evaluation result; Preferably, the construction method further comprises: using the SHapley Additive exPlanations framework to interpret the diagnostic model and verify the importance of key features.

8. A diagnostic system for distinguishing between a subject free of coronary heart disease or having coronary heart disease, wherein, The diagnostic system comprises: A data input module for inputting the obtained personal information of the subject according to claim 1 or 2; A data analysis module connected to the data input module, which inputs the personal information of the subject into the diagnostic model constructed by the construction method of any one of claims 4-7 for analysis; A result output module connected to the data analysis module, which outputs the diagnosis result of whether the subject has coronary heart disease or not; The coronary heart disease includes stable coronary heart disease and acute coronary syndrome.

9. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, The processor executes the computer program to realize: based on the personal information of the subject, the diagnosis result of whether the subject has coronary heart disease or not is distinguished; The coronary heart disease includes stable coronary heart disease and acute coronary syndrome The personal information is the same as that in claim 1 or 2.

10. A computer readable storage medium storing computer program instructions, which when executed realize: based on the personal information of the subject, the diagnosis result of whether the subject has coronary heart disease or not is distinguished; wherein The coronary heart disease includes stable coronary heart disease and acute coronary syndrome The personal information is the same as that in claim 1 or 2.

Images