Cardiovascular metabolic risk factor spectrum identification model construction system, storage medium and kit

Abstract

The present invention relates to the field of biomedicine, and provides a recognition model construction system, storage medium and kit for cardiovascular metabolic risk factor spectrum. The recognition model construction system for cardiovascular metabolic risk factor spectrum includes an acquisition module, a feature selection module, a training module and an identification module. The prediction model of the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum obtained by the present invention has excellent sensitivity, specificity and accuracy of risk assessment ability, only requires a small amount of plasma as a test sample, has low cost, is easily accepted by subjects, is suitable for large-scale promotion of the method, and can also be used in the screening of cardiovascular metabolic risk factor spectrum drugs.

Description

Technical Field

[0001] The present invention relates to the field of biomedicine, and in particular to a recognition model building system, a storage medium and a kit for cardiovascular metabolic risk factor spectrum. Background Art

[0002] The spectrum of cardiovascular metabolic risk factors includes: obesity, hypertension, hyperglycemia and hyperlipidemia.

[0003] The cardiovascular metabolic risk factor spectrum is not only a key risk factor for cardiovascular disease, but has also been reported to be associated with an increased incidence of various cancers and neurodegenerative diseases. Multiple risk factors often co-occur in the cardiovascular metabolic risk factor spectrum. According to the classification criteria recommended by the Chinese Medical Association Diabetes Branch in 2004 (CDS 2004), the cardiovascular metabolic risk factor spectrum can be divided into low risk (0 risk factors), medium risk (1-2 risk factors), and high risk (3-4 risk factors) according to the number of cardiovascular metabolic risk factors (obesity, hypertension, hyperglycemia, and hyperlipidemia).

[0004] However, this traditional risk assessment method relies on the appearance of obvious clinical symptoms and involves complex procedures such as physical examination, questionnaire survey, and hematological examination. It is time-consuming and requires the full cooperation of patients. Currently, there is a lack of fast and convenient risk assessment tools for the spectrum of cardiovascular metabolic risk factors. Summary of the invention

[0005] The purpose of the present invention is to provide a recognition model construction system, storage medium and kit for cardiovascular metabolic risk factor spectrum.

[0006] In order to solve the above problems, the present invention provides a system for building a recognition model of cardiovascular metabolic risk factor spectrum, comprising:

[0007] A collection module is used to collect plasma samples of each subject of low-risk, medium-risk and high-risk types of vascular metabolic risk factor spectrum, and obtain plasma small molecule metabolic biomarkers of low-risk, medium-risk and high-risk types and their relative concentration values ​​from each plasma sample;

[0008] A feature selection module is used to perform machine learning based on the plasma small molecule metabolic biomarkers of low-risk, medium-risk and high-risk types and their relative concentration values, so as to screen out the plasma small molecule metabolic biomarkers with significant differences between the subjects of low-risk, medium-risk and high-risk types of cardiovascular metabolic risk factor spectrum as the plasma small molecule metabolic biomarkers after feature selection;

[0009] A training module, for training a prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of each subject of the low-risk, medium-risk, and high-risk types;

[0010] An identification module is used to obtain the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the subject to be tested, and input the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the subject to be tested into the prediction model to obtain a prediction result of whether the subject to be tested belongs to the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum;

[0011] The plasma small molecule metabolic biomarkers after feature selection include at least the following metabolite biomarkers in peripheral plasma: acetoacetic acid, dimethylglycine, creatinine, carotene, serine, homocysteine ​​sulfate lactone, glyceric acid, niacin, homocysteine, glucose, aminoethanesulfonic acid, uric acid, succinic acid, threonine, cysteine, pipecolic acid, aminoethylsulfinic acid and erythronic acid.

[0012] Furthermore, in the above-mentioned cardiovascular metabolic risk factor spectrum identification model construction system, the acquisition module is used to obtain a first subject group and a second subject group that do not overlap each other, wherein the first subject group and the second subject group both include subjects of low-risk, medium-risk, and high-risk types of non-overlapping vascular metabolic risk factor spectra; the acquisition module is used to collect first plasma samples from the first subject group, and obtain plasma small molecule metabolic biomarkers and their relative concentration values ​​from each first plasma sample as a first set; the acquisition module is used to collect second plasma samples from the second subject group, and obtain plasma small molecule metabolic biomarkers and their relative concentration values ​​from each second plasma sample as a test set.

[0013] Furthermore, in the above-mentioned cardiovascular metabolic risk factor spectrum identification model construction system, the feature selection module is used to use the Kruskal-Wallis rank sum test to perform feature selection on the plasma metabolite biomarkers on the first set, and set a significance threshold to obtain the plasma small molecule metabolic biomarkers after feature selection.

[0014] Furthermore, in the above-mentioned cardiovascular metabolic risk factor spectrum identification model construction system, the training module is used to train a candidate prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of each subject of the low-risk, medium-risk and high-risk types in the first set; based on the test set and the first set, a final prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum is obtained.

[0015] Furthermore, in the above-mentioned cardiovascular metabolic risk factor spectrum identification model construction system, the training module is used to classify the low-risk and medium-risk subjects in the first set into the third category of subjects to obtain the third category of subject group; classify the high-risk subjects in the first set into the fourth category of subject group to obtain the fourth category of subject group; the training module is used to train the first candidate prediction model of the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the third category of subject group and the fourth category of subject group; based on the performance of the prediction results of the first candidate prediction model on the test set, the third category of subject group and the fourth category of subject group, obtain the final high-risk prediction model of the cardiovascular metabolic risk factor spectrum. , i.e., the first prediction model; the training module is used to classify the low-risk subjects in the first set into the fifth category of subjects to obtain the fifth category of subject group; the medium-risk and high-risk subjects in the first set are classified into the sixth category of subjects to obtain the sixth category of subject group; the training module is used to train the second candidate prediction model of the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the fifth category of subject group and the sixth category of subject group; based on the performance of the prediction results of the second candidate prediction model of the cardiovascular metabolic risk factor spectrum on the test set, the third category of subject group and the fourth category of subject group, a final medium- and high-risk prediction model of the cardiovascular metabolic risk factor spectrum, i.e., the second prediction model, is obtained.

[0016] Furthermore, in the above-mentioned cardiovascular metabolic risk factor spectrum recognition model construction system, the training module is used to divide the third group of subjects and the fourth group of subjects into 5 non-overlapping data, and each time, one of the 5 non-overlapping data that has not been selected is used as the first internal validation set, and the remaining 4 data are used as the first training set each time; wherein the first internal validation set and the first training set both include non-overlapping third group of subjects and fourth group of subjects; the training module is used to cyclically train the high-risk prediction model of the cardiovascular metabolic risk factor spectrum for 5 rounds based on the first internal validation set and the first training set each time, and use the extreme gradient boosting algorithm to obtain the corresponding 5 first candidate prediction models that meet the requirements; wherein the extreme gradient boosting algorithm XGBoostd hyperparameters are set to a learning rate eta = 0.3, the minimum weight of all observations in the subset min_child_weight = 1, the maximum depth of the tree max_depth = 6, the minimum objective function reduction required for further branching on the leaf node of the tree gamma = 0, and when constructing each tree, the sampling rate of the sample subsample = 0.8, feature sampling rate colsample_bytree when building each tree = 1, feature sampling rate colsample_bylevel when building each layer = 1, feature sampling rate when building each leaf node; colsample_bynode = 1, L1 regularization weight alpha = 0, L2 regularization weight lambda = 1, maximum number of iterations nrounds = 200; the training module is used to input the test set into 5 first candidate prediction models that meet the performance requirements for prediction, so as to obtain the prediction result of the first candidate prediction model on the test set, wherein the prediction result of the first candidate prediction model on the test set is determined by voting of the 5 first candidate prediction models that meet the performance requirements, and the prediction result of the first candidate prediction model on the test set is obtained by the principle of minority obeys majority; if the performance value of the prediction result of the first candidate prediction model on the test set is higher than the preset performance threshold, and the performance difference between the prediction result on the test set and the prediction result on the first internal validation set is less than the preset difference threshold, then the 5 first candidate prediction models that meet the performance requirements are used as the final first prediction models.

[0017] Furthermore, in the above-mentioned cardiovascular metabolic risk factor spectrum model construction system, the training module is used to divide the fifth category subject group and the sixth category subject group into 5 non-overlapping data, and each time, one of the 5 non-overlapping data that has not been selected is used as the second internal validation set, and the remaining 4 data are used as the second training set each time; the second internal validation set and the second training set both include non-overlapping fifth category subjects and sixth category subjects; the training module is used to cyclically train the medium- and high-risk prediction model of the cardiovascular metabolic risk factor spectrum for 5 rounds based on the second internal validation set and the second training set each time, and use the extreme gradient boosting algorithm to obtain 5 corresponding second candidate prediction models that meet the requirements; wherein the extreme gradient boosting algorithm XGBoostd hyperparameters are set to a learning rate eta = 0.3, the minimum weight sum of all observations in the subset min_child_weight = 1, the maximum depth of the tree max_depth = 6, the minimum objective function reduction required for further branching on the leaf node of the tree gamma = 0, and when constructing each tree, the sample sampling rate subsample = 0.8, feature sampling rate colsample_bytree when constructing each tree = 1, feature sampling rate colsample_bylevel when constructing each layer = 1, feature sampling rate when constructing each leaf node; colsample_bynode = 1, L1 regularization weight alpha = 0, L2 regularization weight lambda = 1, maximum number of iterations nrounds = 200; the training module is used to input the test set into 5 second candidate prediction models that meet the performance requirements for prediction, so as to obtain the prediction result of the second candidate prediction model on the test set, wherein the prediction result of the first candidate prediction model on the test set is determined by voting of the 5 first candidate prediction models that meet the performance requirements, and the minority obeys the majority principle is adopted to obtain the prediction result of the second candidate prediction model on the test set; if the performance value of the prediction result of the second candidate prediction model on the test set is higher than the preset performance threshold, and the performance difference between the prediction result on the test set and the prediction result on the second internal validation set is less than the preset difference threshold, then the 5 second candidate prediction models that meet the performance requirements are used as the final second prediction models.

[0018] According to another aspect of the present invention, there is further provided a computer-readable storage medium having computer-executable instructions stored thereon, wherein when the computer-executable instructions are executed by a processor, the processor is caused to perform the following steps:

[0019] The acquisition module collects plasma samples of each subject of the low-risk, medium-risk, and high-risk types of the cardiovascular metabolic risk factor spectrum, and obtains the plasma small molecule metabolic biomarkers and their relative concentration values ​​of the low-risk, medium-risk, and high-risk types from each plasma sample; the feature selection module performs machine learning based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​of the low-risk, medium-risk, and high-risk types to screen out the plasma small molecule metabolic biomarkers that are significantly different between the subjects of the low-risk, medium-risk, and high-risk types of the cardiovascular metabolic risk factor spectrum as the plasma small molecule metabolic biomarkers after feature selection; the training module performs machine learning based on the low-risk, medium-risk, and high-risk types of plasma small molecule metabolic biomarkers and their relative concentration values. The feature-selected plasma small molecule metabolic biomarkers and their relative concentration values ​​of each subject of the high-risk type are used to train a prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum; the recognition module obtains the feature-selected plasma small molecule metabolic biomarkers and their relative concentration values ​​of the subject to be tested, and inputs the feature-selected plasma small molecule metabolic biomarkers and their relative concentration values ​​of the subject to be tested into the prediction model to obtain a prediction result of whether the subject to be tested belongs to the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum; the feature-selected plasma small molecule metabolic biomarkers at least include the following metabolite biomarkers in peripheral plasma: acetoacetic acid, dimethylglycine, creatinine, carotene, serine, homocysteine ​​sulfate lactone, glyceric acid, niacin, homocysteine, glucose, aminoethanesulfonic acid, uric acid, succinic acid, threonine, cysteine, pipecolic acid, aminoethylsulfinic acid and erythronic acid.

[0020] According to another aspect of the present invention, the present invention also provides a detection kit for identifying a spectrum of cardiovascular metabolic risk factors, wherein the metabolite biomarkers in the detection kit include at least the following metabolite biomarkers in peripheral plasma: acetoacetic acid, dimethylglycine, creatinine, carotene, serine, homocysteine ​​sulfate lactone, glyceric acid, niacin, homocysteine, glucose, aminoethanesulfonic acid, uric acid, succinic acid, threonine, cysteine, pipecolic acid, aminoethylsulfinic acid and erythronic acid.

[0021] Compared with the prior art, the present invention obtains a prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum through machine learning of plasma metabolite biomarkers, showing excellent sensitivity, specificity and accuracy in risk assessment. A highly efficient assessment capability for the risk of the cardiovascular metabolic risk factor spectrum is achieved. The present invention only requires a small amount of plasma as a test sample, which is simple and convenient compared to the traditional cumbersome evaluation process, has low cost, is easily accepted by the subjects, and is suitable for large-scale promotion of the method. And the high sensitivity and specificity enable it to be used in the screening of drugs for the cardiovascular metabolic risk factor spectrum. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The performance of the first prediction model in one embodiment of the present invention for identifying high risk of cardiovascular metabolic risk factor spectrum for 18 metabolic biomarker combinations on the first internal validation set and the test set;

[0023] Figure 2 The second prediction model in one embodiment of the present invention is used to identify the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum using a combination of 18 metabolic biomarkers on the second internal validation set and the test set. DETAILED DESCRIPTION

[0024] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0025] In a typical configuration of the present application, the terminal, the device of the service network and the trusted party all include one or more processors (CPU), input / output interfaces, network interfaces and memories.

[0026] Memory may include non-permanent storage in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0027] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include non-transitory media such as modulated data signals and carrier waves.

[0028] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the various embodiments of the present invention will be described in detail in combination with the embodiments below. The experimental methods without specific conditions in the embodiments are usually carried out under conventional conditions, such as those described in textbooks and experimental guides, or under the conditions recommended by the manufacturer. The supporting software is operated under the recommended conditions.

[0029] The present invention provides a recognition model construction system for cardiovascular metabolic risk factor spectrum, the system comprising the acquisition module, feature selection module, training module and recognition module described below. In one embodiment, the acquisition module is used to collect plasma samples of each subject of low-risk, medium-risk and high-risk types of vascular metabolic risk factor spectrum, and obtain plasma small molecule metabolic biomarkers of low-risk, medium-risk and high-risk types and their relative concentration values ​​from each plasma sample.

[0030] Here, metabolomics is an emerging branch of biology that emerged after genome-transcriptome-proteomics in the early 21st century. It aims to conduct a comprehensive analysis of the chemical processes of metabolites, small molecule substrates, intermediates and cellular metabolites, and occupies an important position in the framework of systems biology. Plasma is an ideal biological fluid for finding biomarkers in metabolomics. At present, plasma metabolomics has shown broad application prospects in many research fields including disease diagnosis, drug development, efficacy and toxicity evaluation. Patients with medium-risk and high-risk cardiovascular metabolic risk factors show different metabolic models from low-risk people due to the disorder of their body's metabolic state. Therefore, plasma metabolomics can become a useful resource for risk assessment of cardiovascular metabolic risk factors, which will effectively promote disease prevention and control.

[0031] The relative concentration value refers to the relative concentration value of a certain type of plasma small molecule metabolic biomarker in plasma, and its unit can be %.

[0032] Preferably, the acquisition module can be used to obtain a non-overlapping first subject group and a second subject group, wherein the first subject group and the second subject group both include subjects of low-risk, medium-risk and high-risk types of non-overlapping vascular metabolic risk factor spectra.

[0033] Here, the first subject group includes subjects of low risk, medium risk, and high risk types in the vascular metabolic risk factor spectrum; the second subject group includes subjects of low risk, medium risk, and high risk types in the vascular metabolic risk factor spectrum; the first subject group and the second subject group do not overlap with each other.

[0034] Preferably, the collection module can be used to collect first plasma samples from the first subject group, and obtain plasma small molecule metabolic biomarkers and their relative concentration values ​​from each first plasma sample as a first set.

[0035] Preferably, the collection module can be used to collect second plasma samples from the second subject group, and obtain plasma small molecule metabolic biomarkers and their relative concentration values ​​from each second plasma sample as a test set.

[0036] Specifically, the group of subjects may be subjects recruited from three centers from January to December 2020 with the approval of a hospital ethics committee.

[0037] The first set of subjects used for training and validation of the prediction model can be 1244 subjects with low risk of cardiovascular metabolic risk factor spectrum, 1336 subjects with medium risk of cardiovascular metabolic risk factor spectrum, and 1144 subjects with high risk of cardiovascular metabolic risk factor spectrum from Center 1 (first community) and Center 2 (second community); the training set can be subsequently split into non-overlapping training sets and internal validation sets. Among them, the 1244 subjects with low risk of cardiovascular metabolic risk factor spectrum, excluding the subjects with medium and high risks that need to be identified in the present invention, serve as controls without risk factors of cardiovascular metabolic risk factor spectrum.

[0038] The test set for the prediction model can be from 584 subjects with low risk of cardiovascular metabolic risk factor spectrum, 623 subjects with medium risk of cardiovascular metabolic risk factor spectrum, and 531 subjects with high risk of cardiovascular metabolic risk factor spectrum from center 3 (third community).

[0039] Plasma can be collected from subjects classified into low-risk, intermediate-risk, and high-risk types of the cardiovascular metabolic risk factor spectrum according to the 2004 Chinese Medical Association Diabetes Society Recommended Criteria (CDS 2004), and metabolite biomarkers can be detected in a nanoparticle-enhanced laser desorption / ionization mass spectrometer.

[0040] The inclusion criteria for the subjects can be formulated according to the criteria recommended by the Chinese Medical Association Diabetes Branch in 2004 (CDS2004). Subjects who meet three or more of the following four items are classified as high-risk subjects in the cardiovascular metabolic risk factor spectrum; subjects who meet one or two of the following items are classified as medium-risk subjects in the cardiovascular metabolic risk factor spectrum; subjects who do not meet any of the following items are classified as low-risk subjects in the cardiovascular metabolic risk factor spectrum.

[0041] ① Hyperglycemia: fasting blood sugar ≥ 6.1mmol / L (110mg / dl) and / or blood sugar ≥ 7.8mmol / L (140mg / dl) two hours after a meal, and / or those who have been diagnosed with diabetes and are receiving treatment;

[0042] ② Hypertension: systolic / diastolic blood pressure ≥ 140 / 90 mmHg, and / or confirmed as having hypertension and receiving treatment;

[0043] ③Dyslipidemia: fasting serum triglycerides ≥1.7mmol / L (150mg / dl), and / or fasting blood high-density lipoprotein <0.9mmol / L (35mg / dl) (male) or <1.0mmol / L (39mg / dl) (female).

[0044] ④ Overweight and / or obesity: BMI ≥ 25.0 kg / m2;

[0045] Preferably, the exclusion criteria within the inclusion criteria are subjects with acute and infectious clinical symptoms within three weeks before sampling, including but not limited to fever, headache, cough, sore throat, loss of smell, runny nose, abdominal pain and diarrhea.

[0046] Collecting plasma samples from the subject group and obtaining plasma small molecule metabolic biomarkers and their relative concentration values ​​from each plasma sample can be implemented through steps S121 to S128 described below.

[0047] In step S121, the collection module is used to treat the plasma sample of peripheral venous blood of the subject in the fasting state in the early morning with EDTA anticoagulation and store it at -80°C.

[0048] Step S122, preparation of instruments and reagents: preparing a nanoparticle enhanced laser desorption / ionization time-of-flight mass spectrometer (MALDI-TOF-MS) and related reagents.

[0049] Here, the rice particle enhanced laser desorption / ionization time-of-flight mass spectrometer can be a product of Bruker, Germany, for example.

[0050] In step S123, the collection module is used to dilute the plasma sample: the plasma sample is diluted with deionized water at a standard ratio to obtain a diluted plasma sample, and the plasma sample is diluted to ensure that the metabolite concentration is suitable for mass spectrometry analysis.

[0051] Specifically, 100 nL of plasma sample can be diluted 10 times with deionized water to obtain a diluted plasma sample.

[0052] In step S124, the acquisition module is used to prepare the inorganic nanoparticles into a nanoparticle matrix solution for enhancing the mass spectrometry signal.

[0053] Specifically, the matrix (inorganic nanoparticles) was prepared into a 1 mg / mL matrix solution with deionized water.

[0054] In step S125, the acquisition module is used to prepare the sample on the mass spectrometry target plate: the diluted plasma sample is spotted onto the mass spectrometry target plate and dried at room temperature.

[0055] Specifically, the sample preparation can be performed on a mass spectrometry target plate, and 500 nL of each diluted plasma sample is spotted and dried at room temperature.

[0056] In step S126, the acquisition module is used to prepare the matrix on the mass spectrometry target plate: spot the nanoparticle matrix solution on the mass spectrometry target plate, ensuring that the nanoparticle matrix solution evenly covers the plasma sample spot to obtain a sample.

[0057] Specifically, the matrix can be prepared on a mass spectrometry target plate, and 500 nL of each matrix solution can be spotted on each plasma sample spot and dried at room temperature.

[0058] In step S127, the acquisition module is used to collect data on the mixture in a nanoparticle enhanced laser desorption / ionization time-of-flight mass spectrometer: mass spectrometry data of metabolites in plasma are collected using MALDI-TOF-MS, and a metabolite peak map in the sample is obtained through delayed extraction and time-of-flight analysis.

[0059] Specifically, data acquisition was performed in a nanoparticle enhanced laser desorption / ionization time-of-flight mass spectrometer, and the acquired data were extracted in positive ion mode with delayed extraction, a repetition rate of 1000 Hz, an acceleration voltage of 20 kV, a delay time of 250 ns, and 2000 laser shots per analysis.

[0060] In step S128, the acquisition module is used to run the built-in data preprocessing pipeline based on the metabolite peak map in the sample under the recommended conditions of the nanoparticle enhanced laser desorption / ionization time-of-flight mass spectrometer supporting software, including: data resampling, spectral line smoothing, baseline correction, and spectral peak matching, so as to obtain the relative concentration values ​​of the full spectrum of 303 metabolic biomarkers.

[0061] Specifically, a Savitzky-Golay (SG) filter can be used for spectral smoothing; then an adaptive iterative algorithm is used to identify baseline points, a polynomial fit is performed on the baseline points identified in the metabolite peak graph, the fitted baseline is subtracted from the original spectrum, and the corrected spectrum is checked to ensure that the baseline is flat and has no negative values; then the peak detection parameters are set: signal-to-noise ratio threshold = 3, mass accuracy tolerance 50ppm, continuous wavelet transform (CWT) is used to identify possible metabolite marker peak positions, and further peak fitting is performed to obtain the corresponding peak position, peak height and peak area, and the metabolite marker levels are relatively quantified according to the peak height; finally, Min-Max normalization is performed to obtain the relative concentration values ​​of 303 markers.

[0062] The relative concentration value obtained here can be the peak intensity value after min-max normalization, in %, which is used to measure the relative concentration value of metabolites in plasma. Normalization ensures that all metabolites have the same dimension.

[0063] By comparing the HMDB (Human Metabolome Database), 303 plasma small molecule metabolic biomarkers can be identified.

[0064] In one embodiment, the feature selection module is used to perform machine learning based on the plasma small molecule metabolic biomarkers of low-risk, medium-risk, and high-risk types and their relative concentration values ​​to screen out plasma small molecule metabolic biomarkers with significant differences between subjects of low-risk, medium-risk, and high-risk types of cardiovascular metabolic risk factor spectrum as plasma small molecule metabolic biomarkers after feature selection;

[0065] Preferably, the feature selection module is used to perform feature selection on the plasma metabolite biomarkers on the first set using the Kruskal-Wallis rank sum test, and set a significance threshold to obtain plasma small molecule metabolic biomarkers after feature selection.

[0066] More preferably, the significance threshold may be set as a Benjamini-Hochberg corrected P value < 0.05.

[0067] Specifically, the Kruskal-Wallis rank sum test can be used on the R (4.3.1) software to perform feature selection on the 303 plasma small molecule metabolic biomarkers in the training set, and the Benjamini-Hochberg method is used for multiple comparisons to control the incidence of type I errors (false positives). Finally, 18 plasma small molecule metabolic biomarkers were screened and there were significant differences among subjects with low risk, medium risk, and high risk types of the cardiovascular metabolic risk factor spectrum (P<0.05).

[0068] R software is a language and environment for statistical computing and graphics. It supports cross-platform support for Windows Mac Linux, etc. It is a GNU project, similar to the S language and environment developed by John Chambers and his colleagues at Bell Labs (formerly AT&T, now Lucent Technologies). R can be thought of as a different implementation of S. There are some important differences, but much of the code written for S remains unchanged when running under R. R provides a wide variety of statistics (linear and nonlinear modeling, classical statistical tests, time series analysis, classification, clustering, etc.) and graphics techniques, and is highly extensible. The S language is often the tool of choice for research in statistical methods, and the R language provides an open source way to participate in that activity.

[0069] Preferably, the plasma small molecule metabolic biomarkers after feature selection include at least the following metabolite biomarkers in peripheral plasma:

[0070] Acetoacetic acid, Dimethylglycine, Creatinine, Malonic acid, L-Serine, Homocysteine ​​thiolactone, Glyceric acid, Nicotinic acid, Homocysteine, D-Glucose, Taurine, Uric acid, Succinic acid, L-Threonine, L-Cysteine, Pipecolic acid, Hypotaurine and Erythronic acid.

[0071] Among them, the normalized relative concentration value range of acetoacetic acid for low-risk type is 17.48±13.09, the normalized relative concentration value range of medium-risk type is 16.33±12.99, and the normalized relative concentration value range of high-risk type is 15.12±11.76. The unit of relative concentration value is %;

[0072] Dimethylglycine, the normalized relative concentration range of low-risk type is 25.58±15.20, the normalized relative concentration range of medium-risk type is 24.20±15.22, and the normalized relative concentration range of high-risk type is 23.24±14.47;

[0073] Creatinine, the normalized relative concentration range of low-risk type is 8.46±5.14, the normalized relative concentration range of medium-risk type is 8.15±5.64, and the normalized relative concentration range of high-risk type is 8.18±5;

[0074] Carotene, the normalized relative concentration range of low-risk type is 15.16±10.78, the normalized relative concentration range of medium-risk type is 14.76±11.07, and the normalized relative concentration range of high-risk type is 14.24±10.32;

[0075] Serine, the normalized relative concentration range of low-risk type was 16.42±12.66, the normalized relative concentration range of medium-risk type was 15.35±12.55, and the normalized relative concentration range of high-risk type was 14.32±11.20;

[0076] Homocysteine ​​sulfate lactone, the normalized relative concentration value range of low-risk type is 20.92±14.00, the normalized relative concentration value range of medium-risk type is 19.45±13.42, and the normalized relative concentration value range of high-risk type is 18.81±12.74;

[0077] Glyceric acid, the normalized relative concentration range of low-risk type was 16.15±12.77, the normalized relative concentration range of medium-risk type was 15.38±12.54, and the normalized relative concentration range of high-risk type was 14.82±12.01;

[0078] Niacin, the normalized relative concentration range of low-risk type was 19.57±12.96, the normalized relative concentration range of medium-risk type was 18.60±13.05, and the normalized relative concentration range of high-risk type was 18.13±12.68;

[0079] Homocysteine, the normalized relative concentration range of low-risk type was 7.80±5.76, the normalized relative concentration range of medium-risk type was 7.51±6.08, and the normalized relative concentration range of high-risk type was 7.21±6.08;

[0080] Glucose, the normalized relative concentration values ​​of low-risk type ranged from 15.26±9.09, the normalized relative concentration values ​​of medium-risk type ranged from 17.22±11.46, and the normalized relative concentration values ​​of high-risk type ranged from 21.47±14.01;

[0081] Aminoethanesulfonic acid, the normalized relative concentration range of low-risk type is 19.69±13.79, the normalized relative concentration range of medium-risk type is 18.75±14.04, and the normalized relative concentration range of high-risk type is 17.94±13.27;

[0082] Uric acid, the normalized relative concentration range of low-risk type was 14.14±10.69, the normalized relative concentration range of medium-risk type was 13.59±10.73, and the normalized relative concentration range of high-risk type was 13.92±11.16;

[0083] Succinic acid, the normalized relative concentration range of low-risk type was 16.09±11.56, the normalized relative concentration range of medium-risk type was 15.17±11.56, and the normalized relative concentration range of high-risk type was 14.62±11.14;

[0084] Threonine, the normalized relative concentration range of low-risk type was 18.82±10.81, the normalized relative concentration range of medium-risk type was 17.75±10.20, and the normalized relative concentration range of high-risk type was 17.82±10.23;

[0085] Cysteine, the normalized relative concentration range of low-risk type was 16.42±12.66, the normalized relative concentration range of medium-risk type was 15.35±12.55, and the normalized relative concentration range of high-risk type was 14.32±11.20;

[0086] Pipecolic acid, the normalized relative concentration range of low-risk type was 17.34±12.43, the normalized relative concentration range of medium-risk type was 16.85±12.78, and the normalized relative concentration range of high-risk type was 16.84±12.87;

[0087] Aminoethylsulfinic acid, the normalized relative concentration range of low-risk type is 17.04±13.30, the normalized relative concentration range of medium-risk type is 15.77±13.28, and the normalized relative concentration range of high-risk type is 14.53±11.85;

[0088] For erythronic acid, the normalized relative concentration values ​​of the low-risk type ranged from 18.42±12.16, the normalized relative concentration values ​​of the medium-risk type ranged from 18.10±12.60, and the normalized relative concentration values ​​of the high-risk type ranged from 17.73±12.42.

[0089] The specific information is shown in Tables 1 and 2.

[0090] Table 1: Information about 18 metabolic biomarkers

[0091]

[0092] Table 2: Mean range of concentrations of 18 metabolic biomarkers

[0093]

[0094] Here, Table 2 lists the range of normalized relative concentration values ​​(relative peak intensity) of low-risk, medium-risk, and high-risk types, in units of %, which are expressed here as mean plus or minus standard deviation. Based on the range of normalized relative concentration values ​​(relative peak intensity) of the low-risk type reference group as a normal reference, the recognition model construction system of the cardiovascular metabolic risk factor spectrum of the present invention can identify medium-high risk types or high risk types based on model training.

[0095] In the mass spectrum, the horizontal axis represents the mass-to-charge ratio (m / z) of the ion. The mass-to-charge ratio increases from left to right. For ions with a single charge, the value represented by the horizontal axis is the mass of the ion; the vertical axis represents the intensity of the ion flow, which is usually expressed in relative intensity, that is, the intensity of the strongest ion flow is set as 100%, and the intensities of other ion flows are expressed as percentages. Sometimes the total ion flow intensity of all recorded ions is taken as 100%, and various ions are expressed as percentages.

[0096] In the present application, the mean range of ion flow intensities corresponding to the low-risk, medium-risk, and high-risk types of each marker can be obtained, and then the mean ion flow intensities of the low-risk, medium-risk, and high-risk types are respectively normalized by min-max to obtain the range of relative concentration values ​​corresponding to the low-risk, medium-risk, and high-risk types, which are used as the input values ​​for model training of machine learning.

[0097] Here, the present invention adopts 18 plasma metabolic biomarkers for subsequent training of a predictive model for medium-risk or high-risk types of the cardiovascular metabolic risk factor spectrum; it can demonstrate excellent risk assessment capabilities; and by performing machine learning on plasma small molecule metabolic biomarkers after feature selection, it can achieve efficient assessment of the risk of the cardiovascular metabolic risk factor spectrum.

[0098] In one embodiment, the training module is used to train a prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of each subject of the low-risk, medium-risk, and high-risk types.

[0099] Preferably, the training module is used to train candidate prediction models for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of each subject of the low-risk, medium-risk and high-risk types in the first set; based on the test set, one of the candidate prediction models is selected as the final prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum.

[0100] More preferably, the training module is used to classify the low-risk and medium-risk subjects in the first set into a third category of subjects to obtain a third category of subject group; and classify the high-risk subjects in the first set into a fourth category of subject group to obtain a fourth category of subject group.

[0101] More preferably, the training module is used to train a first candidate prediction model for the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the third and fourth subject groups; based on the performance of the prediction results of the first candidate prediction model on the first internal validation set of the test set, a final high-risk prediction model for the cardiovascular metabolic risk factor spectrum, i.e., the first prediction model, is obtained.

[0102] Here, the first candidate prediction is a high-risk candidate prediction model.

[0103] In the first prediction model, the low-risk and medium-risk subjects in the first set are classified into one label, i.e., the third category of subjects; the high-risk subjects in the first set are classified into another label, i.e., the fourth category of subjects. Then, training based on these two labels can predict high-risk patients in the cardiovascular metabolic risk factor spectrum.

[0104] More preferably, the training module is used to classify the low-risk subjects in the first set into the fifth category of subjects to obtain the fifth category of subject group; and classify the medium-risk and high-risk subjects in the first set into the sixth category of subjects to obtain the sixth category of subject group.

[0105] More preferably, the training module is used to train a second candidate prediction model for the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the fifth and sixth subject groups; based on the performance of the prediction results of the second candidate prediction model in the cardiovascular metabolic risk factor spectrum on the second internal validation set of the test set, a final medium- and high-risk prediction model for the cardiovascular metabolic risk factor spectrum, i.e., the second prediction model, is obtained.

[0106] Here, different from the first prediction model, in the second prediction model, the low-risk subjects in the first set are classified into one label, i.e., the fifth category of subjects; the medium-risk and high-risk subjects in the first set are classified into another label, i.e., the sixth category of subjects, and then training based on these two labels can predict patients with medium and high risks in the cardiovascular metabolic risk factor spectrum;

[0107] The performance of the first prediction model of machine learning in assessing whether a subject is at high risk of cardiovascular metabolic risk factors was evaluated on the first internal validation set and test set of the third and fourth subject groups. The performance of AUC, sensitivity, specificity and accuracy were selected as model evaluation indicators. Among them, the model diagnostic performance curve overlapped in the first internal validation set and test, such as Figure 1 The specific test data is shown in Table 3.

[0108] Table 3. Performance of machine learning model 1 obtained on the internal validation set and test set of metabolic biomarkers

[0109]

[0110] The model prediction results are represented by a 2x2 confusion matrix, where rows represent actual labels and columns represent predicted labels. The confusion matrix divides samples into four categories:

[0111] TP (True Positive): The number of samples that are actually positive examples and predicted to be positive examples.

[0112] TN (True Negative) True Negative Examples: The number of samples that are actually negative examples but predicted to be negative examples.

[0113] FP (False Positive): False positive examples: The number of samples that are actually negative examples but predicted to be positive examples.

[0114] FN (False Negative) False negative examples: The number of samples that are actually positive examples but predicted to be negative examples.

[0115] Accuracy: The proportion of samples predicted correctly in the total samples, calculated as (TP+FP) / (FN+TN+TP+TN).

[0116] Sensitivity: The proportion of samples predicted to be positive among samples that are actually positive. The calculation formula is TP / (TP+FN).

[0117] Specificity: The proportion of samples predicted to be negative among samples that are actually negative. The calculation formula is TN / (TN+FP).

[0118] AUC: The area under the ROC curve. The ROC curve is a curve composed of the true positive rate and the false positive rate under different thresholds. The AUC is not affected by the ratio of positive and negative samples. It reflects the overall performance of the model under different thresholds.

[0119] The range is 0 to 1. The larger the value, the better the overall performance of the model. It can be calculated by the following formula:

[0120]

[0121] Among them, TPR=TP / (TP+FN), FPR=FP / (TN+FP).

[0122] The performance of machine learning model 2 in assessing whether a subject belongs to the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum was evaluated on the second internal validation set and test set of the fifth and sixth subject groups, respectively. The performance of AUC, sensitivity, specificity, and accuracy were selected as model evaluation indicators. Among them, the model diagnostic performance curve overlapped in the second internal validation set and test set, such as Figure 2 The specific test data is shown in Table 4.

[0123] Table 4. Performance of the second prediction model obtained by machine learning on the internal validation set and test set of metabolic biomarkers

[0124]

[0125] In a specific embodiment, the training module is used to divide the third group of subjects and the fourth group of subjects into five non-overlapping data, and each time use one of the five non-overlapping data that has not been selected as the first internal validation set, and each time use the remaining four data as the first training set; wherein the first internal validation set and the first training set both include non-overlapping third group of subjects and fourth group of subjects.

[0126] Specifically, the first internal validation set includes the third category of subjects and the fourth category of subjects; the first training set includes the third category of subjects and the fourth category of subjects; and the first internal validation set and the first training set do not overlap with each other.

[0127] In a specific embodiment, the training module is used to cyclically train the high-risk prediction model of the cardiovascular metabolic risk factor spectrum for 5 rounds based on the first internal validation set and the first training set each time, and use the extreme gradient boosting (XGBoost, eXtreme Gradient Boosting) algorithm to obtain the corresponding 5 first candidate prediction models that meet the requirements.

[0128] Here, the performance meeting the requirement may mean that the performance meets the requirement, that is, the performance value is relatively high, such as greater than a preset performance threshold. The first candidate prediction model is a high-risk candidate prediction model.

[0129] In a specific embodiment, the training module is used to input the test set into 5 first candidate prediction models whose performance meets the requirements for prediction, so as to obtain the prediction result of the first candidate prediction model on the test set, wherein the prediction result of the first candidate prediction model on the test set is decided by voting of the 5 first candidate prediction models whose performance meets the requirements, and the prediction result of the first candidate prediction model on the test set is obtained by adopting the principle of minority obeys majority; if the performance value of the prediction result of the first candidate prediction model on the test set is higher than a preset performance threshold, and the performance of the prediction result on the test set and the performance of the prediction result on the first internal validation set differ by less than a preset difference threshold, then the 5 first candidate prediction models whose performance meets the requirements are used as the final first prediction models.

[0130] In a specific embodiment, the training module is used to divide the fifth group of subjects and the sixth group of subjects into five non-overlapping data, and each time use one of the five non-overlapping data that has not been selected as the second internal validation set, and each time use the remaining four data as the second training set; the second internal validation set and the second training set both include non-overlapping fifth and sixth group of subjects.

[0131] Specifically, the second internal validation set includes the fifth category of subjects and the sixth category of subjects; the second training set includes the fifth category of subjects and the sixth category of subjects; and the second internal validation set and the second training set do not overlap with each other.

[0132] In a specific embodiment, the training module is used to cyclically train the medium- and high-risk prediction models of the cardiovascular metabolic risk factor spectrum for 5 rounds based on each second internal validation set and the second training set, using an extreme gradient boosting algorithm, to obtain corresponding 5 second candidate prediction models that meet the performance requirements.

[0133] Here, the second candidate prediction is a candidate prediction model with medium to high risk.

[0134] In a specific embodiment, the training module is used to input the test set into 5 second candidate prediction models whose performance meets the requirements for prediction, so as to obtain the prediction result of the second candidate prediction model on the test set, wherein the prediction result of the second candidate prediction model on the test set is decided by voting among the 5 second candidate prediction models whose performance meets the requirements, and the principle of minority obeys majority is adopted to obtain the prediction result of the second candidate prediction model on the test set; if the performance value of the prediction result of the second candidate prediction model on the test set is higher than a preset performance threshold, and the performance of the prediction result on the test set and the performance of the prediction result on the second internal validation set differ by less than a preset difference threshold, then the 5 second candidate prediction models whose performance meets the requirements are used as the final second prediction model module.

[0135] Here, if the performance meets the requirements, the value is relatively high, and the performance of the prediction results of the first prediction model or the second prediction model on the test set is close to the performance on the corresponding internal validation set, and there is no risk of overfitting, it indicates that the model can be generalized and the trained model meets the requirements.

[0136] The first set consisted of 3724 subjects, including: 1244 subjects with low risk of cardiovascular metabolic risk factor spectrum, 1336 subjects with medium risk of cardiovascular metabolic risk factor spectrum, and 1144 subjects with high risk of cardiovascular metabolic risk factor spectrum;

[0137] The test set consisted of 1738 subjects, including 584 subjects with low risk in the cardiovascular metabolic risk factor spectrum, 623 subjects with medium risk in the cardiovascular metabolic risk factor spectrum, and 531 subjects with high risk in the cardiovascular metabolic risk factor spectrum.

[0138] Divide the third group of subjects and the fourth group of subjects or the fifth group of subjects and the sixth group of subjects into five parts, take a different one of the parts as the internal validation set each time, and take the remaining four parts as the training sets, and perform five rounds of training to obtain five corresponding first or second candidate prediction models that meet the requirements;

[0139] You can use the extreme gradient boosting (XGBoost, eXtreme GradientBoosting) algorithm on the R (4.3.1) software to perform a five-fold cross-validation model training on the first set to obtain the performance of the model's prediction results on the internal validation set;

[0140] Preferably, the hyperparameters of the extreme gradient boosting algorithm XGBoost are set to learning rate eta = 0.3, minimum weight and min_child_weight of all observations in the subset = 1, maximum depth of the tree max_depth = 6, minimum objective function reduction required for further branching on the leaf nodes of the tree gamma = 0, sample sampling rate subsample = 0.8 when building each tree, feature sampling rate colsample_bytree = 1 when building each tree, feature sampling rate colsample_bylevel = 1 when building each layer, feature sampling rate when building each leaf node; colsample_bynode = 1, L1 regularization weight alpha = 0, L2 regularization weight lambda = 1, maximum number of iterations nrounds = 200; thereby obtaining 5 candidate prediction models that meet the requirements.

[0141] Preferably, the first set is divided into 5 parts: 4 parts are training sets and 1 part is an internal validation set; a different 1 part of the internal validation set is taken each time, and the remaining 4 parts are used as internal training sets; 5 rounds of cyclic training are performed to obtain 5 candidate prediction models that meet the requirements;

[0142] The hyperparameter settings of XGBoost,specific information is shown in Table 5. The hyperparameters of the first prediction model and the second prediction model are the same,,but the parameters are different.

[0143] Table 5: Hyperparameter settings for XGBoost Hyperparameters

[0144]

[0145] In the 5-round training, the maximum number of iterations in each round is nrounds = 200, and each round of training uses a corresponding internal validation set to verify the performance;

[0146] After 5 rounds of training, if the performance of the 5 models is good, these 5 models will be used as 5 candidate prediction models that meet the performance requirements; if some of the 5 models perform well and some perform poorly, 5 rounds of training will be repeated until 5 models with good performance are obtained.

[0147] Here, the inputs of the five models were the relative concentration values ​​of 18 metabolic biomarkers in the test set, and the outputs were low, intermediate, or high risk on the cardiometabolic risk factor profile.

[0148] The test set can be used to test the accuracy of the trained model.

[0149] Use the trained first prediction model or the second prediction model to make predictions on the test set to obtain the performance of the model's prediction results on the test set; if the performance meets the requirements, that is, the performance value is relatively high, and the performance of the prediction results of the first prediction model or the second prediction model on the test set is close to the performance on the corresponding internal validation set, and there is no risk of overfitting, it indicates that the model can be generalized and the trained model meets the requirements.

[0150] The performance may include: sensitivity, specificity, and accuracy.

[0151] In order to monitor people with higher levels of metabolic abnormalities, the first prediction model had a sensitivity of 89.03%, a specificity of 76.61%, an accuracy of 85.23%, and an AUC value of 0.934 in the first internal validation set; its sensitivity in the test set was 85.58%, specificity was 80.60%, accuracy was 84.06%, and AUC value was 0.925.

[0152] The second prediction model is to monitor people with metabolic abnormalities and distinguish them from normal people, that is, to monitor people with medium to high levels of metabolic abnormalities. When one of the medium risk or high risk cannot be determined, the first prediction model can be used to assist in judging whether it is indeed a high risk; in the second internal validation set, its sensitivity is 85.58%, specificity is 85.20%, accuracy is 85.33%, and AUC value is 0.917; its sensitivity in the test set is 89.72%, specificity is 83.36%, accuracy is 85.55%, and AUC value is 0.909. The model for risk assessment of cardiovascular metabolic risk factor spectrum provided by the present invention has the characteristics of low sample consumption and high reproducibility.

[0153] Here, the prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum is used for training in subsequent steps to assess whether the subject to be tested belongs to the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum.

[0154] In one embodiment, the identification module is used to obtain the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the person to be tested, and input the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the person to be tested into the prediction model to obtain a prediction result of whether the person to be tested belongs to the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum.

[0155] Specifically, the first prediction model can predict the high-risk subjects of the cardiovascular metabolic risk factor spectrum; the second prediction model can predict the medium-high risk subjects of the cardiovascular metabolic risk factor spectrum; if a subject to be tested is predicted as a high-risk patient and a medium-high-risk patient by the first prediction model and the second prediction model respectively, the prediction result of the patient is a high-risk patient;

[0156] The plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the subject to be tested can be obtained in a manner similar to that described above in conjunction with the acquisition module.

[0157] According to another aspect of the present invention, there is further provided a computer-readable storage medium having computer-executable instructions stored thereon, wherein when the computer-executable instructions are executed by a processor, the processor is caused to perform the following steps:

[0158] Step S1, the collection module collects plasma samples of each subject of the low-risk, medium-risk, and high-risk types of the vascular metabolic risk factor spectrum, and obtains plasma small molecule metabolic biomarkers and their relative concentration values ​​of the low-risk, medium-risk, and high-risk types from each plasma sample;

[0159] Step S2, the feature selection module performs machine learning based on the plasma small molecule metabolic biomarkers of low-risk, medium-risk, and high-risk types and their relative concentration values ​​to screen out plasma small molecule metabolic biomarkers with significant differences between subjects of low-risk, medium-risk, and high-risk types of cardiovascular metabolic risk factor spectrum as plasma small molecule metabolic biomarkers after feature selection;

[0160] Step S3, the training module trains a prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of each subject of the low-risk, medium-risk, and high-risk types;

[0161] Step S4, the recognition module obtains the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the subject to be tested, and inputs the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the subject to be tested into the prediction model to obtain a prediction result of whether the subject to be tested belongs to the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum;

[0162] The plasma small molecule metabolic biomarkers after feature selection include at least the following metabolite biomarkers in peripheral plasma: acetoacetic acid, dimethylglycine, creatinine, carotene, serine, homocysteine ​​sulfate lactone, glyceric acid, niacin, homocysteine, glucose, aminoethanesulfonic acid, uric acid, succinic acid, threonine, cysteine, pipecolic acid, aminoethylsulfinic acid and erythronic acid.

[0163] Among them, acetoacetic acid, the normalized relative concentration range of low-risk type is 17.48±13.09, the normalized relative concentration range of medium-risk type is 16.33±12.99, the normalized relative concentration range of high-risk type is 15.12±11.76, the relative concentration value unit is %; dimethylglycine, the normalized relative concentration range of low-risk type is 25.58±15.20, the normalized relative concentration range of medium-risk type is 24.20±15.22, the normalized relative concentration range of high-risk type is 23.24±14.47; creatinine, the normalized relative concentration range of low-risk type is 8.46±5.14, the normalized relative concentration range of medium-risk type is 8.15±5.64, the normalized relative concentration range of high-risk type is 8.18±5; carotene, The normalized relative concentration range of low-risk type was 15.16±10.78, the normalized relative concentration range of medium-risk type was 14.76±11.07, and the normalized relative concentration range of high-risk type was 14.24±10.32; serine, the normalized relative concentration range of low-risk type was 16.42±12.66, the normalized relative concentration range of medium-risk type was 15.35±12.55, and the normalized relative concentration range of high-risk type was 14.32±11.20; homocysteine ​​sulfate lactone, the normalized relative concentration range of low-risk type was 20.92±14.00, the normalized relative concentration range of medium-risk type was 19.45±13.42, and the normalized relative concentration range of high-risk type was 18.81±12.74; glyceric acid, The normalized relative concentration values ​​of low-risk type were 16.15±12.77, the normalized relative concentration values ​​of medium-risk type were 15.38±12.54, and the normalized relative concentration values ​​of high-risk type were 14.82±12.01; niacin, the normalized relative concentration values ​​of low-risk type were 19.57±12.96, the normalized relative concentration values ​​of medium-risk type were 18.60±13.05, and the normalized relative concentration values ​​of high-risk type were 18.13±12.68; homocysteine, the normalized relative concentration values ​​of low-risk type were 7.80±5.76, the normalized relative concentration values ​​of medium-risk type were 7.51±6.08, and the normalized relative concentration values ​​of high-risk type were 7.21±6.08; glucose, The normalized relative concentration values ​​of low-risk type ranged from 15.26±9.09, the normalized relative concentration values ​​of medium-risk type ranged from 17.22±11.46, and the normalized relative concentration values ​​of high-risk type ranged from 21.47±14.01; aminoethanesulfonic acid, the normalized relative concentration range of low-risk type is 19.69±13.79, the normalized relative concentration range of medium-risk type is 18.75±14.04, and the normalized relative concentration range of high-risk type is 17.94±13.27; uric acid, the normalized relative concentration range of low-risk type is 14.14±10.69, the normalized relative concentration range of medium-risk type is 13.59±10.73, and the normalized relative concentration range of high-risk type is 13.92±11.16; succinic acid, the normalized relative concentration range of low-risk type is 16.09±11.56, the normalized relative concentration range of medium-risk type is 15.17±11.56, and the normalized relative concentration range of high-risk type is 14.62±11.14; threonine, The normalized relative concentration values ​​of low-risk type were 18.82±10.81, the normalized relative concentration values ​​of medium-risk type were 17.75±10.20, and the normalized relative concentration values ​​of high-risk type were 17.82±10.23; the normalized relative concentration values ​​of cysteine, low-risk type were 16.42±12.66, medium-risk type were 15.35±12.55, and the normalized relative concentration values ​​of high-risk type were 14.32±11.20; pipecolic acid, The normalized relative concentration values ​​of low-risk type were 17.34±12.43, the normalized relative concentration values ​​of medium-risk type were 16.85±12.78, and the normalized relative concentration values ​​of high-risk type were 16.84±12.87; the normalized relative concentration values ​​of aminoethylsulfinic acid, low-risk type were 17.04±13.30, medium-risk type were 15.77±13.28, and high-risk type were 14.53±11.85; the normalized relative concentration values ​​of erythronic acid, low-risk type were 18.42±12.16, medium-risk type were 18.10±12.60, and high-risk type were 17.73±12.42.

[0164] According to another aspect of the present invention, a detection kit for identifying a cardiovascular metabolic risk factor spectrum is also provided, wherein the metabolite biomarkers in the detection kit include at least the following metabolite biomarkers in peripheral plasma: acetoacetic acid, dimethylglycine, creatinine, malonic acid, serine (L-Serine), homocysteine ​​thiolactone, glyceric acid, nicotinic acid, homocysteine, glucose (D-Glucose), taurine, uric acid, succinic acid, L-threonine, cysteine ​​(L-Cysteine), pipecolic acid, hypotaurine and erythronic acid.

[0165] Among them, the metabolite biomarkers in the detection kit include at least the following metabolite biomarkers in peripheral plasma: acetoacetate, the normalized relative concentration value range of the low-risk type is 17.48±13.09, the normalized relative concentration value range of the medium-risk type is 16.33±12.99, and the normalized relative concentration value range of the high-risk type is 15.12±11.76, and the relative concentration value unit is %; dimethylglycine, the normalized relative concentration value range of the low-risk type is 25.58±15.20, the normalized relative concentration value range of the medium-risk type is 24.20±15.22, and the normalized relative concentration value range of the high-risk type is 23.24±14.47; creatinine, The normalized relative concentration range of low-risk type was 8.46±5.14, the normalized relative concentration range of medium-risk type was 8.15±5.64, and the normalized relative concentration range of high-risk type was 8.18±5; carotene, the normalized relative concentration range of low-risk type was 15.16±10.78, the normalized relative concentration range of medium-risk type was 14.76±11.07, and the normalized relative concentration range of high-risk type was 14.24±10.32; serine, the normalized relative concentration range of low-risk type was 16.42±12.66, the normalized relative concentration range of medium-risk type was 15.35±12.55, and the normalized relative concentration range of high-risk type was 14.32±11.20; homocysteine ​​sulfate lactone, The normalized relative concentration values ​​of low-risk type were 20.92±14.00, the normalized relative concentration values ​​of medium-risk type were 19.45±13.42, and the normalized relative concentration values ​​of high-risk type were 18.81±12.74; the normalized relative concentration values ​​of glyceric acid, low-risk type were 16.15±12.77, medium-risk type were 15.38±12.54, and the normalized relative concentration values ​​of high-risk type were 14.82±12.01; the normalized relative concentration values ​​of niacin, low-risk type were 19.57±12.96, medium-risk type were 18.60±13.05, and the normalized relative concentration values ​​of high-risk type were 18.13±12.68; homocysteine, The normalized relative concentration values ​​of low-risk type were 7.80±5.76, the normalized relative concentration values ​​of medium-risk type were 7.51±6.08, and the normalized relative concentration values ​​of high-risk type were 7.21±6.08; for glucose, the normalized relative concentration values ​​of low-risk type were 15.26±9.09, and the normalized relative concentration values ​​of medium-risk type were 17.22±11.46, the normalized relative concentration range of high-risk type is 21.47±14.01; aminoethanesulfonic acid, the normalized relative concentration range of low-risk type is 19.69±13.79, the normalized relative concentration range of medium-risk type is 18.75±14.04, and the normalized relative concentration range of high-risk type is 17.94±13.27; uric acid, the normalized relative concentration range of low-risk type is 14.14±10.69, the normalized relative concentration range of medium-risk type is 13.59±10.73, and the normalized relative concentration range of high-risk type is 13.92±11.16; succinic acid, The normalized relative concentration values ​​of low-risk type were 16.09±11.56, the normalized relative concentration values ​​of medium-risk type were 15.17±11.56, and the normalized relative concentration values ​​of high-risk type were 14.62±11.14; the normalized relative concentration values ​​of threonine, low-risk type were 18.82±10.81, medium-risk type were 17.75±10.20, and the normalized relative concentration values ​​of high-risk type were 17.82±10.23; the normalized relative concentration values ​​of cysteine, low-risk type were 16.42±12.66, medium-risk type were 15.35±12.55, and the normalized relative concentration values ​​of high-risk type were 14.32±11.20; the normalized relative concentration values ​​of pipecolic acid, The normalized relative concentration values ​​of low-risk type were 17.34±12.43, the normalized relative concentration values ​​of medium-risk type were 16.85±12.78, and the normalized relative concentration values ​​of high-risk type were 16.84±12.87; the normalized relative concentration values ​​of aminoethylsulfinic acid, low-risk type were 17.04±13.30, medium-risk type were 15.77±13.28, and high-risk type were 14.53±11.85; the normalized relative concentration values ​​of erythronic acid, low-risk type were 18.42±12.16, medium-risk type were 18.10±12.60, and high-risk type were 17.73±12.42.

[0166] In summary, the present invention obtains a prediction model for medium-risk or high-risk types of cardiovascular metabolic risk factor spectrum through machine learning of plasma metabolite biomarkers, showing excellent sensitivity, specificity and accuracy in risk assessment. A highly efficient assessment capability for cardiovascular metabolic risk factor spectrum risk is achieved. The present invention only requires a small amount of plasma as a test sample, which is simple and convenient compared to the traditional cumbersome evaluation process, has low cost, is easily accepted by subjects, and is suitable for large-scale promotion of the method. And the high sensitivity and specificity enable it to be used in the screening of cardiovascular metabolic risk factor spectrum drugs.

[0167] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

[0168] It should be noted that the present invention can be implemented in software and / or a combination of software and hardware, for example, it can be implemented using an application-specific integrated circuit (ASIC), a general-purpose computer or any other similar hardware device. In one embodiment, the software program of the present invention can be executed by a processor to implement the steps or functions described above. Similarly, the software program of the present invention (including related data structures) can be stored in a computer-readable recording medium, for example, a RAM memory, a magnetic or optical drive or a floppy disk and the like. In addition, some steps or functions of the present invention can be implemented using hardware, for example, as a circuit that cooperates with a processor to perform various steps or functions.

[0169] In addition, a part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. The program instruction for calling the method of the present invention may be stored in a fixed or removable recording medium, and / or transmitted through a data stream in a broadcast or other signal-bearing medium, and / or stored in a working memory of a computer device that runs according to the program instruction. Here, according to an embodiment of the present invention, a device is included, the device including a memory for storing computer program instructions and a processor for executing program instructions, wherein, when the computer program instruction is executed by the processor, the device is triggered to run the method and / or technical solution based on the aforementioned multiple embodiments of the present invention.

[0170] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved. In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the device claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

Claims

1. A system for building a recognition model for cardiovascular metabolic risk factor spectrum, characterized in that: include: A collection module is used to collect plasma samples of each subject of low-risk, medium-risk and high-risk types of vascular metabolic risk factor spectrum, and obtain plasma small molecule metabolic biomarkers of low-risk, medium-risk and high-risk types and their relative concentration values ​​from each plasma sample; A feature selection module is used to perform machine learning based on the plasma small molecule metabolic biomarkers of low-risk, medium-risk and high-risk types and their relative concentration values, so as to screen out the plasma small molecule metabolic biomarkers with significant differences between the subjects of low-risk, medium-risk and high-risk types of cardiovascular metabolic risk factor spectrum as the plasma small molecule metabolic biomarkers after feature selection; A training module, for training a prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of each subject of the low-risk, medium-risk, and high-risk types; An identification module is used to obtain the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the subject to be tested, and input the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the subject to be tested into the prediction model to obtain a prediction result of whether the subject to be tested belongs to the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum; The plasma small molecule metabolic biomarkers after feature selection include at least the following metabolite biomarkers in peripheral plasma: Acetoacetate, dimethylglycine, creatinine, carotene, serine, homocysteine ​​sulfate lactone, glycerate, niacin, homocysteine, glucose, aminoethanesulfonic acid, uric acid, succinate, threonine, cysteine, pipecolic acid, aminoethylsulfinic acid and erythronic acid.

2. The cardiovascular metabolic risk factor spectrum recognition model construction system according to claim 1, characterized in that: The acquisition module is used to obtain a first subject group and a second subject group that do not overlap each other, wherein the first subject group and the second subject group both include subjects of low-risk, medium-risk, and high-risk types of non-overlapping vascular metabolic risk factor spectra; The collection module is used to collect first plasma samples from the first subject group, and obtain plasma small molecule metabolic biomarkers and their relative concentration values ​​from each first plasma sample as a first set; The acquisition module is used to collect second plasma samples from the second subject group, and obtain plasma small molecule metabolic biomarkers and their relative concentration values ​​from each second plasma sample as a test set.

3. The cardiovascular metabolic risk factor spectrum recognition model construction system according to claim 2, characterized in that: The feature selection module is used to perform feature selection on the plasma metabolite biomarkers on the first set using the Kruskal-Wallis rank sum test, and set a significance threshold to obtain plasma small molecule metabolic biomarkers after feature selection.

4. The cardiovascular metabolic risk factor spectrum recognition model construction system according to claim 3, characterized in that: The training module is used to train a candidate prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of each subject of the low-risk, medium-risk and high-risk types in the first set; based on the test set and the first set, a final prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum is obtained.

5. The cardiovascular metabolic risk factor spectrum recognition model construction system according to claim 4, characterized in that: The training module is used to classify the low-risk and medium-risk subjects in the first set into a third category of subjects to obtain a third category of subject group; classifying the high-risk subjects in the first set into a fourth subject group to obtain a fourth subject group; The training module is used to train a first candidate prediction model for the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the third and fourth subject groups; based on the performance of the prediction results of the first candidate prediction model on the test set, the third and fourth subject groups, a final high-risk prediction model for the cardiovascular metabolic risk factor spectrum, i.e., a first prediction model; The training module is used to classify the low-risk subjects in the first set as the fifth category of subjects to obtain the fifth category of subject group; and classify the medium-risk and high-risk subjects in the first set as the sixth category of subjects to obtain the sixth category of subject group; The training module is used to train a second candidate prediction model for the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the fifth and sixth subject groups; based on the performance of the prediction results of the second candidate prediction model for the cardiovascular metabolic risk factor spectrum on the test set, the third and fourth subject groups, a final prediction model for medium and high risks of the cardiovascular metabolic risk factor spectrum, i.e., the second prediction model, is obtained.

6. The cardiovascular metabolic risk factor spectrum recognition model construction system according to claim 5, characterized in that: The training module is used to divide the third group of subjects and the fourth group of subjects into 5 non-overlapping data, and each time use one of the 5 non-overlapping data that has not been selected as the first internal validation set, and each time use the remaining 4 data as the first training set; wherein the first internal validation set and the first training set both include non-overlapping third group of subjects and fourth group of subjects; The training module is used to cyclically train the high-risk prediction model of the cardiovascular metabolic risk factor spectrum for 5 rounds based on the first internal validation set and the first training set each time, and use the extreme gradient boosting algorithm to obtain 5 corresponding first candidate prediction models that meet the requirements; wherein the extreme gradient boosting algorithm XGBoostd hyperparameters are set to learning rate eta = 0.3, the minimum weight and min_child_weight of all observations in the subset = 1, the maximum depth of the tree max_depth = 6, the minimum objective function reduction required for further branching on the leaf node of the tree gamma = 0, when constructing each tree, the sample sampling rate subsample = 0.8, the feature sampling rate colsample_bytree = 1 when constructing each tree, the feature sampling rate colsample_bylevel = 1 when constructing each layer, and the feature sampling rate when constructing each leaf node; colsample_bynode = 1, L1 regularization weight alpha = 0, L2 regularization weight lambda = 1, and the maximum number of iterations nrounds = 200; The training module is used to input the test set into 5 first candidate prediction models whose performance meets the requirements for prediction, so as to obtain the prediction result of the first candidate prediction model on the test set, wherein the prediction result of the first candidate prediction model on the test set is decided by voting of the 5 first candidate prediction models whose performance meets the requirements, and the prediction result of the first candidate prediction model on the test set is obtained by adopting the principle of minority obeys majority; if the performance value of the prediction result of the first candidate prediction model on the test set is higher than a preset performance threshold, and the performance difference between the prediction result on the test set and the prediction result on the first internal validation set is less than a preset difference threshold, then the 5 first candidate prediction models whose performance meets the requirements are used as the final first prediction models.

7. The cardiovascular metabolic risk factor spectrum recognition model construction system according to claim 5, characterized in that: The training module is used to divide the fifth group of subjects and the sixth group of subjects into 5 non-overlapping data, and each time use one of the 5 non-overlapping data that has not been selected as the second internal validation set, and each time use the remaining 4 data as the second training set; the second internal validation set and the second training set both include non-overlapping fifth group of subjects and sixth group of subjects; The training module is used to cyclically train the medium- and high-risk prediction model of the cardiovascular metabolic risk factor spectrum for 5 rounds based on the second internal validation set and the second training set each time, and use the extreme gradient boosting algorithm to obtain 5 corresponding second candidate prediction models that meet the requirements; wherein the extreme gradient boosting algorithm XGBoostd hyperparameters are set to learning rate eta = 0.3, the minimum weight and min_child_weight of all observations in the subset = 1, the maximum depth of the tree max_depth = 6, the minimum objective function reduction required for further branching on the leaf node of the tree gamma = 0, when constructing each tree, the sample sampling rate subsample = 0.8, the feature sampling rate colsample_bytree = 1 when constructing each tree, the feature sampling rate colsample_bylevel = 1 when constructing each layer, and the feature sampling rate when constructing each leaf node; colsample_bynode = 1, L1 regularization weight alpha = 0, L2 regularization weight lambda = 1, and the maximum number of iterations nrounds = 200; The training module is used to input the test set into 5 second candidate prediction models whose performance meets the requirements for prediction, so as to obtain the prediction result of the second candidate prediction model on the test set, wherein the prediction result of the first candidate prediction model on the test set is decided by voting of the 5 first candidate prediction models whose performance meets the requirements, and the minority obeys the majority principle is adopted to obtain the prediction result of the second candidate prediction model on the test set; if the performance value of the prediction result of the second candidate prediction model on the test set is higher than the preset performance threshold, and the performance difference between the prediction result on the test set and the prediction result on the second internal validation set is less than the preset difference threshold, then the 5 second candidate prediction models whose performance meets the requirements are used as the final second prediction models.

8. A computer-readable storage medium having computer-executable instructions stored thereon, wherein: When the computer executable instructions are executed by a processor, the processor performs the following steps: The acquisition module collects plasma samples of each subject of low-risk, medium-risk, and high-risk types of the vascular metabolic risk factor spectrum, and obtains plasma small molecule metabolic biomarkers and their relative concentration values ​​of low-risk, medium-risk, and high-risk types from each plasma sample; The feature selection module performs machine learning based on the plasma small molecule metabolic biomarkers of low-risk, medium-risk and high-risk types and their relative concentration values ​​to screen out plasma small molecule metabolic biomarkers with significant differences between subjects of low-risk, medium-risk and high-risk types of cardiovascular metabolic risk factor spectrum as plasma small molecule metabolic biomarkers after feature selection; The training module trains a prediction model for the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum based on the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of each subject of the low-risk, medium-risk, and high-risk types; The recognition module obtains the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the subject to be tested, and inputs the plasma small molecule metabolic biomarkers and their relative concentration values ​​after feature selection of the subject to be tested into the prediction model to obtain a prediction result of whether the subject to be tested belongs to the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum; The plasma small molecule metabolic biomarkers after feature selection include at least the following metabolite biomarkers in peripheral plasma: Acetoacetate, dimethylglycine, creatinine, carotene, serine, homocysteine ​​sulfate lactone, glycerate, niacin, homocysteine, glucose, aminoethanesulfonic acid, uric acid, succinate, threonine, cysteine, pipecolic acid, aminoethylsulfinic acid and erythronic acid.

9. A detection kit for identifying cardiovascular metabolic risk factor spectrum, characterized in that: The metabolite biomarkers in the detection kit include at least the following metabolite biomarkers in peripheral plasma: Acetoacetate, dimethylglycine, creatinine, carotene, serine, homocysteine ​​sulfate lactone, glyceric acid, niacin, homocysteine, glucose, aminoethanesulfonic acid, uric acid, succinic acid, threonine, cysteine, pipecolic acid, aminoethylsulfinic acid and erythronic acid; wherein, the metabolic biomarkers of the subject to be tested and their relative concentration values ​​are input into the prediction model to obtain the prediction result of whether the subject to be tested belongs to the medium-risk or high-risk type of the cardiovascular metabolic risk factor spectrum.

Images