Detection of biomarkers for heart failure based on oxidative lipid derivatization and machine learning models

Abstract

The embodiment of the present invention discloses a method for detecting heart failure biomarkers based on oxidized lipid derivatization and machine learning models, comprising the steps of: S1, derivatizing a biological sample to obtain a derivatized product; S2, detecting the derivatized product to obtain detection peaks of derivatives of different oxidized lipids; S3, screening using a random forest algorithm model based on the detection peak data of the derivatives to obtain heart failure biomarkers; wherein step S3 specifically includes: S31, cleaning the detection peak data of the derivatives; S32, converting the data obtained by data cleaning; S33, standardizing the converted data; S34, constructing a random forest algorithm model and training it; S35, screening using the trained random forest algorithm model to obtain ten oxidized lipids as heart failure biomarkers; the detection method has high detection accuracy, high reliability, good detection linearity and low detection limit for biomarkers.

Description

Technical Field

[0001] The present invention belongs to the field of biological detection technology, and specifically relates to a detection method for heart failure detection biomarkers based on oxidized lipid derivatization and machine learning models. Background Art

[0002] Heart failure (HF) is the final stage of many forms of advanced cardiovascular disease. Biomarkers play a vital role in all aspects of clinical practice and patient management of HF, and their application provides a rapid, objective, and low-cost quantitative tool.

[0003] Oxidized lipids are a class of bioactive lipids that play key roles in regulating critical biological processes such as cell signaling, inflammatory responses, and apoptosis. In recent years, there has been growing interest in their involvement in cardiovascular diseases, neurodegenerative diseases, and cancer. Targeted screening of oxidized lipid biomarkers has become a valuable tool for monitoring disease progression. Therefore, accurate and effective identification and quantification of oxidized lipids are essential for advancing biomedical research. Oxidized lipids can participate in the development of heart failure by affecting processes such as cardiomyocyte injury and apoptosis, myocardial fibrosis, inflammatory responses, and vascular function.

[0004] The analysis of oxidized lipids has encountered significant challenges, mainly due to their low abundance in biological matrices and a wide dynamic range, thus requiring highly sensitive analytical methods with broad detection capabilities. Recent developments have shown that liquid chromatography-tandem mass spectrometry (LC-MS / MS) combined with chemical isotope labeling and optimized solid phase extraction have improved the analytical performance of oxidized lipids to some extent.

[0005] Patent document CN117393154A discloses biomarkers for analyzing heart failure and screening methods thereof, but effective methods for analyzing oxidized lipids as biomarkers are still to be developed, and there is still a need to develop simpler and more practical diagnostic markers and detection methods for heart failure. Summary of the invention

[0006] In view of this, some embodiments disclose a method for detecting heart failure biomarkers based on oxidative lipid derivatization and machine learning models, comprising the steps of:

[0007] S1, performing a derivatization treatment on the biological sample to obtain a derivatization product;

[0008] S2, detecting the derivatization product to obtain the detection peaks of derivatives of different oxidized lipids in the biological sample;

[0009] S3. Based on the detection peak data of derivatives of different oxidized lipids, the random forest algorithm model is used for screening to obtain target oxidized lipids as biomarkers for heart failure;

[0010] Among them, the method of using the random forest algorithm to screen the target oxidized lipids specifically includes:

[0011] S31, performing data cleaning on the detection peak data set of the derivative;

[0012] S32, performing data conversion on the data obtained by data cleaning to obtain detection peak data corresponding to detection peaks of different oxidized lipids under different derivatization conditions, and forming a dense matrix;

[0013] S33, standardizing the converted data to obtain secondary processed data that can be used in the random forest algorithm model; in the dense matrix formed by the secondary processed data, the detection peak corresponding to each feature is secondary processed;

[0014] S34. Build a random forest algorithm model and perform training; specifically include:

[0015] Perform regression classification on the secondary processed data;

[0016] Sampling the data obtained from regression classification and building a single decision tree;

[0017] Build a random forest algorithm model based on the obtained single decision tree;

[0018] Use secondary processed data to train the random forest algorithm model;

[0019] S35. Utilize the trained random forest algorithm model and secondary processed data for screening, and obtain target oxidized lipids as biomarkers for heart failure.

[0020] Some embodiments disclose a method for detecting heart failure biomarkers based on oxidative lipid derivatization and a machine learning model, wherein step S1 specifically comprises:

[0021] S11, extract biological samples;

[0022] S12, pre-treating the extracted biological sample to obtain an oxidized lipid sample in the biological sample;

[0023] S13, treating the oxidized lipid sample obtained by pretreatment with a derivatization reagent solution to perform a derivatization reaction to obtain a derivatized product of the oxidized lipid.

[0024] The embodiment of the present invention discloses a method for detecting heart failure biomarkers based on oxidized lipid derivatization and a machine learning model. The oxidized lipids in the biological sample are derivatized using a derivatization reagent solution, and the oxidized lipid derivatives are accurately detected using a liquid chromatography-mass spectrometry system. Ten highly sensitive oxidized lipids that can be used for simplified detection are screened out in combination with a random forest algorithm model, and ten oxidized lipids that can be used as heart failure biomarkers are further determined. The method for detecting heart failure biomarkers has high accuracy, high reliability, good detection linearity, and low detection limit for the detection of biomarkers, and is suitable for the reliable detection of heart failure biomarkers. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 , the effect of different pyridine concentrations on the efficiency of oxidized lipid derivatization;

[0026] Figure 2 , the effect of different concentrations of 3-NPH on the derivatization efficiency of oxidized lipids;

[0027] Figure 3 , the effect of different EDC concentrations on the efficiency of oxidation derivatization;

[0028] Figure 4 , the effect of different reaction times on the efficiency of oxidized lipid derivatization;

[0029] Figure 5 , the effect of different reaction temperatures on the efficiency of oxidized lipid derivatization;

[0030] Figure 6 , five Effects of different derivatization parameters on the properties of different oxidized lipid derivatives;

[0031] Figure 7 , compounds that showed stronger derivatization performance in both groups;

[0032] Figure 8 , comparison of compounds with strong derivatization properties;

[0033] Fig. 9 , the stability of repeated injections of various oxidized lipids;

[0034] Fig.10 , comparison of derivatization and extraction efficiency of various oxidized lipid compounds;

[0035] Fig.11 , the magnitude of matrix effects of different oxidized lipid compounds;

[0036] Fig.12 , the influence of matrix effect on the recovery rate of added standard;

[0037] Fig.13, the change in precision of different compounds as the standard concentration increases;

[0038] Fig.14 , RF-based machine learning modeling;

[0039] Fig.15 , machine learning modeling based on gain index XGBoost;

[0040] Fig.16 , X Construction of GBoost machine learning screening model;

[0041] Fig.17 , model performance evaluation and comparison of training and test sets of two machine learning methods;

[0042] Fig.18 , PLS-DA model validation diagram of oxidized lipid derivatization method in HF clinical analysis, n=60;

[0043] Fig.19 , PLS-DA score chart of oxidized lipid derivatization method in HF clinical analysis, n=60;

[0044] Fig. 20 , RSD of QC internal standard;

[0045] Fig.21 , the stability of 10 highly sensitive oxidized lipid biomarkers in QC;

[0046] Fig. 22 , relative peak areas of 15 HF-related differentially expressed metabolites with potential analytical value, n = 60;

[0047] Fig.23 , 15 ROC curves of potential oxidized lipids;

[0048] Fig.24 , HF clinical machine learning simulation results based on RF;

[0049] Fig.25 , establish HF clinical regression model based on XGB algorithm;

[0050] Fig.26 , HF clinical machine learning simulation results based on XGB gain algorithm;

[0051] Fig. 27 , model performance evaluation and comparison of two machine learning methods on HF clinical training and test sets;

[0052] Fig.28 , correlation analysis between differential metabolites and biochemical indicators. DETAILED DESCRIPTION

[0053] The word "embodiment" used herein as an "exemplary" does not necessarily mean that any embodiment described is superior or better than other embodiments. Performance index tests in the embodiments of the present invention are performed using conventional test methods in the art unless otherwise specified. It should be understood that the terms described in the embodiments of the present invention are only used to describe specific implementation methods and are not intended to limit the contents disclosed in the embodiments of the present invention.

[0054] Unless otherwise specified, the technical and scientific terms used herein have the same meanings as commonly understood by ordinary technicians in the technical field to which the embodiments of the present invention belong; other experimental methods and technical means not specifically specified in the embodiments of the present invention refer to experimental methods and technical means commonly used by ordinary technicians in the field.

[0055] The terms "substantially" and "approximately" used herein are used to describe small fluctuations. For example, they can refer to less than or equal to ±5%, such as less than or equal to ±2%, such as less than or equal to ±1%, such as less than or equal to ±0.5%, such as less than or equal to ±0.2%, such as less than or equal to ±0.1%, such as less than or equal to ±0.05%. The numerical data represented or presented in the range format herein are used only for convenience and brevity, and therefore should be flexibly interpreted as not only including the values ​​clearly listed as the limits of the range, but also including all independent values ​​or sub-ranges contained in the range. For example, the numerical range of "1-5%" should be interpreted as including not only the clearly listed values ​​of 1% to 5%, but also the independent values ​​and sub-ranges within the range shown. Therefore, independent values ​​such as 2%, 3.5% and 4% and sub-ranges such as 1%-3%, 2%-4% and 3%-5% are included in this numerical range. This principle also applies to the range of only one numerical value. In addition, such an explanation applies regardless of the width of the range or the characteristics.

[0056] In this document, including in the claims, transitional words such as "comprises," "includes," "with," "having," "containing," "involving," "accommodating," etc. are understood to be open-ended, i.e., meaning "including but not limited to." Only the transitional words "consisting of" and "composed of" are closed transitional words.

[0057] In order to better illustrate the present invention, numerous specific details are provided in the specific examples below. It should be understood by those skilled in the art that the present invention can be implemented without certain specific details. In the embodiments, some methods, means, instruments, equipment, etc. well known to those skilled in the art are not described in detail in order to highlight the main purpose of the present invention.

[0058] Under the premise of no conflict, the technical features disclosed in the embodiments of the present invention may be arbitrarily combined, and the resulting technical solutions belong to the contents disclosed in the embodiments of the present invention.

[0059] In some embodiments, the method for detecting a heart failure biomarker comprises the steps of:

[0060] S1. Derivatizing the biological sample to obtain a derivatized product; specifically comprising:

[0061] S11, extract biological samples;

[0062] S12, pre-processing the extracted biological sample to obtain an oxidized lipid sample in the biological sample; specifically comprising:

[0063] Oasis HLB column solid phase extraction column is activated with methanol and equilibrated with ultrapure water solution;

[0064] An internal standard solution was added to the biological sample; the internal standard solution included internal standards TXB2-D4, PGF2α-D4, RvD2-D5, LXA4-D5, 15-HEPE-D5, 17-HDHA-D5, 9-OxoODE-D3, and 12-HETE-D8, with a concentration of 250 ng / ml;

[0065] The biological sample containing the internal standard was loaded onto the Oasis HLB column solid phase extraction column, the column was eluted with water, and the oxidized lipids were eluted into an Eppendorf tube with a methanol solution containing 0.02% (V / V) formic acid;

[0066] The eluate was collected and blown dry with nitrogen;

[0067] S13, treating the oxidized lipid sample obtained by pretreatment with a derivatization reagent solution to perform a derivatization reaction to obtain a derivatized product of the oxidized lipid; wherein the derivatization reagent solution comprises: 3-nitrophenylhydrazine as a derivatization reagent, 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride as a cross-linking agent, and pyridine as an acyl transfer reagent, and in the derivatization reagent solution, the concentration of 3-nitrophenylhydrazine is 10 to 35 mg / mL, the concentration of 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride is 100 to 350 mg / mL, and the concentration of pyridine is 0.15 to 4.8 mg / mL; the derivatization reaction time is 10 to 60 min; and the derivatization reaction temperature is 10 to 60° C.;

[0068] The derivatization reaction process includes:

[0069] Adding a derivatization reagent solution to an oxidized lipid sample to obtain a derivatization reaction system; the derivatization system contains a derivatization reagent and oxidized lipids;

[0070] The derivatization reaction system is vortexed, and then the derivatization reaction system is transferred to a sampling bottle, and the oxidized lipid and the derivatization reagent are subjected to a derivatization reaction in the sample plate; the derivatization reagent includes a derivatization reagent, a cross-linking agent, and an acyl transfer reagent;

[0071] In some embodiments, 3-nitrophenylhydrazine (3-NPH) is used as a derivatization agent, 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride (EDC) is used as a cross-linking agent, and the carboxyl group is effectively connected to the primary amine of the oxidized lipid molecule through its unique reaction kinetics, and pyridine (BD) is used as an acyl transfer agent to ensure the rapid reaction of the O-acylurea intermediate with the amine group, forming a stable amide bond in the derivatization product of the oxidized lipid and releasing an isourea byproduct; for example, the reactions occurring during the derivatization reaction process include: the reaction between EDC and the carboxyl group to form an O-acylisourea intermediate; the O-acylisourea intermediate 1,3-rearranges to form an N-acylurea; the O-acylisourea intermediate reacts with the amine group to form an active amide with a stable amide bond, releasing an isourea byproduct; 3-NPH reacts with the active amide to form a carboxyl derivative;

[0072] Generally, in the derivatization process of oxidized lipids, EDC plays a key role as a zero-length crosslinker, effectively connecting the carboxyl group to the primary amine through its unique reaction kinetics; the reaction between EDC and the carboxyl group initially produces a key amine-reactive O-acylisourea intermediate, which is unstable in aqueous solution and may undergo 1,3-rearrangement to form N-acylurea. If it cannot react rapidly with the amine group, the byproduct cannot recombine with the primary amine group, but undergoes a hydrolysis reaction to reform the carboxyl group and possibly release N-substituted urea; in order to reduce the possibility of such side reactions, the nucleophilic reagent pyridine is introduced as an acyl transfer reagent; the addition of pyridine ensures the rapid reaction of the amine-reactive O-acylurea intermediate with the amine group, forming a stable amide bond and releasing the isourea byproduct; in order to further reduce the risk of possible side reactions to generate N-urea, the derivatization reaction conditions, such as temperature and reaction time, need to be strictly controlled to minimize the formation of byproducts and improve the efficiency and selectivity of the derivatization process;

[0073] S2. Detect the derivatization product to obtain the detection peaks of different oxidized lipid derivatives in the biological sample; usually, the derivatization efficiency of the oxidized lipid can be determined based on the detection peak, and the derivatization efficiency of the oxidized lipid under different conditions can be determined by the detection peak size of the oxidized lipid under different conditions; usually, the oxidized lipid derivative product is quantitatively detected using a UPLC-MS / MS system, and the obtained detection peak area represents the derivatization detection value; the detection method for the oxidized lipid derivative specifically includes: quantitatively detecting the oxidized lipid derivative product using a UPLC-MS / MS system; wherein the UPLC-MS / MS system is composed of a Waters ACQUITY UPLC system and a Waters Xevo TQ-S triple quadrupole mass spectrometer;

[0074] S3. Based on the detection peak data of derivatives of different oxidized lipids, the random forest algorithm model is used for screening to obtain the target oxidized lipids as biomarkers for heart failure. Usually, oxidized lipids are metabolites in the human body, involving multiple pathways, and the types of oxidized lipids produced by metabolism are rich and diverse. In order to ensure that the method disclosed in the embodiment of the present invention is universal for the detection of oxidized lipids, and at the same time can minimize the detection amount and improve the detection efficiency, 56 oxidized lipids that can represent the metabolites of the entire pathway are selected to study the detection method; among them, the 56 oxidized lipids include 6-keto-PGF1α, TXB3, RvE1, TXB2, PGF3α, 5-iPF2α-VI, PGF1α, PGF2α, LXA5, PGE2, LXB4, PGE1, PGD1, RVD2, LXA4, PGJ2, RVD1, 11-deoxy-PGE1, RVD5, LTB4, 5, 15-DiHETE, 9,10-DiHOME, 5,6-DiHETE, 9-HOTrE, 18-HEPE, 13-HOTrE, 19-HETE, 15-HEPE, 20-HE TE, 18-HETE, 12-HEPE, 17-HETE, 16-HETE, 5-HEPE, 17,18-EpETE, 13-OxoODE, 15-HETE, 17-HDHA, 15-OxoETE, 9-OxoODE, 14-HDHA, 11-HETE, 12-HETE, 7-HDHA, 9-HETE, 5-HETE, 12,13-EpOME, 14, 15-EET, 16,17-EpDPA, 9,10-EpOME, 4-HDHA, 5-OxoETE, 11,12-EET, 8,9-EET, 5,6-EET, 5-HETrE;

[0075] Usually, 56 oxidized lipid standards are used for methodological research, and 56 oxidized lipid standard solutions are prepared, and further subjected to derivatization reaction to obtain 56 oxidized lipid derivatives. The detection peak data obtained by LC-MS detection of the derivatives are used as the detection peak raw data of the derivatives and as the raw data set of the random forest algorithm model;

[0076] The peak data of derivatives of 56 different oxidized lipids were analyzed, and the derivatization conditions with the overall optimal derivatization efficiency for the 56 oxidized lipids were obtained, namely, pyridine concentration of 1.2%, 3-NPH concentration of 20 mg / mL, EDC concentration of 200 mg / mL, reaction time of 40 min, and reaction temperature of 40°C.

[0077] Based on the derivatization detection peak data of 56 different oxidized lipids, the selected 56 different oxidized lipids were screened, and 10 oxidized lipids with high sensitivity to the derivatization conditions were obtained; namely, 5-HETE, 9-HETE, 12-HETE, 16-HETE, 17-HETE, 20-HETE, 5,6-EET, 7-HDHA, 12-HEPE, and 14,15-EET. These 10 oxidized lipids are highly sensitive to the derivatization conditions and are greatly affected by the derivatization conditions.

[0078] Further, 60 clinical samples were collected, including 30 groups of healthy group samples and 30 groups of heart failure group samples, and 56 oxidized lipids in the clinical samples were detected, including 10 oxidized lipids that were highly sensitive to the derivatization conditions; the test results confirmed that the RSDs of the 56 oxidized lipids, including the 10 highly sensitive oxidized lipids, were all within the reliable range, that is, the oxidized lipid derivatization detection method disclosed in the above embodiment was reliable for the analysis of clinical samples; through the test results of the 10 highly sensitive oxidized lipids in the clinical samples, it can be inferred that the test results of the 56 oxidized lipids in the clinical samples can greatly simplify the process of comprehensive detection of the 56 oxidized lipids in the clinical samples, reduce the workload, and improve work efficiency;

[0079] Furthermore, based on the derivative detection peak data of 60 clinical samples, the derivatization detection peaks of 56 oxidized lipids were screened using the t-test and the screening condition of VIP value greater than 1, and 15 oxidized lipids with significant differences between healthy group samples and heart failure samples were obtained; these 15 oxidized lipids directly related to heart failure have the potential to be used as biomarkers for heart failure detection; the potential biomarkers for heart failure are 5-HETE, 9-HETE, 11-HETE, 12-HETE, 15-HETE, 16-HETE, 18-HETE, 20-HETE, PGD2, 5-iPF2α-VI, 9-OxoODE, R vD2, 4-HDHA, PGE1 and PGD1; among them, 5-HETE, 9-HETE, 11-HETE, 12-HETE, 15-HETE, 16-HETE, 18-HETE, 20-HETE, PGD2, 5-iPF2α-VI and 9-OxoODE were positively correlated biomarkers. Compared with the biological samples of the healthy group, the contents of these biomarkers in the biological samples of the heart failure group increased and were higher than the normal level; RvD2, 4-HDHA, PGE1 and PGD1 were negatively correlated biomarkers. Compared with the biological samples of the healthy group, the contents of these biomarkers in the biological samples of the heart failure group decreased and were lower than the normal level;

[0080] In some embodiments, the normal distribution of metabolomics data was assessed using the Shapiro-Wilk test and quantile-quantile plot. Independent sample t-test and ROC curve analysis were performed using SPSS (v26.0). P values ​​<0.05 and VIP values ​​>1.0 corrected by Bonferroni were set as thresholds for identifying differentially expressed metabolites. Histograms were drawn using GraphPad Prism 8.0 (GraphPad Software Inc., USA).

[0081] Furthermore, based on the derivative detection data of 60 clinical samples, the random forest algorithm model was used to optimize the selection of the selected 15 potential biomarkers, and 10 oxidized lipids were obtained as heart failure biomarkers with clinical practice guidance. The 10 oxidized lipids were 5-HETE, 11-HETE, 12-HETE, 16-HETE, 18-HETE, 5-iPF2α-VI, RvD2, PGE1, PGE2 and PGD1; among them, 5-HETE, 11-HETE, 12-HETE, 16-HETE, 18-HETE, PGD2, 5-iPF2α-VI were positively correlated biomarkers, and RvD2, PGE1 and PGD1 were negatively correlated biomarkers;

[0082] In some embodiments, the method of screening target oxidized lipids using the random forest algorithm specifically includes:

[0083] S31, using the detection peak data of the oxidized lipid derivative as the original data set, and performing data cleaning; wherein the detection peak data is the detection peak area value; the purpose of data cleaning is to remove data that may not belong to the detection peak area, so as to facilitate better data conversion;

[0084] In some embodiments, pandas.DataFrame. fillna in the scikit-learn library is used in python to perform data cleaning, determine the effective detection peak area data corresponding to the oxidized lipids and the derivatization conditions, and clean the above data for better data conversion;

[0085] S32. Perform data conversion on the data obtained from data cleaning, perform unique hot encoding, and convert the data classification vector into a data variable that can be processed by the random forest algorithm model; usually, data conversion refers to standardizing the obtained detection peak area mean to obtain a standardized peak area value as secondary processing data for random forest algorithm modeling and classification; usually, standardization includes calculating the peak area mean μ of all detection peaks, and then calculating the standard deviation σ of each peak area and the peak area mean; for each detection peak area data, the formula z=(x-μ) / σ is used for standardization; wherein x is the original data of the detection peak area, μ is the peak area mean, σ is the difference between the peak area of ​​the detection peak and the peak area mean, i.e., the standard deviation, and z is the detection peak area standardization result; usually, for the converted secondary processing data, the distribution of the diffraction peak area corresponding to each derivatization condition conforms to the normal distribution characteristics;

[0086] In some embodiments, pandas.get_dummies and sklearn.preprocessing.OneHotEncoder are used for one-hot encoding to obtain a dense matrix consisting of columns containing all oxidized lipid detection peak area information and rows corresponding to derivatization conditions; pandas.get_dummies and sklearn.preprocessing.OneHotEncoder are used to convert the oxidized lipid peak area values ​​corresponding to different BD concentrations, different 3NPH concentrations, different EDC concentrations, different reaction times, and different reaction temperatures into a data format that can be standardized and normalized, and the data format is generally represented in binary;

[0087] S33, normalizing the standardized peak area data; usually, after normalization, the data can be used to construct a random forest algorithm model;

[0088] In some embodiments, sklearn.preprocessing.StandardScaler is applied to perform data standardization, and sklearn.preprocessing.MinMaxScal is applied to perform data normalization to obtain unified data that can be used for model construction;

[0089] S34, build a random forest algorithm model and train it;

[0090] In some embodiments, the detection peak data of the derivatization reaction of 56 oxidized lipids are used as raw data to select oxidized lipids with high sensitivity to the derivatization conditions; wherein, constructing a random forest algorithm model and training it specifically includes:

[0091] The secondary processed data is regressed and classified, and the normalized peak area numerical data is used as the input data of the random forest algorithm model for training; for the classification task, the target variable used is the peak area string of different derivatization conditions, representing different categories; for the regression task, the target variable used is the peak area character value of the same derivatization condition to generate a regression task; 56 types of oxidized lipid peak area data involving the total sample data are constructed; then the regression classification data is sampled, and a single decision tree is established under each derivatization condition, and a model is learned based on the input oxidized lipid derivatization peak area data, which can classify or regress new data; then the random forest algorithm model is trained based on the single decision tree;

[0092] For example, sklearn.ensemble.RandomForestClassifier and sklearn.ensemble.RandomForestRegressor were used for classification and regression; make_regression was used to construct analysis data involving 56 oxidized lipids, 5 derivatization conditions, a gradient of 6 for each derivatization condition, and 3 repeated experiments for each derivatization condition; random_state was used to ensure the repeatability of the results;

[0093] Apply sklearn.utils.resample for Bootstrap sampling;

[0094] sklearn.tree.DecisionTreeClassifier and sklearn.tree.DecisionTreeRegressor were used to construct a single decision tree, and the overall random forest model was constructed by integrating the content of the single decision tree. Based on the derivative detection peak data under different derivatization conditions, the characteristics of oxidized lipids under different BD concentrations, different 3NPH concentrations, different EDC concentrations, different reaction times, and different reaction temperatures were classified and regressed. Each oxidized lipid after random sampling can be used to construct a decision tree to constitute the overall random forest model. Among them, sklearn.tree.DecisionTreeClassifier obtained the magnitude of the effect of derivatization conditions on the detection peak areas of 56 oxidized lipid derivatives, and sklearn.tree.DecisionTreeRegressor obtained the predicted value of the effect of derivatization conditions on the detection peak areas of 56 oxidized lipid derivatives. Each oxidized lipid after random sampling can be used to construct a decision tree, and the constructed decision tree further constitutes the overall random forest model.

[0095] In some embodiments, the detection peak data of the derivatization reaction of 15 potential oxidized lipids of 60 clinical samples are used as raw data to select oxidized lipids as biomarkers of heart failure; wherein, constructing a random forest algorithm model and training it specifically includes:

[0096] The secondary processed data were regressed and classified, and the normalized peak area numerical data were used as the input data of the random forest algorithm model for training; for the classification task, the target variable used was the peak area string of the sample under the optimal derivatization conditions, representing different categories; for the regression task, the target variable used was the peak area character value of the same clinical sample to generate the regression task; 15 potential oxidized lipid peak area data involving the total sample data were constructed; then the regression classification data were sampled, and a single decision tree was established under each clinical sample, and a model was learned based on the input oxidized lipid derivatization peak area data, which could classify or regress new data; then the random forest algorithm model was trained based on the single decision tree;

[0097] For example, sklearn.ensemble.RandomForestClassifier and sklearn.ensemble.RandomForestRegressor are used for classification and regression; analysis data involving 15 potential oxidized lipids are constructed according to make_regression; random_state is used to ensure the repeatability of the results;

[0098] Apply sklearn.utils.resample for Bootstrap sampling;

[0099] sklearn.tree.DecisionTreeClassifier and sklearn.tree.DecisionTreeRegressor were used to construct a single decision tree, and the overall random forest model was constructed by integrating the content of the single decision tree. Based on the peak detection data of 15 oxidized lipid derivatives of different clinical samples, the characteristics of oxidized lipids of different clinical samples were constructed, and the predicted values ​​of the peak detection areas of 15 oxidized lipid derivatives of different clinical samples were obtained. Each oxidized lipid after random sampling can be used to construct a decision tree, and the constructed decision tree further constitutes the overall random forest model.

[0100] S35, using the trained random forest algorithm model to make predictions;

[0101] In some embodiments, fit(X_train, y_train) is used to train the random forest algorithm model, and predict(X_test) and predict_proba(X_test) are used to perform predictions to obtain the final target oxidized lipids;

[0102] For example, the detection peak data of the derivatization reaction of 56 oxidized lipids at room temperature are used as the original data, and the oxidized lipids with high sensitivity to the derivatization conditions are selected. The input data format is the same as the input data format during model training, and the training set ratio is 0.7; predict(X_test) is used to classify or regress the test data and return the prediction results, where each row represents an oxidized lipid sample and each column represents a derivatization condition; for the classification task, the output is the predicted category label of each sample; for the regression task, the output is the predicted value of each sample; predict_proba(X_test) is used to classify and predict the test data and return the probability that each sample belongs to each category, where each row represents an oxidized lipid sample and each column represents a derivatization condition; the output data is a two-dimensional array, where each row corresponds to an oxidized lipid in X_test, each column corresponds to a derivatization condition category, and the value represents the probability that the sample belongs to the corresponding category; the target oxidized lipid is screened as a highly sensitive oxidized lipid according to the probability fluctuation, and the obtained highly sensitive oxidized lipid can be used as a biomarker for heart failure;

[0103] For example, the detection peak data of the derivatization reaction of 15 potential oxidized lipids in 60 clinical samples at room temperature are used as the original data, and the heart failure biomarkers are selected. The input data format is the same as the input data format during model training, and the training set ratio is 0.7; predict(X_test) is used to classify or regress the test data and return the prediction results, in which each row represents an oxidized lipid and each column represents a clinical sample; for the classification task, the output is the predicted category label of each clinical sample; for the regression task, the output is the predicted value of each clinical sample; predict_proba(X_test) is used to classify the test data and return the detection peak area of ​​each clinical sample belonging to each category; the output data is a two-dimensional array, in which each row corresponds to a clinical sample in X_test, each column corresponds to an oxidized lipid category, and the value represents the detection peak area of ​​the sample belonging to the corresponding category; heart failure biomarkers are screened according to the importance score.

[0104] In some embodiments, the evaluation of the constructed random forest algorithm model is also included; wherein,

[0105] Apply sklearn.metrics.r2_score to calculate the coefficient of determination R²:

[0106]

[0107] in, SS res is the residual sum of squares, that is, the sum of squares of the differences between the actual value and the predicted value, SS tot is the total sum of squares, that is, the sum of the squares of the differences between the actual values ​​and the mean.

[0108] Apply sklearn.metrics.mean_absolute_error to calculate the mean absolute error MAE:

[0109]

[0110] Where n is the number of samples, y i is the measured value, y i ^ is the predicted value.

[0111] Apply sklearn.metrics.mean_squared_error to calculate the mean square error MSE:

[0112]

[0113] Where n is the number of samples, y i is the measured value, y i ^is the predicted value.

[0114] Apply sklearn.metrics.root_mean_squared_error to calculate the root mean square error RMSE:

[0115]

[0116] Where n is the number of samples, y i is the measured value, ^ is the predicted value.

[0117] The technical details are further illustrated below in conjunction with embodiments.

[0118] Chemicals and reagents

[0119] LC-MS grade acetonitrile (ACN), formic acid (FA), isopropanol (IPA), and methanol were obtained from Fisher Scientific (Nidderau, Germany); oxidized lipids were purchased from Cayman Chemicals (Ann Arbor, MI, USA); 3-NPH was purchased from Sigma-Aldrich Laboratories, Inc. (St. Louis, MO); 1-ethyl-3(3-(dimethylamino)propyl)carbodiimide (EDC) and pyridine were purchased from CFW Laboratories, Inc. (Newark, Delaware, USA); high-throughput, fully automated solid phase extraction (SPE) instrumentation was purchased from Reeko (Raykol Group Corp., Ltd.);

[0120] Serum samples were collected from Liaocheng People's Hospital; male Sprague-Dawley rats (6–8 weeks old, 240 g) were purchased from Huafukang Biotechnology Co., Ltd. (Beijing, China).

[0121] Sample extraction and derivatization

[0122] Extraction of rat serum and HF patient serum samples

[0123] Serum samples were collected in blood collection tubes; each serum sample was allowed to stand at room temperature for 30–60 min and then centrifuged at 3000 g for 10 min at 4°C; after centrifugation, serum samples were collected and each serum sample was divided into 50 μL portions, and all serum samples were stored at −80°C before analysis.

[0124] Pretreatment of biological samples

[0125] The Oasis HLB column solid phase extraction column was activated with 2 ml of methanol and equilibrated with 2 ml of ultrapure water solution. 50 μL of biological sample was taken and 2 μL of 250 ng / ml internal standard solution was added. The internal standards included TXB2-D4, PGF2α-D4, RvD2-D5, LXA4-D5, 15-HEPE-D5, 17-HDHA-D5, 9-OxoODE-D3, and 12-HETE-D8. The biological sample was loaded onto the Oasis HLB column, and the column was washed with 2 ml of water. The target analytes were eluted into an Eppendorf tube with 1.5 μL of 0.02% formic acid in methanol; the eluate was collected and blown dry with nitrogen.

[0126] Preparation of derivatization solution

[0127] A derivatization solution was prepared by mixing an aqueous solution of EDC with a concentration of 200 mg / mL, a 50% acetonitrile solution of 3-NPH with a concentration of 20 mg / mL, and an acetonitrile solution of pyridine (BD) with a concentration of 1.2%; wherein the volume ratio of EDC to 3-NPH was 1:1, and the volume ratio of EDC, 3-NPH, and BD was 1:1:1;

[0128] A 50 μL aliquot of the derivatization solution was added to each blow-dried sample; after vortexing for 2 min, the samples were transferred to sampling vials and placed in a sample plate maintained at 40 °C; after 40 min of derivatization, the samples were centrifuged at 4 °C for 5 min, and 5 μL of the supernatant was taken for LC-MS analysis.

[0129] LC-MS / MS analysis conditions

[0130] Separation was performed using a CSH C18 analytical column (2.1 × 100 mm, 1.7 μm, Waters); this mobile phase consisted of ACN / water (40:60, v / v, containing 0.02% FA) as solvent A and ACN / IPA (50:50, v / v) as solvent B;

[0131] The binary gradient application method was: 0-5.5min, 100%A; 5.5-6.5min, 100-93%A; 6.5-9min, 93-82%A; 9-13.5min, 82-80.5%A; 13.5-18.5min, 80.5-75%A; 18.5-22.5min, 75-70%A; 22.5-22.6min, 70-55%A; 22.6-24.1min, 55-0%A; 24.1-26min, 0-100%A; the column temperature was maintained at 40°C, the flow rate was 0.4mL / min, and the injection volume was 5μL; oxidized lipid derivatives were detected in the negative eMRM;

[0132] The parameters of the mass spectrometer were set as follows: capillary voltage was 2 kV, desolvation temperature was 400 °C, ionization source temperature was 100 °C, desolvation flow rate was 800 L / h, and cone gas flow rate was 150 L / h. Data acquisition was performed using Masslynx 4.1 software and processed using TargetLynx (Waters, Milford, MA, USA).

[0133] LC-MS / MS analysis method

[0134] Calibration curves were constructed by plotting peak area ratios (analyte / internal standard) and the concentrations of oxidized lipids were calculated. Limits of detection (LODs) and limits of quantification (LOQs) were determined at signal-to-noise ratios (S / N) of 3 and 10, respectively. The accuracy and precision of the method were evaluated by recoveries and relative standard deviations (RSDs). The accuracy and precision of the method were evaluated by calculating recoveries and intra-day and inter-day relative standard deviations (RSDs).

[0135] Three different levels of oxidized lipid standards, 1 ng / mL, 10 ng / mL and 50 ng / mL, were added to rat serum samples to calculate the recovery rate and intra-day and inter-day precision. Each sample was measured three times. The intra-day precision was evaluated by repeating the injection analysis three times within one day, and the intra-day precision was evaluated by repeating the injection analysis for three consecutive days.

[0136] To investigate the extraction efficiency of oxidized lipids in biological samples, all oxidized lipid standards were spiked into rat serum before extraction and analyzed by UPLC−MS / MS in triplicate; matrix effects were assessed by comparing the peak areas of the standards spiked in matrix with those in neat solution.

[0137] In some embodiments, 56 standard solutions of oxidized lipids are prepared, each standard solution is derivatized under different derivatization conditions, and then the derivative products of the 56 selected oxidized lipids under different conditions are detected;

[0138] Different derivatization conditions included: pyridine concentration: 0.15, 0.3, 0.6, 1.2, 2.4, 4.8 mg / mL; 3-NPH concentration: 10, 15, 20, 25, 30, 35 mg / mL; EDC concentration: 100, 150, 200, 250, 300, 350 mg / mL; reaction time: 10, 20, 30, 40, 50, 60 min; reaction temperature: 10, 20, 30, 40, 50, 60 °C;

[0139] Generally, increasing the concentration of the derivatization reagent can increase the effective collision of the reactants and promote the transfer of equilibrium to the product, thereby effectively improving the derivatization efficiency. However, due to the existence of multiple reaction pathways, increasing the concentration of the derivatization reagent may lead to an increase in side reactions, thereby reducing the derivatization efficiency; in addition, increasing the concentration of the derivatization reagent may also lead to changes in solubility, so changing the solubility and other properties of the compound will affect the efficiency of derivatization; high concentrations of derivatization reagents may affect the stability of the reactants, which may also affect the derivatization efficiency; at the same time, reaction time and reaction temperature also have a great influence on the derivatization efficiency; therefore, the effects of different derivatization conditions on the derivatization efficiency are investigated;

[0140] Effect of Derivatization Conditions on the Efficiency of Oxidized Lipid Derivatization

[0141] Derivatization reactions were performed on 56 oxidized lipid standards under different derivatization conditions, and the derivatives were detected by LC-MS / MS; the detected peaks were analyzed and processed, and the results were as follows: Figure 1-Figure 5 shown by Figure 1 to Figure 5 From the data, we can see that when the pyridine concentration is 1.2% (v / v), the 3-NPH concentration is 20 mg / mL, the EDC concentration is 200 mg / mL, the reaction time is 40 min, and the reaction temperature is 40 °C, the overall derivatization efficiency of 56 oxidized lipids is high.

[0142] Analysis of the difference in derivatization efficiency of different types of oxidized lipids

[0143] In order to comprehensively study the effects of five derivatization parameters, namely pyridine concentration, 3-NPH concentration, EDC concentration, reaction temperature, and reaction time, on different types of oxidized lipids, radial plots were created based on the derivatization efficiency data of 56 oxidized lipids under different conditions, e.g. Figure 6 ; The effects of various derivatization parameters on different compounds were intuitively compared; the results showed that time significantly affected HDHAs, temperature had a greater impact on HETEs, pyridine and EDC concentrations significantly affected PGs, and 3-NPH concentration had a significant effect on oxo and HEPEs; based on the above analysis results, although the overall optimal derivatization parameters of different oxidized lipids can be determined, specific oxidized lipids have higher derivatization efficiency under specific conditions. Therefore, differential analysis of different oxidized lipid compounds was performed to improve the derivatization efficiency of individual compounds and improve the applicability of the method;

[0144] Further improvement of conditions, such as Figure 7 , Figure 8As shown, 9-HOTrE, 12-EpOME, 13-EpOME and 9, 10-EpOME showed the best derivatization efficiency at 30min and 50min. HETE compounds, such as 16-HETE, 11-HETE, 14-EET, 15-EET, and PGs, such as PGF2α and PGD1, showed the best derivatization efficiency at EDC concentrations of 250mg / mL and 300mg / mL; 13-OxoODE, 9-OxoODE, 11-HETE, 12-HETE and 9-HETE had the best derivatization efficiency at pyridine concentrations of 2.4mg / mL and 4.8mg / mL;

[0145] Further comparisons under relevant conditions such as Figure 8 As shown; the research results showed that oxidized lipid compounds such as 5-HETE, 11-HETE and 12-HETE had the best derivatization efficiency at 50°C, while 17, 18-EpETE, 19-HETE, 12-HEPE and LTB4 had the best derivatization efficiency when the 3-NPH concentration was 25 mg / mL.

[0146] In general, the differences in the derivatization performance of oxidized lipids may be related to their properties; taking HETEs as an example, HETEs exist in a variety of different isomers, and the differences in the molecular structures of these compounds affect their biological activities. The biosynthesis of HETE involves different enzyme pathways, such as the lipoxygenase pathway, the cyclooxygenase pathway, and the cytochrome P450 pathway. Different HETE isomers exert their biological effects by binding to different receptors; for example, 5-OxoODE acts through selective OXE receptors, while 12-HETE acts through GPR31 receptors; derivatization is a chemical structural modification used to enhance the detectability of target analytes or improve chromatographic performance. Therefore, differences in the structure and other properties of oxidized lipids will inevitably lead to specificity in the derivatization efficiency.

[0147] Linear analysis of derivatization efficiency test results

[0148] The results of the derivatization reaction of 56 oxidized lipid standards were analyzed and the limit of detection (LOD), limit of quantification (LOQ) and linear relationship were determined. Fig. 9 Indicates the stability of repeated injections of various oxidized lipid compounds. Fig.10 Comparison of derivatization and extraction efficiency of various oxidized lipid compounds. Fig.11 Indicates the size of matrix effect of different oxidized lipid compounds, Fig.12 It indicates the influence of matrix effect on the recovery rate of added standard. Fig.13 It indicates the change of precision of different compounds as the standard concentration increases;

[0149] The results showed that the LOD of 56 oxidized lipid standards ranged from 0.07 to 20 pg / mL, while the LOQ ranged from 0.16 to 40 pg / mL; the relative standard deviation (RSD) of the peak area of ​​six consecutive derivatization analyses was 0.49-6.68%; the extraction recovery, matrix effect and spike recovery of 1 ng, 10 ng and 50 ng oxidized lipid derivatization ranged from 80 to 120%, which met the methodological standards; the intra-day precision of oxidized lipid derivatization ranged from 0.29 to 14.71%, while the inter-day precision ranged from 0.22 to 14.76%;

[0150] The derivatization data of 56 oxidized lipids showed a high grayscale linear correlation coefficient (R²) ranging from 0.990 to 0.999, indicating that the oxidized lipid derivatization method disclosed in the embodiment of the present invention has good accuracy, reliability and repeatability; the comparison of the derivatization extraction efficiency of different types of oxidized lipids showed that OxoODE, TXs and HETEs had higher extraction efficiency, while HDHA and RVs had lower extraction efficiency; the results of the matrix effect study of 56 oxidized lipid compounds showed that the fluctuation range of the 56 compounds was within the optimal range.

[0151] Random Forest Algorithm Model

[0152] The original data set obtained from 56 oxidized lipids under different derivatization conditions was imported into the decision tree of the random forest algorithm model RF. Each tree independently generated predictions based on the training data. The RF algorithm aggregated these predictions through voting or averaging mechanisms to produce the final result. This method is particularly effective in processing complex data sets, effectively preventing overfitting, and providing satisfactory prediction performance. Throughout the training process, cross-validation techniques were used to evaluate model performance, and the model's prediction performance was optimized through parameter adjustment. In the model evaluation stage, indicators such as R square (R²), mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE) were used to evaluate the model's prediction accuracy.

[0153] Specifically, the data were imported into SIMCA-P software (version 14.1) for partial least squares discriminant analysis (PLS-DA) to identify differences between groups. To validate the PLS-DA model, a permutation test (n = 200) was performed. The normal distribution of metabolomics data was assessed using the Shapiro-Wilk test and quantile-quantile plots. Independent sample t-tests and ROC curve analysis were performed using SPSS (v26.0). Bonferroni-corrected P values ​​< 0.05 and VIP values ​​> 1.0 were set as thresholds for identifying differentially expressed metabolites. Histograms were plotted using GraphPad Prism 8.0 (GraphPad Software Inc., USA).

[0154] Gradient Boosted Tree Model

[0155] The gradient boosting tree model (XGB) is used to process the data of oxidized lipid derivatives, and the method of screening target oxidized lipids includes:

[0156] Import the peak area data of the oxidized lipid derivative detection peak into the XGBoost library; for example, in Python, import xgboost as xgb is used to import the XGBoost library, and the data is processed. The feature data obtained is a two-dimensional NumPy array or Pandas DataFrame, in which each row is an oxidized lipid sample, each column is a derivatization condition or clinical sample, and the label data is a one-dimensional NumPy array or Pandas Series, i.e., oxidized lipid sample, derivatization condition or clinical sample;

[0157] Data cleaning was performed on the peak area data of oxidized lipid derivatives. For example, X = data[features] and y = data[target] were used to divide the data into feature matrix (X) and target variable (y); the feature matrix included the LC-MS peak area data of 56 oxidized lipids in different clinical samples, and the target variable included the high or low expression of oxidized lipids in each clinical sample. From sklearn.model_selection import train_test_split and X_train, X_test, y_train, y_test = train_test_split were used to divide the data set into training set and test set, with the training set ratio being 0.7;

[0158] The detection peak data division of oxidized lipids affected by different derivatization conditions and the detection peak data division of differential metabolites in clinical samples; from sklearn.model_selection import train_test_split and X_train, X_test, y_train, y_test = train_test_split were used to divide the data set into training set and test set, where the training set ratio was 0.7;

[0159] Conversion of detection peak data of oxidized lipids affected by different derivatization conditions and detection peak data of differential metabolites in clinical samples; for example, use dtrain = xgb.DMatrix(X_train, label=y_train) and dtest = xgb.DMatrix(X_test, label=y_test) in DMatrix for data conversion; XGB model parameter setting; for example, use params = {'max_depth', 'eta', 'reg:squarederror'} for parameter setting to obtain secondary processing data that can be used for model training, where the data comes from the combination data of parameters and parameter values, and obtains the peak area of ​​oxidized lipids of different samples under different derivatization conditions and clinical samples for use in model training. Convert the peak area data of oxidized lipid derivative detection peaks into secondary processing data that can be used for simulation training according to the parameter settings, and the data format is a string;

[0160] Train the constructed XGB model; for example, apply num_round and bst = xgb.train (params, dtrain, num_round) to train the model to obtain the final screening variable, which is converted under dtrain = xgb.DMatrix(X, label=y), where num_round is used to train the model;

[0161] Among them, the secondary processed data of the detection peak data of 56 oxidized lipid standard derivatives at different BD concentrations, different 3NPH concentrations, different EDC concentrations, different reaction times and different reaction temperatures were used as model training data. The trained model can be used to screen highly sensitive oxidized lipids that are greatly affected by derivatization conditions; the secondary processed data of the detection peak data of 15 differential metabolites selected from 60 clinical samples were used as model training data. The obtained model can be used to screen biomarkers with diagnostic value for patients with heart failure.

[0162] Furthermore, it also includes XGB model evaluation, and the evaluation method is carried out with reference to the RF model evaluation method.

[0163] Specifically, the data were imported into SIMCA-P software (version 14.1) for partial least squares discriminant analysis (PLS-DA) to identify differences between groups. To validate the PLS-DA model, a permutation test was performed (n = 200).

[0164] Results and comparison of two algorithm models

[0165] The results show that the average Gini value of RF decreases, as Fig.14, indicating the importance of the features of compounds such as 16-HETE, 5,6-EET, 9-HETE and 7-HDHA in the RF model. In the XGB analysis, Fig.15 , Fig.16 The compounds 15-OxoODE, 11-deoxy-PGE1, 5-HEPE and 6-keto-PGF1α were identified as key feature indicators by total gain index analysis. The model performance indicators (R 2 , MAE, MSE, RMSE), such as Fig.17 ; The results show that the R 2 The results are significantly worse than those of the RF model, while the training set of the XGB model has worse results in MAE, MSE, and RMSE than the RF model, but little difference in other aspects;

[0166] Therefore, considering all factors, the RF model outperformed the XGB model in terms of performance for this particular analysis; the 10 key oxidized lipids 5-HETE, 9-HETE, 12-HETE, 16-HETE, 17-HETE, 20-HETE, 5,6-EET, 7-HDHA, 12-HEPE, and 14,15-EET were highly sensitive during the derivatization reaction.

[0167] In order to verify the feasibility of the detection method, it was applied to detect and analyze serum samples from heart failure (HF) patients and healthy individuals; 60 clinical samples were collected, including 30 healthy people and 30 HF patients, and derivatization reactions were performed using the 60 clinical samples. The derivative data of 10 key oxidized lipids with high sensitivity were detected to verify the stability of clinical sample analysis; and the detection data were analyzed;

[0168] Principal component analysis (PCA) showed that PC1 and PC2 accounted for 47.2% and 10.86% of the variance, respectively, indicating significant differences between the HF group and the healthy group, such as Fig.18 ;PLS-DA model shows that Q 2 and R 2 The values ​​meet the basic requirements for evaluating model fit and prediction performance, such as Fig.19 ; To verify the accuracy of data detection, the detection data were compared with the human metabolome database, confirming that they were within the theoretical range; Statistical analysis of internal standards (IS) in quality control samples (QC), such as Fig. 20 , showing a stable fluctuation range, with relative standard deviations (RSD) ranging from 2.11% to 5.92%, indicating the stability of the instrument and derivatization method;

[0169] In the QC samples, the RSD values ​​of 10 key highly sensitive oxidized lipids were always kept below 10%, such as Fig.21 This indicates that even under the influence of external factors such as changes in the derivatization environment and differences in human operation, oxidized lipids that are highly sensitive to derivatization conditions still maintain high stability as characteristic markers. This further confirms the good stability and reliability of the detection method and clinical sample analysis disclosed in the embodiments of the present invention.

[0170] Based on the test result data of 60 clinical samples, the important variables in projection (VIP) values ​​and t-test analysis were performed, such as Fig. 22 The variable importance in the HF group was analyzed, and a total of 15 significantly differential metabolites were obtained. Compared with the healthy group, the levels of 11 oxidized lipids in the HF group were significantly increased, including 5-HETE, 9-HETE, 11-HETE, 12-HETE, 15-HETE, 16-HETE, 18-HETE, 20-HETE, PGD2, 5-iPF2α-VI, and 9-OxoODE; on the contrary, the levels of RvD2, 4-HDHA, PGE1, and PGD1 were significantly decreased. These results suggest that these 15 significantly different oxidized lipid metabolites may play a role in the occurrence and development of HF.

[0171] Studies have confirmed that 9-OxoODE contributes to oxidative stress-related diseases in terms of mechanism and can be used as a biomarker of oxidative stress; HETE is an important arachidonic acid metabolite with certain proinflammatory effects; 18-HETE and 16-HETE have been identified as key lipid mediators under normal physiological and disease conditions; 15-HETE, 5-HETE, and 11-HETE can be used as proinflammatory mediators; PGE1 is a potent vasodilator that can reduce vasoconstriction and lower blood pressure; PGE1 alleviates Ang II through EP3 receptor activation and netrin-1 upregulation. II-induced cardiac hypertrophy; PGE2 inhibits the synthesis of TGF-β in cardiomyocytes and prevents HF through crosstalk between GRK1 and GRK2; 4-HDHA is a mediator of the anti-angiogenic effect of ωωH polyunsaturated fatty acids; 5-IPF2α-VI acid anti-angiogenic is positively correlated with reactive oxygen species (ROS) levels; RvD2 can prevent vascular remodeling, excessive contraction and endothelial dysfunction in obese hypertensive mice by regulating vascular and pro-inflammatory factors; therefore, the 15 differentially oxidized lipids selected and determined in the embodiment of the present invention as biomarkers for the diagnosis and analysis of HF disease have important prospects.

[0172] To verify the feasibility of the 15 different differential metabolites screened as potential diagnostic biomarkers for HF, ROC analysis was performed on these indicators.

[0173] The results show that the AUC values ​​of these indicators meet the requirements, and their effective peak areas are above the central axis area, such as Fig.23 , indicating its potential as a discriminant indicator.

[0174] RF analysis showed that compounds such as PGE1, PGD1, RvD2, and 16-HETE were important in the model, e.g. Fig.24 ; Similarly, XGB analysis proves a complete model, such as Fig.25 , 26 , the gain index showed that 4-HDHA, 5-iPV2α-VI, RvD2, PGE2 and other compounds were used as the main feature indicators for constructing the XGB model; based on the performance of the two machine learning methods, the R2, MAE, MSE and RMSE indicators of the training set and the test set were deeply analyzed, such as Fig. 27 ,The results show that the MAE, MSE and RMSE indicators of the XGB model are significantly higher ,compared with the RF model.

[0175] In conclusion, the results showed that the RF model had higher accuracy and stability in diagnosing and predicting HF, thus demonstrating its significant advantage and value.

[0176] Based on the RF model, the detection peak area data of 60 clinical samples of 15 selected potential oxidized lipids were used as the original data of the established RF model, and were screened again to further obtain 10 oxidized lipids as HF diagnostic markers; therefore, PGE1, PGD1, RvD2, 5-HETE, 11-HETE, 12-HETE, 16-HETE, 18-HETE, PGE2, and IPF2a screened under this model can be used as HF diagnostic biomarkers.

[0177] Based on these 10 HF diagnostic biomarkers, a comprehensive correlation analysis was performed with clinical biochemical indices, such as Fig.28 ; The results showed that most of the markers were significantly positively correlated with NT-proBNP and negatively correlated with LVEF. In addition, some of these markers were also significantly correlated with indicators such as total cholesterol (TC), triglycerides (TG) and serum creatinine (SCR), which fully demonstrated the close correlation between the markers selected in the embodiment of the present invention and key HF indicators, further strengthened the feasibility of using them as HF biomarkers, and verified the reliability of the detection method.

[0178] The detection method of heart failure biomarkers based on oxidized lipid derivatization and machine learning model disclosed in the embodiment of the present invention uses a derivatization reagent solution to derivatize oxidized lipids in biological samples, uses a liquid chromatography-mass spectrometry system to accurately detect oxidized lipid derivatives, and combines machine learning algorithms to screen out ten highly sensitive oxidized lipids that can be used to simplify detection. The stability of this method and these ten highly sensitive oxidized lipids is verified through a methodology investigation of oxidized lipid derivatization and oxidized lipid detection in clinical heart failure samples. Further, based on the qualified verification method and the stability of clinical samples, ten oxidized lipids are screened out as heart failure biomarkers in combination with a random forest algorithm model; the detection method of heart failure biomarkers has high detection accuracy, high reliability, good detection linearity, and low detection limit for biomarkers, and is a reliable detection method for heart failure biomarkers.

[0179] The technical solutions disclosed in the embodiments of the present invention and the technical details disclosed in the embodiments are merely illustrative of the inventive concept of the present invention and do not constitute a limitation on the technical solutions of the embodiments of the present invention. Any conventional changes, replacements or combinations of the technical details disclosed in the embodiments of the present invention have the same inventive concept as the present invention and are within the protection scope of the claims of the present invention.

Claims

1. A method for detecting heart failure biomarkers based on oxidative lipid derivatization and machine learning models, characterized in that: Includes steps: S1. Derivatizing the biological sample to obtain a derivatized product; specifically comprising: S11, extract biological samples; S12, pre-treating the extracted biological sample to obtain an oxidized lipid sample in the biological sample; S13, treating the oxidized lipid sample obtained by pretreatment with a derivatization reagent solution to perform a derivatization reaction to obtain a derivative of the oxidized lipid; wherein the derivatization reagent in the derivatization reagent solution includes: 3-nitrophenylhydrazine, 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride, and pyridine; in the derivatization reagent solution, the concentration of 3-nitrophenylhydrazine is 10 to 35 mg / mL, the concentration of 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride is 100 to 350 mg / mL, and the concentration of pyridine is 0.15 to 4.8 mg / mL; the derivatization reaction time is 10 to 60 min; and the derivatization reaction temperature is 10 to 60°C; S2. Detecting the derivatized products using LC-MS / MS method to obtain detection peaks of derivatives of different oxidized lipids in the biological sample; S3. Based on the detection peak data of derivatives of different oxidized lipids, the random forest algorithm model is used for screening to obtain target oxidized lipids as biomarkers for heart failure; Among them, the method of using the random forest algorithm to screen the target oxidized lipids specifically includes: S31, performing data cleaning on the detection peak data of the derivative; S32, performing data conversion on the data obtained by data cleaning to obtain detection peak data corresponding to detection peaks of different oxidized lipids under different derivatization conditions, and forming a dense matrix; S33, standardizing the converted data to obtain secondary processed data that can be used in the random forest algorithm model; in the dense matrix formed by the secondary processed data, the detection peak corresponding to each feature is secondary processed; S34. Build a random forest algorithm model and perform training; specifically include: Perform regression classification on the secondary processed data; Sampling the data obtained from regression classification and building a single decision tree; Build a random forest algorithm model based on the obtained single decision tree; Use secondary processed data to train the random forest algorithm model; S35. Utilize the trained random forest algorithm model and secondary processed data for screening, and obtain target oxidized lipids as biomarkers for heart failure.

2. The detection method according to claim 1, characterized in that: Step S2 includes using a UPLC-MS / MS system to quantitatively detect the derivatives of oxidized lipids to obtain a quantitative detection peak; wherein the UPLC-MS / MS system is composed of a Waters ACQUITYUPLC system and a Waters Xevo TQ-S triple quadrupole mass spectrometer; Chromatographic conditions included: A CSH C18 analytical column was used for separation; the mobile phase consisted of solvent A and solvent B, wherein solvent A consisted of ACN / water with a volume ratio of 40:60, containing 0.02% FA; solvent B consisted of ACN / IPA with a volume ratio of 50:50; the gradient was: 0-5.5 min, 100% A; 5.5-6.5 min, 100-93% A; 6.5-9 min, 93-82% A; 9-13.5 min, 82-80.5% A; 13.5-18.5 min, 80.5-75% A; 18.5-22.5 min, 75-70% A; 22.5-22.6 min, 70-55% A; 22.6-24.1 min, 55-0% A; the column temperature was maintained at 40 °C, the flow rate was 0.4 mL / min, and the injection volume was 5 μL; Mass spectrometry conditions included: Detection of oxidized lipid derivatives in negative mode MRM; The mass spectrometer parameter settings included: capillary voltage 2 kV, desolvation temperature 400 °C, ion source temperature 100 °C, desolvation gas N2, flow rate 800 L / h, cone gas flow rate 150 L / h; data acquisition was performed using Masslynx 4.1 software and processed using TargetLynx.

3. The detection method according to claim 1, characterized in that: In step S3, the detection peak data is the detection peak area of ​​the derivatives of different oxidized lipids in the biological sample under different derivatization conditions.

4. The detection method according to claim 1, characterized in that: In step S3, the derivatization product detection peak data of 56 oxidized lipids are used as original data, and 10 oxidized lipids with high sensitivity to derivatization conditions are selected, namely 5-HETE, 9-HETE, 12-HETE, 16-HETE, 17-HETE, 20-HETE, 5,6-EET, 7-HDHA, 12-HEPE, and 14,15-EET.

5. The detection method according to claim 1, characterized in that: In step S3, the peak data of 56 oxidized lipid derivatives from 60 clinical samples were detected and screened using the t-test and the screening condition of VIP value greater than 1 to obtain 15 potential biomarkers, namely 5-HETE, 9-HETE, 11-HETE, 12-HETE, 15-HETE, 16-HETE, 18-HETE, 20-HETE, PGD2, 5-iPF2α-VI, 9-OxoODE, RvD2, 4-HDHA, PGE1 and PGD1.

6. The detection method according to claim 5, characterized in that: Based on the detection peak data of oxidized lipid derivatives of 60 clinical samples, 10 potential biomarkers were screened out from 15 potential biomarkers using the random forest algorithm model as heart failure biomarkers, namely 5-HETE, 11-HETE, 12-HETE, 16-HETE, 18-HETE, PGD1, RvD2, PGE1, PGE2 and 5-IPF2α-VI.

7. The detection method according to claim 1, characterized in that: Biological samples include serum samples.

8. The detection method according to claim 1, characterized in that: Step S12 specifically includes: Oasis HLB column solid phase extraction column is activated with methanol and equilibrated with ultrapure water solution; An internal standard solution is added to the biological sample; the internal standard solution includes internal standards TXB2-D4, PGF2α-D4, RvD2-D5, LXA4-D5, 15-HEPE-D5, 17-HDHA-D5, 9-OxoODE-D3, and 12-HETE-D8; the concentration of the internal standard solution is 250 ng / ml; The biological sample containing the internal standard solution was loaded onto the Oasis HLB column solid phase extraction column, the column was eluted with ultrapure water, and the oxidized lipids were eluted into an Eppendorf tube with a methanol solution containing 0.02% formic acid by volume; The eluate was collected and blown dry with nitrogen, and the solid sample was collected.

9. The detection method according to claim 6, characterized in that: Step S13 specifically includes: adding a derivatization reagent solution to the concentrated solid sample to obtain a derivatization reaction system; The derivatization reaction system is vortexed, and then the derivatization reaction system is transferred to a sampling bottle, and the oxidized lipids in the derivatization system and the derivatization reagent solution are derivatized in the sample plate.

Images