Metabolic syndrome onset risk prediction system and method based on baseline amino acid

Abstract

The invention belongs to the field of metabolic syndrome onset risk prediction, and particularly relates to a metabolic syndrome onset risk prediction system and method based on baseline amino acid. According to the scheme, the system comprises an acquisition unit, a model construction unit and a prediction unit, wherein the acquisition unit is used for acquiring various baseline amino acid data, demographic information data and onset risk data of a subject; the model construction unit is used for respectively constructing a logistic regression model obeying binomial distribution single amino acid prediction and a BKMR model and a qgcomp model for multiple amino acid combination prediction by taking multiple baseline amino acids and demographic information as independent variables and taking the onset risk as dependent variables; various data acquired by an acquisition unit are used for fitting a logistic regression model, a BKMR model and a qgcomp model, and meanwhile, the fitting process of the three prediction models is regulated and controlled by checking model convergence. According to the scheme, a Bayesian kernel nonparametric regression method and a quantile g parameter calculation method are used for predicting the onset risk of the metabolic syndrome through baseline amino acid.

Description

Technical Field

[0001] This scheme belongs to the field of metabolic syndrome risk prediction, and specifically relates to a metabolic syndrome risk prediction system and method based on baseline amino acids. Background Art

[0002] Metabolic syndrome is a pre-stage of type 2 diabetes and cardiovascular and cerebrovascular diseases, characterized by central obesity, hyperlipidemia, hypertension and hyperglycemia. In recent years, the prevalence of metabolic syndrome has been gradually increasing, which has led to a significant increase in the prevalence, disability and mortality of its proximal outcomes, type 2 diabetes and cardiovascular and cerebrovascular diseases.

[0003] At present, the risk prediction of metabolic syndrome is mainly based on proteins as markers. In the protein markers and applications for predicting the probability of metabolic syndrome in the future time period proposed in Chinese patent CN117310171A, 11 protein markers with representative meanings are screened out based on serum proteomics data, combined with the recursive feature elimination algorithm based on Shapley values ​​and the Bourta feature selection method based on Shapley values; in addition, the age characteristics and gender characteristics of the subjects are input into the LGBM model, and the age characteristics are selected by using RFE based on feature importance. The relative expression of any combination of age characteristics and 11 protein markers is combined to predict the probability of the subject suffering from metabolic syndrome in the future time period. Among them, the recursive feature elimination algorithm of Shapley values ​​is often based on the assumption that there is a relatively simple linear relationship between the feature and the target variable during the feature selection process. However, in the relationship between serum proteomics and metabolic syndrome, the association between many protein markers and diseases may be nonlinear. For example, some protein markers may have different effects on metabolic syndrome at low and high concentrations, showing a complex "U"-shaped or inverted "U"-shaped relationship. This nonlinear relationship may cause the recursive feature elimination algorithm of Shapley value to fail to accurately capture the true relationship between features and diseases, thus affecting the accuracy of feature selection.

[0004] Amino acids are at the basic level of metabolism and are the starting materials or intermediates of many metabolic pathways. In the early stages of metabolic syndrome, metabolic disorders in cells will first cause fluctuations in amino acid levels. For example, branched-chain amino acids (BCAAs, including leucine, isoleucine, and valine) will change in the early stages of insulin resistance. When the insulin signaling pathway begins to show slight abnormalities, the metabolic regulation of BCAAs will be affected, and its concentration in the blood may increase, thereby indicating potential metabolic problems earlier than protein.

[0005] BKMR is a Bayesian nonparametric regression method used to model and predict the relationship between response variables and predictor variables. It uses kernel functions to map the original predictor variables (independent variables) to a high-dimensional space, so that it can handle nonlinear relationships between variables. In this high-dimensional space, the parameters of the model are estimated by Bayesian statistical methods, and the regression problem is converted into an inner product calculation in the feature space to measure the similarity between samples. In medical research, it is sometimes difficult to obtain a large amount of sample data, especially for studies of some rare diseases or specific populations. BKMR can also achieve good fitting results with small sample data and can handle high-dimensional exposure vectors, which provides a powerful tool for the analysis of limited data in medical research.

[0006] In this study, qgcomp was further used to evaluate the predictive effect of amino acid levels on MetS. This method is an improvement on weighted quantile sum (WQS) regression. It is based on the g calculation of the parameterized generalized linear model and combines the simplicity of reasoning and interpretability of WQS regression with the flexibility, simplicity and fast operation speed of g calculation. Different from WQS, this method does not assume directional homogeneity and is less affected by sample size. It also allows nonlinear or non-additive relationships between single amino acids and mixed amino acids and the dependent variable, and can more accurately estimate the overall effect of multiple amino acid corresponding variables. The weighted index (positive and negative weights) in different directions represents the relative contribution (i.e., importance) of each amino acid to the total effect of the outcome in that direction, and the sum of the weights of pollutants in the same direction is 1. Since this method is also robust in modeling in small sample size data, this study used the qgcomp method to divide the mixed amino acid level into quartiles (q = 4) to fit the multi-amino acid co-prediction model to evaluate the changes in the total effect OR and 95% CI of the MetS risk for each additional quartile of the multi-amino acid mixture.

[0007] At present, there is no system or method for predicting the risk of metabolic syndrome through baseline amino acids using the Bayesian kernel non-parametric regression method and the quantile g calculation parameter method. Summary of the invention

[0008] The purpose of this solution is to provide a system and method for predicting the risk of metabolic syndrome based on baseline amino acids to solve the above-mentioned problems.

[0009] In order to achieve the above purpose, this scheme provides a metabolic syndrome risk prediction system based on baseline amino acids, including:

[0010] Collection unit: used to collect various baseline amino acid data, demographic information data and disease risk data of the subjects;

[0011] Model building unit: used to use multiple baseline amino acids and demographic information as independent variables and the risk of disease as the dependent variable to respectively build a logistic regression model for predicting a single amino acid that obeys a binomial distribution, as well as a BKMR model and a qgcomp model for predicting a combination of multiple amino acids, and to fit the logistic regression model, the BKMR model and the qgcomp model using the data collected by the collection unit, and to regulate the fitting process of the three prediction models by checking the convergence of the models; the model building unit is used to put a type of baseline amino acid data into the logistic regression model after the training of the logistic regression model is completed, and to put a combination of multiple baseline amino acid data into the BKMR model and the qgcomp model after the training of the BKMR model and the qgcomp model is completed to deduce the correlation between the baseline amino acid and the risk of metabolic syndrome;

[0012] Prediction unit: The baseline amino acid data of the subjects were put into the logistic regression model, BKMR model and qgcomp model to calculate the risk of disease.

[0013] And, a method for predicting the risk of developing metabolic syndrome based on baseline amino acids using a system for predicting the risk of developing metabolic syndrome based on baseline amino acids

[0014] The principle and technical effect of this scheme is that this scheme takes into account a variety of baseline amino acids and demographic information, and compared with the single-factor evaluation method, it can more accurately reflect the actual risk of an individual. At the same time, this scheme can rank the importance of these factors by analyzing the contribution of each independent variable in the model (including baseline amino acids and demographic information) to the risk of disease. This helps to determine the factors that need to be focused on in research and clinical practice. Furthermore, this scheme can improve the accuracy of the model by regulating the fitting process by looking at the convergence of the model. A well-converged model can output the probability of disease risk more stably and accurately when predicting new data. For example, on the validation data set, a model that has been regulated for convergence may have higher accuracy and recall than an unregulated model, thereby better meeting the actual needs of disease risk prediction.

[0015] In summary, this protocol used the Bayesian kernel nonparametric regression method and the quantile g calculation parameter method to predict the risk of metabolic syndrome through baseline amino acids.

[0016] Furthermore, when the model construction unit is used to construct a logistic regression model, the matrix constructed from multiple baseline amino acid data is used as the exposure variable, the matrix constructed from demographic information data is used as the covariate, and the risk data corresponding to the exposure variable and the covariate are used as the dependent variable, and the Bayesian nonparametric regression method is used to fit the covariate, exposure variable and dependent variable data.

[0017] The Bayesian nonparametric regression method provides a flexible way to fit the data. Unlike traditional parametric regression models, it does not require the number of parameters in the form of functions to be specified in advance, but instead adaptively determines the complexity of the model based on the data. This flexibility is particularly important when dealing with complex metabolic syndrome data. At the same time, the relationship between baseline amino acids and the risk of disease may not be a simple linear relationship. Bayesian nonparametric regression can adapt to this complexity and discover potential complex associations in a data-driven way. This scheme uses Bayesian nonparametric regression to fit covariates, exposure variables, and dependent variable data, which can improve the accuracy of disease risk prediction, because this method can fit various relationships in the data more finely, so when new subject data is substituted into the model for disease risk calculation, it can give prediction results that are closer to the actual situation.

[0018] Furthermore, before fitting the model, the model building unit standardizes the exposure variables and sets the random number seed; when fitting the model, the Markov chain Monte Carlo method is used for fitting.

[0019] Standardization can help improve the performance and stability of the model, allowing the model to better capture the relationship between the exposure variable and the response variable. Especially when dealing with multiple exposure variables, standardization can ensure that each variable has a similar weight in the model, avoiding some variables dominating the model results due to their large dimensions.

[0020] Furthermore, when checking the convergence of the logistic regression model, the model construction unit evaluates the convergence of the model and the stability of the parameters by drawing a parameter trajectory diagram during the Bayesian model fitting process; checks the impact of baseline amino acids on the risk of disease by drawing a trajectory diagram of the regression coefficient, and checks the degree of difference between the model predicted value and the actual observed value by drawing a trajectory diagram of the variance of the error term.

[0021] If the trajectory of the regression coefficient is unstable or has no obvious convergence trend, it may mean that the model needs more sampling iterations or there are other problems, such as the model is not suitable for the data, the prior distribution is not properly selected, etc.; if the variance trajectory fluctuates greatly, it may mean that the model estimates the error inaccurately, or there is a lot of noise or outliers in the data. The regression coefficients drawn by this scheme can intuitively see the direction and intensity changes of the effects of different amino acids on the risk of disease, and can also easily compare their relative importance on the risk of disease, as well as some complex relationships. The variance trajectory drawn by this scheme can show the changes of the model prediction error in real time during the fitting process, and find out the overfitting or underfitting problems in the fitting process, so as to take corresponding measures, such as adjusting the complexity of the model, increasing the amount of data, or performing regularization.

[0022] Furthermore, when the model building unit is used to calculate the association between baseline amino acids and the risk of metabolic syndrome, the importance of each variable in the logistic regression model is measured, and then the overall risk summary of the fitted model is calculated, and the calculation results are plotted into data graphs.

[0023] This approach can help you visualize how risk changes by plotting estimates and confidence intervals versus quantiles. It can also help you visualize the overall risk summary to make it easier to compare risk differences under different conditions.

[0024] Furthermore, when the model building unit is used to infer the correlation between baseline amino acids and the risk of metabolic syndrome, the effect of any baseline amino acid on the risk of disease is calculated for the fitted model, and the preset quantiles are used for calculation one by one.

[0025] Furthermore, when the model building unit is used to infer the association between baseline amino acids and the risk of metabolic syndrome, the two baseline amino acids are combined and a data frame of exposure variable pairs is created, and then the data bars of the data frame are placed one by one into the fitted model for binary prediction, and the prediction is plotted into a two-dimensional color matrix; finally, the binary prediction results under different preset quantiles are calculated and plotted into a graph.

[0026] By calculating the impact of any baseline amino acid on the risk of disease, and using the preset quantiles to calculate one by one, the contribution of each amino acid to the risk of disease at different levels can be more accurately assessed. The use of quantiles can take into account the different positions of the amino acid concentration distribution, not just the average or typical value. At the same time, this calculation method fully considers the distribution characteristics of the data. The baseline amino acid levels of different individuals may vary greatly. The use of quantiles can cover the entire data range, thereby more comprehensively reflecting the relationship between amino acids and the risk of disease. This helps to discover the threshold effect of amino acid concentration, that is, when the amino acid concentration exceeds or falls below the level corresponding to a specific quantile, the risk of disease may change significantly, providing a basis for determining the risk warning value.

[0027] Furthermore, the model building unit is also used to fit the covariates, exposure variables and dependent variable data using the qgcomp model, and set the quantile sequence of exposure levels, draw a risk summary graph at different exposure levels in the quantile sequence of exposure levels based on the covariates, exposure variables and dependent variable data, and calculate the overall risk change trend of multiple baseline amino acid mixtures at different exposure levels based on the risk estimation values ​​in the risk summary graph.

[0028] The qgcomp model can handle different types of distributions, such as Gaussian distribution, binomial distribution, etc., and can adapt to different data characteristics and research needs. In addition, the qgcomp model allows nonlinear relationships between exposure variables and can capture the complex interactions and effects between baseline amino acids. By introducing nonlinear terms such as polynomial terms or interaction terms into the model, the joint effects of baseline amino acids and changes in health risks at different exposure levels can be more accurately described.

[0029] Furthermore, when the model building unit is used to calculate the association between baseline amino acids and the risk of metabolic syndrome, a combination of any two of leucine, isoleucine and valine or any two of tyrosine, phenylalanine and tryptophan is selected.

[0030] Most of the existing technologies focus on the study of single amino acids, exploring the relationship between them and certain diagnostic criteria of metabolic syndrome (such as hypertension and hyperlipidemia). Most of the existing technologies focus on the study of single amino acids, exploring the relationship between them and certain diagnostic criteria of metabolic syndrome (such as hypertension and hyperlipidemia). The relationship between the level of amino acids in pairs and the overall risk of metabolic syndrome has not been fully studied. However, since the level of a single amino acid is susceptible to many factors such as individual diet, metabolic state, genetic differences, and environmental factors, its prediction results often lack stability and reliability. These fluctuations may cause the level of a single amino acid to vary greatly between different individuals. In contrast, the level of amino acids in pairs (for example, the combination of leucine and isoleucine) can not only enhance the predictive ability, but also effectively reduce the random fluctuations of single amino acid variables. This stability enables the model to provide more consistent and reliable prediction results when facing individual differences, dietary changes, and other external influences, thereby improving the accuracy and operability of this scheme for predicting the risk of metabolic syndrome. In addition, since there may be a counteracting effect between amino acids, the predictive ability of simply considering the total amino acid level is weak. This scheme predicts the risk of metabolic syndrome by combining key amino acids, which can not only avoid the waste of time and resources caused by testing multiple amino acids, but also circumvent the redundancy in complex models. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Schematic diagram of the structure of the metabolic syndrome risk prediction system based on baseline amino acids.

[0032] Figure 2 This is a graph showing the correlation between the 17 baseline amino acids and the risk of developing metabolic syndrome in Example 1 of this protocol.

[0033] Figure 3 This is a graph showing the correlation between the six groups of two amino acid combinations and the risk of developing metabolic syndrome in Example 2 of this scheme. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the concept and technical effects of the present invention in combination with the embodiments, so as to fully understand the purpose, features and effects of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative work are all within the scope of protection of the present invention:

[0035] Example 1

[0036] like Figure 1 As shown in the figure, the risk prediction system for metabolic syndrome based on baseline amino acids includes:

[0037] Collection unit: used to collect various baseline amino acid data, demographic information data and disease risk data of the subjects;

[0038] Model building unit: used to construct a logistic regression model and a BKMR model that obey the binomial distribution by taking multiple baseline amino acids and demographic information as independent variables and the risk of disease as the dependent variable, fit the logistic regression model and the BKMR model using the data collected by the collection unit, and adjust the fitting process of the model by checking the convergence of the model; used to put one or more combinations of baseline amino acid data into the logistic regression model after the model training is completed, and put multiple combinations of baseline amino acid data into the BKMR model to deduce the association between baseline amino acids and the risk of metabolic syndrome;

[0039] Prediction unit: The baseline amino acid data of the subjects were put into the logistic regression model and BKMR model to calculate the risk of disease.

[0040] When the model building unit is used to construct a logistic regression model, the matrix constructed from multiple baseline amino acid data is used as the exposure variable, the matrix constructed from demographic information data is used as the covariate, and the risk data corresponding to the exposure variable and the covariate are used as the dependent variable. The Bayesian nonparametric regression method is used to fit the covariate, exposure variable and dependent variable data.

[0041] In the embodiment of this solution, the model building unit constructs the logistic regression model using the following code:

[0042] model<-glm(MS~logileJ+ageJ+gender+education+smoking2J+drinkingJ+exercise+taste+sweetness+grease+meats+milk+egg+beans,data=USE,family=binomial)

[0043] odds.ratio(model).

[0044] In the embodiment of this scheme, the data collected by the collection unit are all from healthy people who did not suffer from metabolic syndrome 4 years ago, a total of 362 research subjects, and after 4 years of follow-up, various data of the 362 research subjects are collected. Among them, the types of baseline amino acids include but are not limited to: polycyclic aromatic hydrocarbon metabolites (OHNap2, OHNap1, OHFlu3, OHFlu2, OHPhe2, OHPhe9, OHPhe1, OHPyr1, ∑OH-PAHs,), amino acids (branched-chain amino acids, leucine, isoleucine, valine), aromatic amino acids (tyrosine, tryptophan, phenylalanine), sulfur-containing amino acids (methionine, cysteine, cystine), amino acids related to ammonia transport (alanine, glutamic acid, glutamine), and others (serine, threonine, histidine, lysine, glycine).

[0045] In the embodiment of this scheme, the Gaussian kernel function (a mixed exposure model) method is used to construct the BKMR model. This method can not only estimate the nonlinear non-additive effects, but also take into account the interactions between exposures, and can flexibly estimate mixed exposures. In BKMR, the median of all mixed exposure substances is used as a reference, and the risk of disease at different levels of mixed exposure is compared with the risk of disease at the median of mixed exposure substances. Explore the changing trajectory of disease risk when mixed exposure is at different levels. BKMR is simulated using R software (version 4.4.0).

[0046] Among them, the following codes are used to set covariates, mixed exposure variables and dependent variables:

[0047] covar=data.matrix(USE[,c('ageS','gender','education','smoking2J','drinkingJ','exercise','taste','sweetness','grease','meats','milk','egg','beans')])

[0048] expos=

[0049] data.matrix(USE[,c('logleuJ','logileJ','logvalJ','logtyrJ','logtrpJ','logph eJ')])

[0050] USE$MS<-as.numeric(USE$MS)

[0051] USE$MS[USE$MS==1]<-0

[0052] USE$MS[USE$MS==2]<-1

[0053] Y=USE$MS.

[0054] Before fitting the model in the model building unit, the exposure variables were standardized and the random number seed was set; when fitting the model, the Markov chain Monte Carlo method was used for fitting.

[0055] In the example of this protocol, the standardized exposure factors use the following codes:

[0056] scale_expos=scale(expos).

[0057] The following code was used to fit the model:

[0058] set.seed(20230404)

[0059] fitkm=kmbayes(Y,Z=scale_expos,X=covar,iter=10000,verbose=FALSE,varsel=TRUE,family='binomial',est.h=TRUE)

[0060] Among them, Varsel indicates whether to perform variable screening on exposure.

[0061] When checking the convergence of the BKMR model, the model building unit evaluates the convergence of the model and the stability of the parameters by drawing the parameter trajectory diagram during the Bayesian model fitting process; the influence of baseline amino acids on the risk of disease is checked by drawing the trajectory diagram of the estimated coefficients, and the degree of difference between the model predicted values ​​and the actual observed values ​​is checked by drawing the trajectory diagram of the variance of the error term.

[0062] In the embodiment of this scheme, the following code is used to check the convergence of the model:

[0063] TracePlot(fit=fitkm,par='beta')

[0064] TracePlot(fit=fitkm,par="sigsq.eps")

[0065] TracePlot(fit=fitkm,par="r",comp=1).

[0066] When the model building unit is used to deduce the association between baseline amino acids and the risk of metabolic syndrome, the importance of each variable in the BKMR model is measured, and then the overall risk summary of the fitted model is calculated, and the calculation results are plotted into data graphs.

[0067] When the model building unit is used to infer the association between baseline amino acids and the risk of metabolic syndrome, the effect of any baseline amino acid on the risk of disease is calculated for the fitted model, and the preset quantiles are used for calculation one by one.

[0068] When the model building unit is used to infer the association between baseline amino acids and the risk of metabolic syndrome, the two baseline amino acids are combined and a data frame of exposure variable pairs is created. The data bars of the data frame are then placed one by one into the fitted model for binary prediction, and the prediction is plotted into a two-dimensional color matrix; finally, the binary prediction results under different preset quantiles are calculated and plotted into a graph.

[0069] In the embodiment of this scheme, the model building unit uses the following code to calculate the correlation between baseline amino acids and the risk of metabolic syndrome:

[0070] The code for filtering variables is as follows:

[0071] ExtractPIPs(fitkm)

[0072] Pip can be considered as a measure of variable importance, with higher values ​​(closer to 1) indicating higher importance and lower values ​​(closer to 0) indicating lower importance.

[0073] The code for joint exposure estimation is as follows:

[0074]

[0075] The code for estimating the single factor effect is as follows:

[0076]

[0077] The code for the univariate response relationship is as follows:

[0078]

[0079]

[0080] The code for the bivariate interaction is as follows:

[0081]

[0082] The code for calculating the binary prediction results under different preset quantiles (selecting 0.25, 0.5, and 0.754 quantiles) and drawing them into graphs is as follows:

[0083]

[0084] This scheme also provides a method for predicting the risk of metabolic syndrome based on baseline amino acids, and a method for predicting the risk of metabolic syndrome based on baseline amino acids uses the above-mentioned metabolic syndrome risk prediction system based on baseline amino acids.

[0085] The correlation between a single baseline amino acid and the risk of metabolic syndrome is calculated in this example and is shown in Table 1 below:

[0086]

[0087] Table 1

[0088] Logistic regression was used in the fitting process to adjust age, gender, smoking, drinking, exercise, and dietary habits (taste, fat, sweetness, livestock meat, milk and dairy products, eggs, beans and bean products). All amino acids were log-transformed with the natural logarithm as the base.

[0089] This example calculates the correlation between all 17 baseline amino acids (the 17 baseline amino acids shown in Table 1) and the risk of metabolic syndrome. Figure 2 shown.

[0090] Example 2

[0091] The only difference between this embodiment and Embodiment 1 is that the model building unit is also used to fit the covariates, exposure variables and dependent variable data using the qgcomp model, and set the quantile sequence of exposure levels, draw a risk summary graph at different exposure levels in the quantile sequence of exposure levels according to the covariates, exposure variables and dependent variable data, and calculate the overall risk change trend of multiple baseline amino acid mixtures at different exposure levels according to the risk estimate values ​​in the risk summary graph.

[0092] Specifically, when fitting the data using the qgcomp model, the baseline amino acid data is used as the exposure variable, the risk of metabolic syndrome is used as the dependent variable, and demographic information is considered as a covariate. For example, the qgcomp.boot function in the R language is used for fitting, and part of the code is as follows:

[0093]

[0094] Among them, risk is the risk variable for metabolic syndrome, amino1, amino2, and amino3 are baseline amino acid variables, age and gender are demographic information covariates, expnms specifies the name of the exposure variable, data is the data set, family specifies the model type (such as binomial distribution), and q is the quantile division of the exposure variable.

[0095] When calculating the risk summary, use the function provided by the qgcomp model to calculate the risk summary, for example: OverallRiskSummaries function, part of the code is as follows:

[0096] OverallRiskSummaries(fit=qc.fit,

[0097] qs=seq(0.25,0.75,by=0.05),

[0098] q.fixed = 0.5)

[0099] Among them, fit is the fitting result of the qgcomp model, qs is the quantile sequence of exposure levels, and q.fixed is the quantile of other fixed exposure variables.

[0100] Specifically, the ggplot2 package is used to draw a risk summary graph to show the changes in risk under different exposure levels. Part of the code is as follows:

[0101]

[0102]

[0103] Among them, geom_hline draws the reference line, geom_pointrange draws the risk point and its confidence interval, est is the risk estimate, and sd is the standard deviation.

[0104] In this embodiment, the overall risk change trend of the baseline amino acid mixture at different exposure levels can be intuitively observed through the risk summary graph. If the risk estimate increases significantly with the increase in exposure level, it means that the baseline amino acid mixture has a significant joint effect on the risk of metabolic syndrome. For example, if the risk estimate at a high exposure level (such as the 75% quantile) is significantly higher than that at a low exposure level (such as the 25% quantile), it indicates that the joint effect of the baseline amino acid mixture is strong and the health risk at a high exposure level is significantly increased.

[0105] In this example, the function provided by the qgcomp model is used to calculate the univariate effect, and the ggplot2 package is used to draw the effect curve. Through the effect curve, the effect change of a single baseline amino acid at different exposure levels can be observed. If the effect estimate of a baseline amino acid increases significantly at a high exposure level, it means that the baseline amino acid contributes more to the risk of metabolic syndrome at a high exposure level. For example, if the effect estimate at the 75% quantile is significantly higher than the 25% quantile, it indicates that the health risk of the baseline amino acid at a high exposure level increases more significantly. In addition, the effect size and change trend between different baseline amino acids can be compared to identify the baseline amino acids that have a greater impact on the risk of metabolic syndrome, providing a basis for subsequent intervention measures. Part of the code is shown below:

[0106] risks.sigvar<-SingVarRiskSummaries(fit=qc.fit,

[0107] qs.diff=c(0.25,0.75),

[0108] q.fixed = c(0.25,0.5,0.75))

[0109] Among them, fit is the fitting result of the qgcomp model, qs.diff is the quantile difference of exposure level, and q.fixed is the quantile of other fixed exposure variables.

[0110]

[0111] Among them, geom_pointrange plots the effect point and its confidence interval, coord_flip swaps the coordinate axes so that the variable name is displayed on the y-axis, variable is the baseline amino acid variable name, est is the effect estimate, sd is the standard deviation, and col=q.fixed sets the color according to the fixed quantile.

[0112] When the model building unit is used to infer the association between baseline amino acids and the risk of metabolic syndrome, a combination of any two of leucine, isoleucine and valine or any two of tyrosine, phenylalanine and tryptophan is selected.

[0113] There is too much data in the fitting process, so the results are not shown here.

[0114] The influence of the amino acid mixture effect on MetS (Qgcomp) calculated in this example is shown in Table 2 below:

[0115]

[0116] Table 2

[0117] Among them, age, gender, education level, smoking, drinking, exercise and dietary habits (taste, fat, sweetness, livestock meat, milk and dairy products, eggs, beans and bean products) were adjusted in the fitting process.

[0118] Among them, the association between the combination of 6 groups of two amino acids in Table 2 (i.e., leucine and isoleucine, leucine and valine, isoleucine and valine, tyrosine and phenylalanine, tryptophan and phenylalanine, and tyrosine and tryptophan) and the risk of metabolic syndrome is as follows Figure 3 shown.

[0119] In this embodiment, the "Guidelines for the Prevention and Treatment of Hypertension in China (2018 Revised Edition)" is used to diagnose MS, and a diagnosis can be made if three or more of the following items are met. (1) Abdominal obesity: waist circumference ≥90cm for men and ≥85cm for women; (2) Increased blood pressure: blood pressure ≥130 / 85mmHg and (or) those who have been diagnosed with hypertension and treated; (3) Dyslipidemia: fasting triglycerides ≥1.7mmol / L, fasting HDL-C <1.04mmol / L, or those who have been diagnosed with dyslipidemia and treated with drugs; (4) Hyperglycemia: fasting blood glucose ≥6.1mmol / L or blood glucose ≥7.8mmol / L 2h after sugar load, and (or) those who have been diagnosed with diabetes and treated. Metabolic syndrome can be diagnosed if any three or more of the above items are met. It should be noted that amino acid levels are not part of the diagnostic criteria for metabolic syndrome.

[0120] In this embodiment, when consulting the prior art, it was found that existing studies mainly focused on the correlation between a single amino acid level and a certain diagnostic criterion for metabolic syndrome. For example, the paper "The inverse associations of glycine and histidine in diet with hyperlipidemia and hypertension" (Nutr J. 2024Aug 22; 23: 98. doi: 10.1186 / s12937-024-01005-4) found that the levels of glycine and histidine in the diet, as well as high dietary glycine and high histidine groups, were negatively correlated with hyperlipidemia and hypertension. Most of these studies focus on the study of a single amino acid and explore its relationship with certain diagnostic criteria for metabolic syndrome (such as hypertension and hyperlipidemia). However, the relationship between the level of amino acid combinations and the overall risk of metabolic syndrome has not been fully studied.

[0121] A metabolic syndrome risk prediction method based on amino acid combination levels is proposed in this implementation. In the present embodiment, some single amino acids (such as isoleucine and tyrosine) are significant in metabolic syndrome risk prediction. However, due to the fact that single amino acid levels are susceptible to the influence of individual diet, metabolic state, genetic differences, environmental factors and other aspects, its prediction results often lack stability and reliability. These fluctuations may cause single amino acid levels to vary greatly between different individuals. In contrast, amino acid combination levels (for example, the combination of leucine and isoleucine) can not only enhance prediction ability, but also effectively reduce the random fluctuations of single amino acid variables, and this stability enables the model to provide more consistent and reliable prediction results when facing individual differences, dietary changes and other external influences, thereby improving the accuracy and operability of metabolic syndrome risk prediction. Due to the possibility of mutual cancellation effects between amino acids, the prediction ability of simply considering the total amino acid level is weak.

[0122] From the data on the influence of the amino acid mixture effect on MetS calculated by this embodiment, it can be seen that considering too many amino acid levels does not show excellent prediction effects, and by combining key amino acids, higher stability and prediction reliability can be provided. Therefore, this scheme proposes a metabolic syndrome risk prediction method based on the level of amino acid combinations. Through the analysis of amino acid combinations in pairs, this method significantly improves the accuracy of metabolic syndrome risk prediction. Compared with traditional methods, this method can not only avoid the waste of time and resources caused by detecting multiple amino acids, but also avoid redundancy in complex models.

[0123] This scheme can effectively improve the accuracy of metabolic syndrome risk prediction by analyzing the interactions and combination effects between amino acids, and provide a new early screening tool for clinical use. Compared with traditional single amino acid level predictions, this method can more accurately identify high-risk populations and avoid misjudgment problems caused by fluctuations in a single amino acid. By improving the stability of the prediction results, false positives and false negatives can be effectively reduced, thereby providing more solid data support for clinical decision-making. In summary, the advantage of the stability of pairwise combinations of amino acids in the risk prediction of metabolic syndrome can effectively improve the practicality and clinical value of the prediction model, especially under the high requirements for stability and reliability in practical applications, showing more competitive prediction performance.

[0124] The above is only an embodiment of the present invention, and the common knowledge such as the known specific structure and characteristics in the scheme is not described in detail here. It should be pointed out that for those skilled in the art, several deformations and improvements can be made without departing from the structure of the present invention, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A metabolic syndrome risk prediction system based on baseline amino acids, characterized in that: include: Collection unit: used to collect various baseline amino acid data, demographic information data and disease risk data of the subjects; Model building unit: used to use multiple baseline amino acids and demographic information as independent variables and the risk of disease as the dependent variable to respectively build a logistic regression model for predicting a single amino acid that obeys a binomial distribution, as well as a BKMR model and a qgcomp model for predicting a combination of multiple amino acids, and to fit the logistic regression model, the BKMR model and the qgcomp model using the data collected by the collection unit, and to regulate the fitting process of the three prediction models by checking the convergence of the models; the model building unit is used to put a type of baseline amino acid data into the logistic regression model after the training of the logistic regression model is completed, and to put a combination of multiple baseline amino acid data into the BKMR model and the qgcomp model after the training of the BKMR model and the qgcomp model is completed to deduce the correlation between the baseline amino acid and the risk of metabolic syndrome; Prediction unit: The baseline amino acid data of the subjects were put into the logistic regression model, BKMR model and qgcomp model to calculate the risk of disease.

2. The metabolic syndrome risk prediction system based on baseline amino acids according to claim 1, characterized in that: When the model building unit is used to build a logistic regression model, a matrix constructed from multiple baseline amino acid data is used as an exposure variable, a matrix constructed from demographic information data is used as a covariate, and the risk data corresponding to the exposure variables and covariates are used as dependent variables. The Bayesian nonparametric regression method is used to fit the covariate, exposure variable and dependent variable data.

3. The metabolic syndrome risk prediction system based on baseline amino acids according to claim 2, characterized in that: Before fitting the model, the model building unit standardizes the exposure variables and sets the random number seed; when fitting the model, the Markov chain Monte Carlo method is used for fitting.

4. The metabolic syndrome risk prediction system based on baseline amino acids according to claim 3, characterized in that: When checking the convergence of the logistic regression model, the model building unit evaluates the convergence of the model and the stability of the parameters by drawing a parameter trajectory diagram during the Bayesian model fitting process; The effect of baseline amino acids on the risk of disease can be viewed by plotting the trajectory of the regression coefficients, and the degree of difference between the model predictions and the actual observed values ​​can be viewed by plotting the trajectory of the variance of the error terms.

5. The metabolic syndrome risk prediction system based on baseline amino acids according to claim 4, characterized in that: When the model building unit is used to deduce the association between baseline amino acids and the risk of metabolic syndrome, the importance of each variable in the logistic regression model is measured, and then the overall risk summary of the fitted model is calculated, and the calculation results are plotted into data graphs.

6. The metabolic syndrome risk prediction system based on baseline amino acids according to claim 5, characterized in that: When the model building unit is used to infer the correlation between the baseline amino acid and the risk of metabolic syndrome, the effect of any baseline amino acid on the risk of disease is calculated for the fitted model, and the preset quantiles are used for calculation one by one.

7. The metabolic syndrome risk prediction system based on baseline amino acids according to claim 6, characterized in that: When the model building unit is used to infer the association between baseline amino acids and the risk of metabolic syndrome, two baseline amino acids are combined to create a data frame of exposure variable pairs, and then the data bars of the data frame are placed one by one into the fitted model for binary prediction, and the prediction is plotted into a two-dimensional color matrix; finally, the binary prediction results under different preset quantiles are calculated and plotted into a graph.

8. The metabolic syndrome risk prediction system based on baseline amino acids according to claim 7, characterized in that: The model building unit is also used to fit the covariates, exposure variables and dependent variable data using the qgcomp model, and set the quantile sequence of exposure levels, draw a risk summary graph at different exposure levels in the quantile sequence of exposure levels based on the covariates, exposure variables and dependent variable data, and calculate the overall risk change trend of multiple baseline amino acid mixtures at different exposure levels based on the risk estimation values ​​in the risk summary graph.

9. The metabolic syndrome risk prediction system based on baseline amino acids according to claim 8, characterized in that: When the model building unit is used to calculate the correlation between baseline amino acids and the risk of metabolic syndrome, any two of leucine, isoleucine and valine or a combination of any two of tyrosine, phenylalanine and tryptophan are selected.

10. A method for predicting the risk of metabolic syndrome based on baseline amino acids, characterized in that: Any one of the baseline amino acid-based metabolic syndrome risk prediction systems described in claims 1-9 is used.

Images