Method for constructing sepsis risk prediction model
By constructing a sepsis risk prediction method that integrates physiological indicators and gene prediction models, the problems of incomplete feature selection and overfitting in existing models are solved, achieving more accurate and stable risk assessment and supporting personalized medical decision-making.
Patent Information
- Application Number
- CN202510780346.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-12
- Publication Date
- 2025-11-07
AI Technical Summary
Existing sepsis risk prediction models are prone to overfitting when feature selection is not comprehensive enough, data volume is small, and model complexity is high. This results in poor generalization ability on test sets or in real-world applications, and the decision-making process of complex models is difficult to interpret.
By collecting biological sample data from sepsis patients and healthy controls, preprocessing the data, comparing gene expression data, screening differentially expressed genes, constructing physiological indicators and gene prediction models, and integrating physiological indicators and gene markers to form a comprehensive risk prediction model.
It improves the accuracy and stability of sepsis risk prediction, provides a more comprehensive risk assessment, reduces model overfitting, and can explain the causes of risk.
Smart Images
Figure CN120913871A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application provides a sepsis risk prediction model construction method, and relates to the technical field of sepsis risk prediction. BACKGROUND
[0002] Sepsis is a dysregulated immune response caused by infection, leading to life-threatening organ dysfunction. Early identification and intervention are crucial for improving prognosis. Traditional scoring systems such as SIRS, SOFA, and APACHE II are widely used, but they have low specificity and limited predictive ability. Therefore, developing more accurate and real-time risk prediction models has become a research hotspot.
[0003] In recent years, machine learning techniques have been widely applied in sepsis risk prediction. Models based on big data and machine learning can analyze clinical data such as vital signs and laboratory test results to predict the occurrence of sepsis several hours in advance, with an accuracy rate close to 90%, significantly better than traditional scoring systems. The development of machine learning models relies on high-quality clinical data. For example, the MIMIC-III and MIMIC-IV databases provide researchers with rich ICU patient data, including demographic information, vital signs, laboratory test results, etc. These data are used to train and validate prediction models to improve model accuracy and reliability.
[0004] However, the sepsis risk prediction models in the prior art may not be comprehensive or accurate enough, as some factors closely related to sepsis risk are not included, or some redundant and irrelevant features are included, affecting the performance of the model. For example, only the patient's basic physiological indicators are considered, while important factors such as the patient's immune status and medical history are ignored.
[0005] The sepsis risk prediction models in the prior art may overfit if the training data is relatively small and the model complexity is high. That is, the model performs well on the training set, but has poor generalization ability on new data in the test set or actual clinical application, resulting in inaccurate prediction results. Some complex machine learning models such as deep learning models may have certain advantages in prediction accuracy, but the decision-making process of the model is difficult to understand, and it is difficult to explain the causes of sepsis risk based on the output of the model, which is not conducive to the promotion and application of the model in practice. SUMMARY
[0006] To solve the above technical problems, the application provides a sepsis risk prediction model construction method, comprising the following steps: Step S1, collecting biological sample data of sepsis patients and healthy control population, preprocessing the collected biological sample data to obtain gene expression data of sepsis patients and healthy control population; Step S2, comparing the gene expression data of sepsis patients and healthy control population to find the differentially expressed genes between the two groups; Step S3, screening the differentially expressed genes by using a feature selection algorithm to obtain gene markers; Step S4, constructing a physiological index prediction model, inputting the personal physiological parameters affecting uremia into the physiological index prediction model; constructing a gene prediction model, inputting the gene markers obtained in step S3 into the gene prediction model, fusing the physiological index prediction model and the gene prediction model to construct a risk prediction model, and predicting the comprehensive risk value of sepsis occurrence.
[0007] In a preferred embodiment, in step S3, the feature selection algorithm uses a parameter linear screening model to select gene markers; The parameter linear screening model is: ; Wherein, y is the response variable, is the initial feature coefficient, x i is the i-th differentially expressed gene, is the corresponding feature coefficient, is the error term, and n is the number of differentially expressed genes involved in screening; The control function of the parameter linear screening model is: ; Wherein, Y i is the predicted value, is the control parameter, and the feature coefficient of the differentially expressed gene to be excluded is controlled to 0 by solving .
[0008] In a preferred embodiment, step S4 includes: S41, constructing a physiological index prediction model, inputting the personal physiological parameters affecting uremia into the physiological index prediction model, and outputting the physiological occurrence probability of sepsis; S42, constructing a gene prediction model, inputting the gene markers obtained in step S3 into the gene prediction model, and outputting the gene potential risk value of sepsis; S43, constructing a risk prediction model, inputting the physiological occurrence probability of sepsis predicted in step S41 and the gene potential risk value of sepsis predicted in step S42, and predicting the comprehensive risk value of sepsis occurrence.
[0009] In the preferred embodiment, in the step S41, assuming that the number of physiological parameters is m, and each physiological parameter has p indexes, the Jth index of the Ith physiological parameter is denoted as X IJ , wherein I = 1, 2, …, m; J = 1, 2, …, p; The occurrence probability of the Ith physiological parameter is calculated as : ; wherein, wherein is the initial weight, is the reference weight of the Jth index; The physiological index prediction model is: ; The greater the model output value S(X) is, the greater the occurrence probability of sepsis physiology is; The smaller the model output value S(X) is, the smaller the occurrence probability of sepsis physiology is.
[0010] In the preferred embodiment, in the step S42, assuming that the number of genetic markers is N, and the number of sepsis risk categories is M r , the rth genetic marker is denoted as M , and the genetic classification model is constructed as follows: Construct a classification function F(M r ) and a target fitting function : ; ; wherein, is a fitting control term, is a classification variable, W is a fitting parameter; C is a balance parameter, and the balance parameter C is used to balance the balance between the fitting control term and the classification variable . is a classification variable, which is used to handle the case of unclassifiable genetic markers.
[0011] In the preferred embodiment, in the step S43, the sepsis physiological occurrence probability predicted in the step S41 and the sepsis genetic potential risk value predicted in the step S42 are fused into input parameters of a risk prediction model, and a risk prediction model is constructed, taking the sepsis comprehensive risk value R as the prediction output. The risk prediction model is: ; wherein, is a weight parameter.
[0012] In the preferred embodiment, a risk threshold R T is set, and when the sepsis comprehensive risk value R output by the risk prediction model is greater than the risk threshold RT When the comprehensive risk value R of sepsis output by the risk prediction model is not greater than the risk threshold R T , it is considered that the target individual has a low risk of sepsis.
[0013] In the preferred embodiment, in step S2, the gene expression data of the sepsis patients and the healthy control population are compared using statistical methods to calculate the fold change: Fold change = average gene expression of sepsis patient group / average gene expression of healthy control group A fold change threshold is set, and the gene expression data with a fold change higher than the fold change threshold is selected as the differential expression gene.
[0014] Compared with the prior art, the present application has the following beneficial technical effects: 1. Collecting biological sample data of sepsis patients and healthy control population and preprocessing can ensure the data quality of subsequent analysis and lay the foundation for accurately finding differential expression genes. By removing noise, outliers and missing values in the data, the gene expression data can more truly reflect the differences between sepsis patients and healthy population, reducing false results caused by data quality problems.
[0015] 2. Comparing the gene expression data of sepsis patients and healthy controls to find differential expression genes helps to reveal the key genes in the development of sepsis. These genes may be involved in the pathophysiological processes of sepsis, such as inflammatory response, immune regulation, apoptosis, etc., providing important clues for in-depth understanding of the pathogenesis of sepsis.
[0016] 3. Differential expression genes can be used as potential biomarkers to provide a basis for early diagnosis of sepsis. By detecting the expression level of these genes, it is possible to achieve early detection of sepsis and improve accuracy and timeliness.
[0017] 4. Using feature selection algorithm to screen differential expression genes to obtain gene markers can screen out the most representative and judgmental genes from a large number of differential expression genes. This not only can reduce the workload of subsequent analysis, but also can improve the accuracy and stability of the model, avoiding the problem of model overfitting caused by too many irrelevant or redundant genes.
[0018] 5. The risk prediction model is constructed by fusing the physiological index prediction model and the gene prediction model, combining individual physiological parameters and gene markers, which can more comprehensively and accurately predict the comprehensive risk value of sepsis occurrence. This multi-factor prediction model considers the dual influence of physiological factors and genetic factors, which is more advantageous than single-factor prediction model, and can provide more reliable risk assessment results for clinicians. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 This is a flowchart of the method for constructing a sepsis risk prediction model according to the present invention.
[0021] Figure 2 This is a flowchart illustrating the process of obtaining gene expression data from sepsis patients and healthy controls according to the present invention.
[0022] Figure 3 This is a flowchart illustrating the method for constructing the risk prediction model of the present invention. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0024] In the accompanying drawings of specific embodiments of the present invention, in order to better and more clearly describe the working principle of each component in the system and show the connection relationship of each part in the device, only the relative positional relationship between each component is clearly distinguished. It does not constitute a limitation on the signal transmission direction, connection sequence, or size, dimension, and shape of each part within the component or structure.
[0025] like Figure 1 The diagram shows a flowchart of the sepsis risk prediction model construction method of the present invention, which includes the following steps: Step S1: Collect biological sample data from sepsis patients and healthy controls, preprocess the collected biological sample data to obtain gene expression data from sepsis patients and healthy controls.
[0026] like Figure 2 As shown, the specific process for obtaining gene expression data from sepsis patients and healthy controls is as follows: S11. Collect biological sample data from sepsis patients and healthy controls.
[0027] Collecting biological sample data of sepsis patients and healthy control population, including blood, tissue, etc., to ensure the representativeness and sufficient quantity of the samples to meet the needs of subsequent analysis.
[0028] Specifically, blood samples of patients diagnosed with sepsis can be collected from intensive care units (ICUs) of different hospitals, and blood samples of healthy people undergoing physical examination at the same time can be collected as controls.
[0029] S12, gene transcription profile analysis is performed on the collected biological sample data to obtain gene expression data.
[0030] High-throughput sequencing technology or gene chip technology is used to perform gene transcription profile analysis on the collected biological samples to obtain gene expression data. These gene expression data will reflect the differences in gene transcription levels under sepsis and normal physiological conditions.
[0031] S13, cleaning the gene expression data.
[0032] Remove noise, error values and missing values in gene expression data to ensure the quality and integrity of gene expression data.
[0033] Preferably, for RNA-seq data with low sequencing quality reads, filter out probe data with abnormally low signal intensity in gene chip data.
[0034] S14, standardizing the gene expression data.
[0035] Standardize the gene expression data to eliminate expression differences between different samples due to technical differences and other factors, making the data comparable.
[0036] The standardization method can use RPKM, FPKM, TMM.
[0037] Step S2, comparing the gene expression data of sepsis patients and healthy control population to find the differentially expressed genes between the two groups.
[0038] Screening for differentially expressed genes between the two groups of gene expression data.
[0039] Using statistical methods such as limma, DESeq2, etc., to compare the gene expression data of sepsis patients and healthy controls to find differentially expressed genes between the two groups, and calculate the fold change: Fold change = average gene expression of sepsis patients / average gene expression of healthy controls; According to the characteristics of the data, set the threshold of the fold change as the standard for screening differentially expressed genes, preferably 2 or 1.5 times as the lower limit of differential expression.
[0040] Step S3, applying a feature selection algorithm to screen the differentially expressed genes to obtain gene markers.
[0041] Applying a feature selection algorithm to screen the differentially expressed genes most relevant to the occurrence and development of sepsis from the differentially expressed genes, thereby as potential gene markers.
[0042] The feature selection algorithm sorts and screens the differentially expressed genes according to their predictive ability for sepsis status or interaction with other genes.
[0043] The feature selection algorithm uses a parameter linear screening model, so that the feature coefficients of some differentially expressed genes are controlled to be 0, thereby achieving the purpose of selecting gene markers.
[0044] The parameter linear screening model is: ; Wherein, y is the target variable, x i is the i-th differentially expressed gene, is the initial feature coefficient, is the corresponding feature coefficient, is the error term, and n is the number of differentially expressed genes participating in screening.
[0045] The control function of the parameter linear screening model is: ; Wherein, is the control parameter, the larger the , the stronger the restriction on the feature coefficient, which will make more feature coefficients tend to 0.
[0046] Residual sum of squares represents the difference between the actual value and the predicted value Y i , reflecting the goodness of fit of the model to the data. In the control function, this part encourages the model to fit the data as much as possible, so that the difference between the predicted value and the actual value is minimized.
[0047] The feature coefficient is screened by optimizing the control function , preferably, by solving to obtain the optimal solution. Some unimportant features (to be excluded) will be controlled to be 0, thereby achieving the effect of feature selection.
[0048] The coefficients of unimportant features refer to the coefficients of features that have weak relationship with the target variable y and make small contribution to the prediction ability of the model when constructing the parameter linear screening model. In this embodiment, the genes whose coefficients are not 0 are the genes most related to the occurrence and development of sepsis, and these genes can be used as potential gene markers. These genes play an important role in predicting the state of sepsis, and their expression levels have a strong linear relationship with the occurrence and development of sepsis.
[0049] In the preferred embodiment, to ensure the reliability of the screened gene markers, the linear regression model is verified using an independent test data set, or other biological experimental methods such as qRT-PCR, Western blot, etc. are used to verify the correlation between the expression of these genes and sepsis in different samples.
[0050] In the preferred embodiment, the stability and reliability of the screened gene markers are verified by external data sets or independent experiments. For example, the expression of these gene markers is detected in another group of sepsis patients and healthy control samples to observe whether they can also distinguish sepsis patients from healthy people. At the same time, the gene markers can be further optimized and adjusted according to the verification results.
[0051] It should be noted that the gene markers are related to multiple biological processes such as inflammatory response, immune regulation, coagulation function, etc. In this embodiment, the following gene markers are preferably used: A. Related to inflammatory response Proinflammatory cytokine genes: such as tumor necrosis factor-α (TNF-α), interleukin-1β (IL-1β), interleukin-6 (IL-6) genes. When sepsis occurs, pathogen-associated molecular patterns (PAMPs) and damage-associated molecular patterns (DAMPs) will activate immune cells, promote the up-regulation of transcription and expression of these genes, release a large amount of proinflammatory cytokines, trigger an inflammatory cascade, and cause fever, tissue damage, etc. Persistent excessive inflammatory response can cause organ dysfunction, which is an important link in the development of sepsis.
[0052] Anti-inflammatory cytokine genes: Anti-inflammatory cytokine genes such as interleukin-10 (IL-10) also play a key role in sepsis. In the early stage of sepsis, the body activates anti-inflammatory mechanisms, and the expression of anti-inflammatory factor genes such as interleukin-10 (IL-10) increases to inhibit the production of proinflammatory cytokines and inflammatory response, thereby reducing the damage of inflammation to the body. However, if the anti-inflammatory response is excessive, it will lead to immune suppression, increase the risk of secondary infection, and further worsen the condition of sepsis.
[0053] B. Related to immune regulation Human leukocyte antigen (HLA) gene: HLA gene is involved in the process of antigen presentation, which determines the immune recognition ability of the body to pathogens. The distribution of different HLA alleles in sepsis patients is different, and certain HLA genotypes may lead to abnormal antigen presentation, affecting the activation of T cells and the intensity of immune response, and reducing the body's ability to clear pathogens, thereby promoting the occurrence and development of sepsis.
[0054] Toll-like receptor (TLR) gene: TLR gene family can recognize specific molecular structure of pathogens and initiate immune signal transduction. For example, TLR4 gene can recognize bacterial lipopolysaccharide (LPS), activate downstream signaling pathway, and induce the production of inflammatory factors and the activation of immune cells. However, the polymorphism of TLR gene may affect its recognition and signal transduction ability to pathogens, leading to different susceptibility and immune response of individuals to sepsis.
[0055] C. Coagulation function related Tissue factor (TF) gene: During sepsis, inflammatory stimuli can increase the expression of TF gene in vascular endothelial cells, monocytes and other cells. TF binds with coagulation factor Ⅶa to initiate the extrinsic coagulation pathway, promote the generation of thrombin and deposition of fibrin, leading to microcirculation thrombosis, affecting the blood perfusion of tissues and organs, and aggravating tissue injury and organ dysfunction.
[0056] Protein C gene: Protein C is an important anticoagulant substance, and its gene expression product activated protein C (APC) can inactivate coagulation factors Ⅴa and Ⅷa, and inhibit the coagulation process. In sepsis, the expression of protein C gene may be affected by inflammation and other factors, and when the level of protein C decreases, the anticoagulant effect is weakened, the coagulation system is excessively activated, and thrombosis is easily formed, further aggravating the condition of sepsis.
[0057] D. Metabolic related Insulin-like growth factor binding protein-1 (IGFBP-1) gene: During sepsis, the body is in a state of hypercatabolism, and the expression of IGFBP-1 gene is up-regulated, leading to an increase in IGFBP-1 level. IGFBP-1 can bind with insulin-like growth factor (IGF), reduce the binding of IGF with receptor, affect the growth, proliferation and metabolism of cells, lead to an increase in muscle protein degradation, glucose metabolism disorder and other conditions, and affect the recovery of the body and organ function.
[0058] Hepatic fatty acid binding protein (FABP1) gene: FABP1 gene is involved in the transport and metabolism of intracellular fatty acids. In sepsis, FABP1 gene expression may change, affecting the metabolism and utilization of fatty acids, leading to lipid metabolism disorder, producing excessive free fatty acids and lipid peroxides, damaging cells and tissues, and also affecting the function of immune cells, further aggravating the pathophysiological process of sepsis.
[0059] Step S4, constructing a physiological index prediction model, inputting the personal physiological parameters affecting uremia into the physiological index prediction model; constructing a gene prediction model, inputting the gene markers obtained in step S3 into the gene prediction model, fusing the physiological index prediction model and the gene prediction model to construct a risk prediction model, and predicting the comprehensive risk value of sepsis occurrence.
[0060] As shown in Figure 3 , the method for constructing the risk prediction model is as follows: S41, constructing a physiological index prediction model, inputting the personal physiological parameters affecting uremia into the physiological index prediction model, and outputting the physiological probability of sepsis occurrence.
[0061] Suppose the number of personal physiological parameters participating in the study is m, and the physiological index prediction model is used for analysis. By constructing the physiological index prediction model, the physiological probability of sepsis occurrence is taken as the output label, which indicates whether sepsis is likely to occur on the physiological index.
[0062] Independent variables: personal physiological parameters affecting uremia, such as age, gender, underlying diseases, etc.
[0063] Output label: whether sepsis is likely to occur on the physiological index.
[0064] Suppose the number of physiological parameters is m, and each physiological parameter has p indicators. The Jth indicator of the Ith physiological parameter is denoted as X IJ , where I=1,2,…,m, I represents the sample number, J=1,2,…,p, and J represents the index number.
[0065] The occurrence probability of the Ith physiological parameter is calculated as :
[0066] wherein, wherein is the initial weight, is the reference weight of the Jth indicator.
[0067] The physiological index prediction model is: ; The greater the model output value S(X) is, the greater the physiological occurrence probability of sepsis is; The smaller the model output value S(X) is, the smaller the physiological occurrence probability of sepsis is.
[0068] S42, constructing a gene prediction model, inputting the gene markers obtained in step S3 into the gene prediction model, and outputting a sepsis gene potential risk value.
[0069] Supposing that the number of gene markers is N, M r is the rth gene marker.
[0070] The gene classification model construction includes the following multiple equations: First, a classification function F(M r ) and a target fitting function are constructed: ; ; Wherein, W is a fitting parameter; C is a balance parameter, and the balance parameter C is used to balance the balance between the fitting control term and the classification variable ; is a classification variable, which is used to process the case of unclassifiable gene markers.
[0071] The target fitting function is represented, and in the model training process, the fitting parameter W and the classification variable are found by optimizing this function, so that the classification function F(M r ) can classify M r , that is, whether the sepsis gene has potential risk, and “1” represents potential high risk, and “0” represents potential low risk.
[0072] S43, constructing a risk prediction model, inputting the physiological occurrence probability of sepsis predicted in step S41 and the potential risk value of sepsis gene predicted in step S42, and predicting a comprehensive risk value of sepsis.
[0073] The physiological occurrence probability of sepsis predicted in step S41 and the potential risk value of sepsis gene predicted in step S42 are fused as input parameters of the risk prediction model, and the risk prediction model is constructed, with the comprehensive risk value R of sepsis as the output label.
[0074] The construction process of the risk prediction model is as follows: ; Wherein, is a weight parameter.
[0075] The risk prediction model performs logarithmic transformation on the input parameters, which compresses the numerical range to some extent, making the risk prediction model more robust to inputs of different magnitudes. and are weight parameters that need to be determined according to actual conditions and take values in the range [0, 1] to balance the contributions of the two input parameters.
[0076] A risk threshold R is set T When the sepsis comprehensive risk value R output by the risk prediction model is greater than the risk threshold R T , it means that the sepsis incidence risk of the target individual is higher than the set standard, and the individual is in a high-risk state and needs to be detected comprehensively in a medical institution. When the sepsis comprehensive risk value R output by the risk prediction model is not greater than the risk threshold R T , it is considered that the sepsis incidence risk of the target individual is relatively low at present and is within an acceptable range.
[0077] Step S4 analyzes and predicts from two different dimensions of physiological indicators and genetic markers. Physiological indicators reflect the functional state of the target individual's body at the moment, and genetic markers reveal the potential of the target individual from the genetic level. The combination of the two can achieve a comprehensive assessment of the risk of sepsis, which is more accurate and comprehensive than single-dimensional assessment.
[0078] Considering individual physiological parameters and genetic markers embodies the concept of personalized medicine. The physiological conditions and genetic backgrounds of different individuals differ, and the prediction and evaluation based on this make the decision more in line with the individual's actual situation.
[0079] The model of each step is optimized through a large amount of data training, continuously improving the prediction accuracy, providing objective and reliable basis for clinical decision-making, and reducing the errors that may be caused by subjective judgment.
[0080] In one embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the above method embodiments when executing the computer program.
[0081] In one embodiment, a computer readable storage medium is provided, storing a computer program, which is executed by a processor to implement the steps in the above method embodiments.
[0082] In one embodiment, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to make the computer device execute the steps in the above method embodiments.
[0083] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, and the like, but is not limited thereto. The processor involved in the embodiments provided in the present application can be a general processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, and the like, but is not limited thereto.
[0084] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.
[0085] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.
Claims
1. A method for constructing a sepsis risk prediction model, characterized in that, The method comprises the following steps: Step S1, collecting biological sample data of sepsis patients and healthy control population, and preprocessing the collected biological sample data to obtain gene expression data of the sepsis patients and the healthy control population; Step S2, comparing the gene expression data of the sepsis patients and the healthy control population to find differentially expressed genes between the two groups; Step S3, screening the differentially expressed genes by using a feature selection algorithm to obtain gene markers; Step S4, constructing a physiological index prediction model, inputting personal physiological parameters affecting uremia into the physiological index prediction model; constructing a gene prediction model, inputting the gene markers obtained in step S3 into the gene prediction model, and fusing the physiological index prediction model and the gene prediction model to construct a risk prediction model to predict a comprehensive risk value of sepsis occurrence. 2.The method of claim 1, wherein, In the step S3, the feature selection algorithm adopts a parameter linear screening model to realize selection of the gene markers. The parameter linear screening model is as follows: ； where y is the response variable, is the initial feature coefficient, xi is the ith differentially expressed gene, is the corresponding feature coefficient, is the error term, and n is the number of differentially expressed genes involved in the screening. The control function of the parameter linear screening model is as follows: ； wherein Yi is the predicted value, is the control parameter, and the feature coefficients of the differentially expressed genes to be excluded are controlled to 0 by solving to screen the feature coefficients. 3.The method of claim 2, wherein, The step S4 comprises the following steps: S41, constructing a physiological index prediction model, inputting personal physiological parameters affecting uremia into the physiological index prediction model, and outputting a physiological occurrence probability of sepsis; S42, constructing a gene prediction model, inputting the gene markers obtained in step S3 into the gene prediction model, and outputting a gene potential risk value of sepsis; S43, constructing a risk prediction model, inputting the physiological occurrence probability of sepsis predicted in step S41 and the gene potential risk value of sepsis predicted in step S42, and predicting a comprehensive risk value of sepsis occurrence.
4. The method of claim 3, wherein the method is characterized by, In the step S41, the number of physiological parameters is m, each physiological parameter has p indexes, the Jth index of the Ith physiological parameter is denoted as XIJ, wherein I=1, 2, …, m; J=1, 2, …, p; The probability of occurrence of the first physiological parameter is calculated as : ； wherein, wherein is an initial weight, is a reference weight for the Jth indicator; The physiological index prediction model is as follows: ； The greater the model output value S(X) is, the greater the physiological occurrence probability of sepsis is; The smaller the model output value S(X) is, the smaller the physiological occurrence probability of sepsis is.
5. The method for constructing a sepsis risk prediction model according to claim 4, characterized in that, In the step S42, the number of gene markers is N, and Mr is the rth gene marker. The gene classification model is constructed as follows: Constructing the classification function F(Mr) and the target fitting function : ； ； wherein, is a fitting control term, is a categorical variable, W is a fitting parameter; C is a balancing parameter, the balancing parameter C is used to balance the fitting control term and the categorical variable between them; is a categorical variable, used to handle the case of non-separable genetic markers.
6. The method of claim 5, wherein the method further comprises: In the step S43, the physiological occurrence probability of sepsis predicted in step S41 and the gene potential risk value of sepsis predicted in step S42 are fused as input parameters of the risk prediction model, and the risk prediction model is constructed to take the comprehensive risk value R of sepsis as a prediction output. The risk prediction model is as follows: ； wherein is a weight parameter.
7. The method of claim 6, wherein the method further comprises: determining the risk of sepsis of the patient based on the determined risk score. Setting a risk threshold R T When the sepsis comprehensive risk value R output by the risk prediction model is greater than the risk threshold R T , it is considered that the target individual has a high risk of sepsis, and when the sepsis comprehensive risk value R output by the risk prediction model is not greater than the risk threshold R T , it is considered that the target individual has a low risk of sepsis. 8.The method of claim 1, wherein, In the step S2, the gene expression data of the sepsis patients and the healthy control population are compared by using a statistical method to calculate a difference ratio: The difference ratio = average gene expression amount of the sepsis patient group / average gene expression amount of the healthy control group; A difference ratio threshold is set, and the gene expression data with a difference ratio higher than the difference ratio threshold are taken as the screened differentially expressed genes.