A method, device, medium and product for predicting venous thromboembolism based on genetic and non-genetic characteristics

Abstract

The present application discloses a method, device, medium and product for predicting venous thromboembolism based on genetic and non-genetic characteristics, and relates to the field of medical data processing. The method includes using the LASSO regression algorithm to perform feature screening on the screened genetic biological factor indicators and non-genetic factor indicators; using the screened genetic and non-genetic characteristics corresponding to patients with and without venous thromboembolism as positive sets and negative sets, and randomly dividing them into training sets and test sets using stratified sampling; using the training set and test set to train and test multiple machine learning models until multiple machine learning models meet the set conditions, integrating multiple trained machine learning models to obtain a venous thromboembolism prediction model; inputting the genetic and non-genetic characteristics of the tester into the venous thromboembolism prediction model to obtain a venous thromboembolism prediction result. The present application can significantly improve the real-time and accuracy of venous thromboembolism prediction.

Description

Technical Field

[0001] The present application relates to the field of medical data processing, and in particular to a method, device, medium and product for predicting venous thromboembolism based on genetic and non-genetic characteristics. Background Art

[0002] Thrombophilia refers to a pathological state that is prone to thrombosis and thromboembolism due to various genetic or acquired factors. The main clinical manifestation of thrombophilia is venous thromboembolism (VTE). VTE is commonly seen in superficial thrombophlebitis, deep venous thrombosis (DVT) of the lower extremities, and pulmonary embolism (PE). Other manifestations include Budd-Chiari syndrome (hepatic vein occlusion) and thrombosis of the inferior vena cava and cerebral, portal, mesenteric, renal, and retinal veins.

[0003] The incidence of VTE ranks third among cardiovascular and cerebrovascular diseases, second only to acute myocardial infarction and stroke. There are three recognized causes of VTE: hypercoagulable state, blood stasis and vascular endothelial damage, which often exist at the same time and interact with each other. The occurrence of venous thrombosis is the result of the combined effects of genetic and acquired factors, and the severity of individual genetic defects is closely related to the occurrence and mortality of venous thrombosis.

[0004] About 60% of the differences in individual susceptibility to VTE are attributed to genetic factors. Inherited thrombophilia is common in physiological anticoagulant proteins, such as antithrombin (AT), protein C (PC), and protein S (PS), which lead to a lack of protein anticoagulant function and cause thromboembolism. Anticoagulant protein deficiency is common in thrombophilia in the Chinese population. Procoagulant proteins, such as the coagulation factor V Leiden mutation and the prothrombin G20210A gene mutation, lead to enhanced protein procoagulant function and cause thromboembolism. Enhanced coagulation factor function is common in thrombophilia in the European and American population.

[0005] Acquired VTE risk factors include antiphospholipid syndrome, autoimmune diseases, malignant tumors, acute stroke, chronic cardiopulmonary diseases, chronic kidney disease, advanced age, obesity, surgery, limb immobilization or long-term bed rest, multiple trauma, fractures, etc. VTE is a preventable disease in many cases, and early risk stratification of patients and identification of high-risk patients may lead to more effective treatment strategies.

[0006] VTE is a vascular disease with high morbidity and hidden lethality. Its diagnosis is mainly based on clinical symptoms, signs, and imaging (refer to the Caprini score and Padua score items). The possibility of VTE is assessed based on a comprehensive score of the patient's symptoms, signs, and risk factors.

[0007] In recent years, with the development of gene sequencing technology, methods for predicting the risk of deep vein thrombosis based on genetic characteristics have emerged one after another. The Caprini Thrombosis Assessment Model and the Padua Prediction Score have incorporated the coagulation factor V Leiden mutation and the prothrombin G20210A mutation. Thrombophilia is race-specific. The coagulation factor V Leiden mutation and the prothrombin G20210A mutation are more common in Caucasians and cannot be directly applied to the Chinese population. Developing a model that incorporates genetic factors is a hot direction for VTE risk prediction.

[0008] The accuracy of existing thrombophilia risk prediction methods needs to be improved. The complexity of the etiology limits the accuracy and clinical application value of existing technical solutions for predicting the genetic risk of venous thromboembolism, resulting in doctors being unable to provide accurate guidance on patient prevention, thereby increasing the risk of clinical VTE. Summary of the invention

[0009] The purpose of this application is to provide a method, device, medium and product for predicting venous thromboembolism based on genetic and non-genetic characteristics, which can significantly improve the real-time and accuracy of venous thromboembolism prediction.

[0010] To achieve the above objectives, this application provides the following solutions:

[0011] In a first aspect, the present application provides a method for predicting venous thromboembolism based on genetic and non-genetic characteristics, comprising:

[0012] Screening of genetic biological factor indicators and non-genetic factor indicators;

[0013] Using LASSO regression algorithm to perform feature screening on the genetic biological factor index and the non-genetic factor index to obtain genetic and non-genetic characteristics;

[0014] The genetic and non-genetic characteristics corresponding to patients who developed VTE were taken as the positive set, and the genetic and non-genetic characteristics corresponding to patients who did not develop VTE were taken as the negative set;

[0015] The positive set and the negative set are randomly divided into a training set and a test set by using a stratified sampling method;

[0016] The training set and the test set are used to train and test a plurality of machine learning models until the plurality of machine learning models meet the set conditions, thereby obtaining a plurality of trained machine learning models;

[0017] Integrate multiple trained machine learning models to obtain a venous thromboembolism prediction model;

[0018] The genetic and non-genetic characteristics of the test subject are obtained, and the genetic and non-genetic characteristics of the test subject are input into the venous thromboembolism prediction model to obtain the venous thromboembolism prediction result.

[0019] Optionally, the process of screening for genetic biological factor indicators includes:

[0020] m single nucleotide polymorphism sites are screened from a database based on keywords; the keywords are determined based on the name, clinical manifestations and existing research of venous thromboembolism;

[0021] According to the distribution frequency of genotypes in the case group and the control group recorded in the public data corresponding to the m single nucleotide polymorphism sites, the random function is used to randomly allocate various genotypes in the case group and the control group;

[0022] m single nucleotide polymorphism sites were added one by one according to the OR value from high to low, and the PRS value of different single nucleotide polymorphism site combinations for each individual in the case group and the control group was calculated using the genetic scoring method. The AUCs of 1 to m combinations were obtained and compared, and the single nucleotide polymorphism sites included in the PRS-15 genetic scoring model were screened out, and these single nucleotide polymorphism sites were used as indicators of genetic biological factors.

[0023] Optionally, the process of screening for non-genetic factor indicators includes:

[0024] Obtain clinical data from patients with venous thromboembolism and healthy subjects;

[0025] Relatively stable parameters are screened out from the clinical data; the relatively stable parameters include one or more of total cholesterol, D-dimer, fasting blood glucose, C-reactive protein, platelet count, age, fibrinogen, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, gender and triglycerides.

[0026] Optionally, the method for predicting venous thromboembolism based on genetic and non-genetic characteristics further comprises:

[0027] The genetic biological factor index and the non-genetic factor index are subjected to univariate analysis.

[0028] Optionally, the LASSO regression algorithm implements feature screening by adding an L1 regularization term to the linear regression model.

[0029] Optionally, the multiple machine learning models include: linear regression analysis model, linear discriminant analysis model and random forest model.

[0030] Optionally, the venous thromboembolism prediction result is expressed as:

[0031] ;

[0032] In the formula, For the tester Prediction results of venous thromboembolism, For the j The detectors predicted by the machine learning model The prediction results of venous thromboembolism, For the j The weight coefficients of a machine learning model.

[0033] In a second aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-mentioned method for predicting venous thromboembolism based on genetic and non-genetic characteristics.

[0034] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned method for predicting venous thromboembolism based on genetic and non-genetic characteristics.

[0035] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for predicting venous thromboembolism based on genetic and non-genetic characteristics.

[0036] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0037] The present application provides a method, device, medium and product for predicting venous thromboembolism based on genetic and non-genetic characteristics. By screening genetic biological factor indicators and non-genetic factor indicators and using LASSO regression algorithm to perform feature screening on genetic biological factor indicators and non-genetic factor indicators, the site information and non-genetic indicators most relevant to the increased risk of venous thromboembolism can be obtained, the amount of data can be simplified, and the real-time prediction of venous thromboembolism can be improved. By using stratified sampling method to randomly divide positive sets and negative sets into training sets and test sets to train and test multiple machine learning models, and integrating multiple trained machine learning models, the obtained venous thromboembolism prediction model can obtain venous thromboembolism prediction results more accurately, thereby improving the accuracy of venous thromboembolism prediction, and providing a scientific basis for doctors to determine venous thromboembolism. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0039] Figure 1 This is an application environment diagram of a method for predicting venous thromboembolism based on genetic and non-genetic characteristics in an embodiment of the present application;

[0040] Figure 2 A schematic diagram of a process for predicting venous thromboembolism based on genetic and non-genetic characteristics provided in one embodiment of the present application;

[0041] Figure 3 A ROC curve diagram of a 57-SNP locus genetic scoring model provided in another embodiment of the present application;

[0042] Figure 4 AUC trend graph of 57 groups of SNP sites provided in another embodiment of the present application;

[0043] Figure 5 A ROC curve diagram of a 15-SNP locus genetic scoring model provided in another embodiment of the present application;

[0044] Figure 6 A LASSO screening variable result diagram provided in another embodiment of the present application;

[0045] Figure 7 A flowchart of a method for predicting venous thromboembolism based on genetic and non-genetic characteristics provided in another embodiment of the present application;

[0046] Figure 8 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0048] In 2021, Eduardo Salas et al. combined clinical and genetic risks to establish the inCode (TiC) scoring model (involving only two mutations). The model was established based on a Spanish population cohort and verified the VTE risk prediction ability of the TiC score in the Nordic population (from Sweden). The AUC value was 0.673, the sensitivity was 72.25%, and the specificity was 60.62%, which was more reliable than the analysis of only the genetic variation of factors VLeiden and G20210A. However, many studies have shown that the frequency of these two gene loci in the Han Chinese population is extremely low and has no significant association with the occurrence of thrombophilia in the Chinese population. Because thrombophilia has clear ethnic specificity, this model cannot be used directly in the Chinese population. In 2022, Hunan Times Genetic Medical Testing Technology Co., Ltd. developed a set of methods for predicting the genetic risk of venous thromboembolism in the Han Chinese population (see Chinese patent application CN114317724A). This method screened 8 genes and 10 biomarkers, and constructed a SNP site combination genetic risk prediction model in the Chinese population based on the polygenic genetic risk score (PRS) method. In a report by Chao Liu MD et al. in 2023, the AUC of the model was 0.718. However, the model only analyzed the genetic risk of thrombophilia and did not take acquired factors into account. The accuracy and sensitivity of the model screening need to be further improved in clinical applications.

[0049] Based on this, the development of a VTE risk prediction and assessment model based on artificial intelligence covering genetic and non-genetic factors can simplify the process of venous thrombosis prevention and treatment, accurately identify the risk factors for VTE, and make early prevention and individualized intervention more practical and targeted.

[0050] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0051] The method for predicting venous thromboembolism based on genetic and non-genetic characteristics provided in the embodiments of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, or it can be integrated on the server 104, or it can be placed on the cloud or other servers. The terminal 102 can send the genetic and non-genetic characteristics of the tester to the server 104. After the server 104 receives the genetic and non-genetic characteristics of the tester, the server 104 inputs the genetic and non-genetic characteristics of the tester into the constructed venous thromboembolism prediction model for the genetic and non-genetic characteristics of the tester to obtain the venous thromboembolism prediction result. The server 104 can feedback the obtained venous thromboembolism prediction result to the terminal 102.

[0052] In addition, in some embodiments, the method for predicting venous thromboembolism based on genetic and non-genetic characteristics can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly predict venous thromboembolism based on the genetic and non-genetic characteristics of the test subject, or the server 104 can obtain the genetic and non-genetic characteristics of the test subject from the data storage system and predict venous thromboembolism based on the genetic and non-genetic characteristics of the test subject.

[0053] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0054] In an exemplary embodiment, Figure 2 As shown, a method for predicting venous thromboembolism based on genetic and non-genetic characteristics is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate the method, which includes the following steps 200 to 206. Among them:

[0055] Step 200: Screening genetic biological factor indicators and non-genetic factor indicators.

[0056] Step 201: Using LASSO regression algorithm to perform feature screening on genetic biological factor indicators and non-genetic factor indicators to obtain genetic and non-genetic features.

[0057] Step 202: The genetic and non-genetic characteristics corresponding to the patients who developed venous thromboembolism are taken as a positive set, and the genetic and non-genetic characteristics corresponding to the patients who did not develop venous thromboembolism are taken as a negative set.

[0058] Step 203: Randomly divide the positive set and the negative set into a training set and a test set using a stratified sampling method.

[0059] Step 204: Use the training set and the test set to train and test the multiple machine learning models until the multiple machine learning models meet the set conditions, thereby obtaining multiple trained machine learning models.

[0060] Step 205: Integrate multiple trained machine learning models to obtain a venous thromboembolism prediction model.

[0061] Step 206: Obtain the genetic and non-genetic characteristics of the subject, and input the genetic and non-genetic characteristics of the subject into a venous thromboembolism prediction model to obtain a venous thromboembolism prediction result.

[0062] The implementation of the above steps 200 to 206 can significantly improve the real-time and accuracy of venous thromboembolism prediction.

[0063] In another exemplary embodiment of the present application, the process of screening genetic biological factor indicators in the above step 200 is described by taking the prediction of venous thromboembolism in the Chinese Han population as an example, wherein:

[0064] Step 1: m single nucleotide polymorphism (SNP) sites were screened from the database based on keywords. Keywords were determined based on the name, clinical manifestations and existing research of venous thromboembolism.

[0065] For example, (1) a systematic search for thrombosis was conducted in the GWAS database using the keywords (“venous thromboembolism” or “pulmonary embolism” or “deep vein thrombosis” or “thrombophilia”). The selection criteria for gene loci were set as a correlation with venous thromboembolism P < 5.0 × 10 -8 . Based on "venous thromboembolism", 421 loci were screened for research; based on "pulmonary embolism", 16 loci were screened for research. In summary, 437 loci were preliminarily screened in the GWAS database.

[0066] (2) A systematic search was conducted in PubMed, Embase, and Scope, with the keywords ("venous thromboembolism" or "pulmonary embolism" or "deep vein thrombosis" or "thrombophilia") AND ("genetic risk score" or "polygenic risk score" or "genome-wide association study" or "meta-analysis") AND ("Chinese population" or "Han nationality" or "East Asian population" or "Asian population"). Ten loci were selected from a large study on venous thromboembolism based on the Chinese Han population by Liu Chao et al. Three SNP loci were selected from a GWAS study on venous thromboembolism completed by Zhang Zhu et al. A total of 10 loci were selected from several other related meta-studies by the same scholar. One locus was selected from a meta-study on the prediction of venous thromboembolism conducted by Hu Bei et al. Two loci were selected from a study on susceptibility loci for pulmonary embolism by Wang Baoyan et al. One locus was selected from a study on genetic polymorphisms in patients with venous thrombosis by Wang Ziran et al. One locus was selected from the study of risk loci for venous thrombosis in the Han population by LTang et al. Two loci were selected from the study of hotspot mutation assessment for thrombosis by Li Lei et al. Thirty loci based on East Asian population studies in the past two years were screened.

[0067] (3) The variation frequencies of the selected gene loci were studied using the dbSNP database of the National Center for Bioinformation of the United States, and SNP loci with variation frequencies < 0.001 in East Asian or Asian populations were eliminated.

[0068] (4) For sites that appeared repeatedly in multiple studies, the Hardy-Weinberg equilibrium (HWE) was performed on the control genotype distribution of each study. P>0.05 was considered to be unbiased, and the corresponding 95% CI of the odds ratio (OR) was recalculated.

[0069] (5) Exclude SNP sites with OR values ​​≤ 1 associated with VTE (including sites weakly associated with VTE risk).

[0070] Based on the above steps (1) to (5), 57 SNP sites were finally screened (as shown in Tables 1 to 6), which were initially used for multi-gene collective characterization of the genetic risk prediction of venous thromboembolism in the Chinese Han population. These sites have been repeatedly confirmed to be associated with an increased risk of venous thromboembolism in the Chinese Han population. In addition, the references of the selected articles can be manually retrieved for summary. For repeated research sites, the latest research data can be integrated through meta-analysis.

[0071] Table 1 Partial positions of 57 SNP loci Table 1

[0072]

[0073] Table 2 Partial positions of 57 SNP loci Table 2

[0074]

[0075] Table 3 Partial positions of 57 SNP loci Table 3

[0076]

[0077] Table 4 Partial positions of 57 SNP loci Table 4

[0078]

[0079] In Tables 1 to 6, rsID is the reference single nucleotide polymorphism identifier (Reference SNPID, abbreviated as rs or RefSNP) of the dbSNP database, which is a string of digital numbers set by the dbSNP database to identify the variant site and is unique.

[0080] Table 5 Partial positions of 57 SNP loci Table 5

[0081]

[0082] Table 6 Partial positions of 57 SNP loci Table 6

[0083]

[0084] Step 2: According to the distribution frequency of genotypes in the case group and the control group recorded in the public data corresponding to the m single nucleotide polymorphism sites, use the random function to randomly assign various genotypes in the case group and the control group.

[0085] For example, a simulated study population of 2000 cases was used, including 1000 cases of venous thromboembolism and 1000 healthy controls. The distribution frequencies of genotypes in the case group and the control group were strictly recorded in the public data corresponding to the selected and complete 57 SNP sites, and the random function was used to randomly assign various genotypes in the case group and the control group.

[0086] Step 3: Use the polygenic genetic risk scoring method to determine the PRS value of each individual in the case group and the control group.

[0087] The PRS for each individual was calculated as log(OR) multiplied by the sum of weighted genotypes. The PRS value for each individual in the case group and the control group was calculated using the polygenic risk score (PRS) method.

[0088] .

[0089] In the formula, represents the number of SNP sites, =1 ,2 ,..., m . It represents the effect value of the SNP site (i.e., the Ln of the odds ratio [OR]). The genotype of the SNP locus is represented by {0, 1, 2}, that is, the number of risk alleles carried. If it carries 0, it is recorded as 0 (wild type), if it carries 1, it is recorded as 1 (heterozygous mutant), and if it carries two, it is recorded as 2 (homozygous mutant).

[0090] The 57 sites obtained above are used as the 57-SNP model. The 57-SNP model is as follows.

[0091] In this polygenic genetic risk score, To express the SNPs ( The genotypes of the 57 SNP loci were used to determine the risk alleles in Tables 1 to 6. and The PRS is calculated by summing the products of . The detailed calculation process of PRS is as follows:

[0092] .

[0093] The predictive ability of the genetic risk score PRS model was estimated by calculating the area under the curve (AUC) using the pROC package in the R program. The PRS value was used as the independent variable in the AUC calculation process, and whether the disease was present was used as the dependent variable.

[0094] Figure 3 The AUC of the 57-SNP genetic scoring model for predicting venous thromboembolism was 0.7653 (95% CI: 0.7448-0.7858).

[0095] Step 4: Add m single nucleotide polymorphism sites (OR values ​​from high to low) one by one for combination, and use the genetic scoring method to calculate the PRS value of different single nucleotide polymorphism site combinations for each individual in the case group and the control group, obtain the AUC of 1 to m combinations and compare them, and screen out the single nucleotide polymorphism sites included in the PRS-15 genetic scoring model as indicators of genetic biological factors.

[0096] The purpose of optimizing the risk gene data in step 4 above is to improve the accuracy and practicality of the prediction model, control the economic cost at a low level, and facilitate clinical application.

[0097] The SNP sites (OR values ​​from high to low) in the 57 SNP sites were added one by one and combined, and a total of 57 groups of SNP site combinations were obtained. The 57 groups of SNP site combinations were fitted by the above-mentioned PRS method, and the 1000 VTE cases and 1000 healthy people given above were selected as the research subjects, and the AUC was obtained. With AUC as the ordinate and SNP site combination as the abscissa (such as Figure 4 as shown).

[0098] Figure 4 The results showed that with the increase of SNP combination sites, the AUC value of the model first increased rapidly and then tended to be flat. When the PRS-15 SNP site was combined, the AUC reached 0.7451 (95% CI: 0.7238-0.7664), which was not statistically different from the 0.7653 (95% CI: 0.7448-0.7858) of the PRS-57 SNP site combination (P>0.05). The AUC of the PRS-15 SNP site combination was higher than the AUC of the venous thrombosis prediction model reported by Chao Liu et al. (based on the Chinese population) in 2023 [0.718 (95% CI: 0.696-0.740)]. The results show that the PRS-15 genetic scoring model has a better ability to predict the genetic risk of venous thromboembolism. Among them, the ROC curve of the 15-SNP site genetic scoring model is as follows Figure 5 shown.

[0099] The loci included in the PRS-15 genetic scoring model are:

[0100] rs146922325, rs199469469, rs16984852, rs8176719, rs77175619, rs2066865, rs1799762, rs8178847, rs1042579, rs1799883, rs1799983, rs2036914, rs1801133, rs2289252, and rs1801131 (as shown in Table 7).

[0101] Table 7 15 SNP loci included in the RRS index

[0102]

[0103] In another exemplary embodiment of the present application, the process of screening non-genetic factor indicators in the above step 200 is described by taking the prediction of venous thromboembolism in the Chinese Han population as an example, wherein:

[0104] Step 1: Obtain clinical data of patients with venous thromboembolism and healthy people.

[0105] For example, patients diagnosed with deep vein thrombosis (DVT) by ultrasound contrast imaging and patients diagnosed with PE by physical examination (PE) by pulmonary ventilation and pulmonary perfusion were included in the VTE group. Clinical data of patients with venous thromboembolism and healthy people were collected, and 841 patients with venous thromboembolism were included in the VTE group, and 9920 healthy people were included in the control group (Controls) for subsequent data analysis.

[0106] Step 2: Filter out relatively stable parameters from clinical data. Relatively stable parameters include total cholesterol (TC), D-dimer (D2), fasting blood glucose (FBG), C-reactive protein (CRP), platelet count (PLT), age (Age), fibrinogen (FIB), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), gender (Gender) and triglyceride (TG).

[0107] In another exemplary embodiment of the present application, in order to use non-parametric hypothesis testing to perform univariate analysis of PRS scores and 12 indicators including total cholesterol, D-dimer, fasting blood glucose, C-reactive protein, platelet count, age, fibrinogen, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, gender, and triglycerides to measure the difference between the VTE group and the Controls group, univariate analysis of genetic biological factor indicators and non-genetic factor indicators can be performed. In this embodiment, the categorical variable gender is expressed as a percentage, and the chi-square test is used to analyze the difference between the VTE group and the Controls group. The other 11 clinical indicators such as PRS scores are continuous variables, and these characteristics are used to find the differences between the two groups using Student's test (as shown in Table 8).

[0108] The results showed that when only single-factor indicators were used to predict venous thromboembolism, the AUCs of D-dimer and PRS were higher, and the individual predictive abilities of other indicators could not be optimized by the AUC of univariate analysis.

[0109] Table 8 Comparative analysis results between the VTE group and the Controls group for 12 indicators

[0110]

[0111] In Table 8, the P value (pvalue) is used to evaluate whether the results of the hypothesis test are statistically significant. When comparing two groups of data, hypothesis testing is a commonly used statistical method. T-test is used for continuous numerical variables, and chi-square test is used for grouping variables. The pvalue is used to measure the degree of difference between the two groups of data. Generally, pvalue<0.05 means that there is a significant statistical difference between the two groups of data.

[0112] In another exemplary embodiment of the present application, except for the prediction of D-dimer and PRS model, the prediction ability of genetic and non-genetic indicators cannot be distinguished by univariate analysis alone. In order to further optimize the combination of indicators, LASSO feature screening is adopted, and the non-zero coefficient variables are expected to be calculated to achieve the purpose of indicator screening.

[0113] The LASSO (Least Absolute Shrinkage and Selection Operator) regression algorithm used in step 202 is a commonly used feature selection method, which implements feature selection by adding an L1 regularization term to the linear regression model. The LASSO regression algorithm can reduce the coefficients of some unimportant features or even turn them to zero, thereby achieving the purpose of feature selection.

[0114] Linear Regression Model:

[0115] .

[0116] In the formula, yes The observation vector, for The design matrix for p, Yes The unknown parameter vector, is a random error, is the variance of the random error, is a random error The average expectation of is a random error The covariance of , I is the parameter term.

[0117] Estimated parameter vector The basic method is the least square method, the idea of ​​which is that random errors As small as possible, that is, to make ) reaches a minimum.

[0118] The goal of a general linear regression model is to minimize the residual sum of squares, i.e., to predict the target variable by fitting a linear equation.

[0119] However, in the study of this example, there are 12 independent variables (as shown in Table 8), some of which may have weak predictive power for venous thromboembolism or be redundant.

[0120] Regularization adds additional constraints or penalties to the linear regression model (loss function) to prevent overfitting and improve generalization ability.

[0121] The generalization function is the calculation function of the LASSO regression algorithm feature screening. The generalization function is changed from the original becomes , Usually it is L1 or L2 norm, is an adjustable parameter that controls the strength of the regularization.

[0122] When used in linear regression models, L1 regularization and L2 regularization are also called Lasso and Ridge. 1-norm of a vector , , that is, the elements of the vector The L1 regularization factor The L1 norm of is added to the loss function as a penalty term. Since the regularization term is non-zero, this forces the coefficients corresponding to those weak features to become 0. Therefore, L1 regularization often makes the learned model very sparse (coefficients is often 0), which makes L1 regularization a good feature selection method. The value of , the obtained model will become more and more sparse, that is, more and more feature coefficients β will become 0.

[0123] By using the LASSO regression algorithm, the coefficients of some unimportant variables (such as LDL, Fbg and Fib) are set to zero, so as to achieve the purpose of screening variables. After statistically calculating the non-zero coefficient variables, six variables including total cholesterol, D-dimer, C-reactive protein, platelet count, age and polygenic risk score PRS can be obtained as shown in Table 9.

[0124] Table 9 Screening results based on LASSO regression algorithm

[0125]

[0126] Figure 6 Shows the difference The change of MSE (mean square error) under the value of (λ). Figure 6 The two vertical lines are the results of the LASSO regression algorithm selection. .min, at this time can minimize the MSE. Another is .1se, at this time The MSE can be made within 1 times the standard error of the minimum mean square error, and the complexity of the model can be reduced.

[0127] In another exemplary embodiment of the present application, the construction of a venous thromboembolism prediction model is described by taking the prediction of venous thromboembolism in the Chinese Han population as an example.

[0128] According to whether the patient had VTE, 10761 subjects were divided into positive set and negative set. Patients with VTE were regarded as positive set, and patients without VTE were regarded as negative set. The positive set and negative set were randomly divided into independent training set and test set by stratified sampling method, and the training set feature matrix and test set feature matrix were obtained according to the sample feature matrix. The training set accounted for 70% of the data, and the test set accounted for 30% of the data.

[0129] Accuracy is a commonly used indicator in the evaluation of machine learning models. It is used to measure the proportion of samples that the model predicts correctly in the total number of samples. The higher the accuracy, the better the prediction ability of the model. Specificity is a measure of the probability that the model correctly predicts the negative class. 1-specificity is the missed diagnosis rate, which is an important indicator for evaluating the missed diagnosis rate in the medical field.

[0130] Therefore, this example uses accuracy and specificity to evaluate the predictive ability of the integrated model.

[0131] (1) Model 1 (i.e. Model-1): Logistic regression model.

[0132] A machine learning model was established through logistic regression analysis to predict the risk of VTE. The most relevant indicators or variables were selected through feature screening using the LASSO algorithm: total cholesterol, D-dimer, C-reactive protein, platelet count, age, and polygenic risk score, and their ability to actually predict VTE was calculated and evaluated.

[0133] Calculation formula: .

[0134] In the formula, is the predicted value, It is a measure of the classification model The predicted value of samples and the true value label The previous error, is the value parameter. The model is to fit the training samples, and this item is required to be minimum, that is, the model is required to fit the training data as much as possible. In order to ensure the minimum training error, and to hope that the model test error is small, a second item is added, that is, the parameter The regularization function Keep the model as simple as possible.

[0135] The risk prediction model based on genetic and non-genetic factors was constructed using the logistic regression algorithm based on the 10-fold cross-validation method and the package "glmnet" in the R program. The prediction results of machine learning model Model-1 are: , statistical analysis The true category of the test set samples , the prediction accuracy of Model-1 was 0.9176 and the specificity was 0.9898.

[0136] (2) Model 2 (i.e. Model-2): Linear Discriminant Analysis (LDA) model.

[0137] For this VTE prediction study, the hypothetical dataset , where any sample for dimensional vector, .definition For the The number of class samples, For the is a collection of class samples, and For the The mean vector of class samples is defined as For the The covariance matrix of the class samples (strictly speaking, it is the covariance matrix without the denominator).

[0138] The expression is: .

[0139] The expression is: . For the A sample from a collection of class samples.

[0140] For binary data, we only need to project the data onto a straight line. The assumed projection line is the vector , then for any sample , it is in the straight line The projection on , the center point of the two categories and In a straight line The projection on and Since LDA needs to make the distance between the category centers of different categories of data as large as possible, that is, to maximize At the same time, we hope that the projection points of the same category of data are as close as possible, that is, the covariance of the projection points of the same type of samples should be and As small as possible, i.e. minimized .

[0141] In summary, the optimization objective of this model is expressed as:

[0142] .

[0143] Define the intra-class scatter matrix for:

[0144] .

[0145] At the same time, define the inter-class scatter matrix for: .

[0146] The optimization goal is: In the formula, is the generalized Rayleigh entropy, whose maximum value is the matrix The maximum eigenvalue of for The eigenvector corresponding to the maximum eigenvalue of . In the binary classification, The direction is always ,set up , substitute , we can get , that is, the optimal projection direction can be determined by only finding the mean and variance of the original binary classification samples .

[0147] The linear discriminant analysis (LDA) algorithm was used to construct a risk prediction model based on genetic and non-genetic factors using the package "MASS" in the R program. The tolerance parameter was set to the default value of 1.0E-4. The actual ability to predict venous thromboembolism was calculated and evaluated. The prediction results of the test sample by the machine learning model Model-2 were as follows: , statistical analysis The true category of the test set samples , the prediction accuracy of Model-2 was 0.9191 and the specificity was 0.9979.

[0148] (3) Model 3: Random forest model.

[0149] The essence of Random Forest (RF) is a bagging algorithm, which averages the prediction results of the base evaluator or uses the majority voting principle to determine the results of the ensemble evaluator. The specific steps of the algorithm are as follows: Assume that there is a training data set of size N, and each time a sub-data set of size M is selected from the data set with replacement, for a total of K times. Based on these K sub-data sets, K models are trained and learned. When prediction is required, these K models are used for prediction, and then the final prediction result is obtained by taking the average or majority classification.

[0150] The calculation formula is: In the formula, is the preset number of trees, is the number of incorrect judgments, that is, the number of trees with incorrect judgments. is the probability of a tree making an incorrect judgment (i.e., random error), is the probability of correct judgment, and the total number of correct judgments is Second-rate. is the intermediate prediction result, The final prediction result.

[0151] The random forest algorithm was used to construct a risk prediction model based on genetic and non-genetic factors using the package "randomForest" in the R program, where the number of ntree decision trees was set to 500. The calculation was used to evaluate its ability to actually assess venous thromboembolism. The prediction results of the test sample by the machine learning model Model-3 were as follows: , statistical analysis The true category of the test set samples , the prediction accuracy of Model-3 was 0.9222 and the specificity was 0.9913.

[0152] (4) Construction of the integrated model Model-E, that is, construction of a venous thromboembolism prediction model.

[0153] The construction of the integrated model Model-E in this embodiment is based on the three machine learning models Model-1, Model-2 and Model-3 constructed based on the characteristic vectors of 6 non-zero coefficients (as shown in Table 9). Through the evaluation results analysis of the model construction in the above steps (1), (2) and (3), the three machine learning models all have good predictive capabilities for the risk of venous thromboembolism. The prediction accuracy of Model-1 is 0.9176 and the specificity is 0.9898. The prediction accuracy of Model-2 is 0.9191 and the specificity is 0.9979. The prediction accuracy of Model-3 is 0.9222 and the specificity is 0.9913. The three models each have their own advantages, and it is impossible to select the optimal model. Therefore, the concept of an integrated model is proposed.

[0154] The ensemble model (Model Ensemble, Model-E) is a fusion of multiple trained models, based on a certain method to achieve multi-model fusion of test data, so that the final result can "complement each other's strengths and weaknesses", integrate the learning ability of each model, and improve the generalization ability of the final model. Definition The risk prediction score of venous thromboembolism in the Chinese Han population based on genetic and non-genetic factors is obtained by using the integrated model Model-E, which comprehensively calculates the prediction results of Model-1, Model-2 and Model-3. The risk prediction score of venous thromboembolism in the Chinese Han population based on genetic and non-genetic factors is obtained by using the integrated model Model-E, which comprehensively calculates the prediction results of Model-1, Model-2 and Model-3. The calculation formula is: . For the tester A risk prediction score for venous thromboembolism based on genetic and non-genetic factors.

[0155] For the tester , It is the prediction result of Model-j model. is the weight coefficient of model Model-j, , .

[0156] The integrated model Model-E (i.e., the venous thromboembolism prediction model) integrates three machine learning models. These three models are independent of each other and have equal weights. , The accuracy of the integrated model Model-E was 0.9291, which was higher than that of Model-1, Model-2 and Model-3. When the C-score cutoff value of the risk prediction score of venous thromboembolism in the Chinese Han population based on genetic and non-genetic factors was 0.5, the specificity was 99.71%.

[0157] In another exemplary embodiment of the present application, threshold determination and stratification of the integrated model Model-E are important means to improve model performance and generalization ability. The core goal is to find an optimal segmentation point so that the integrated model Model-E can achieve the best performance in predicting the risk of venous thromboembolism in the Chinese Han population.

[0158] This example calculates the risk prediction scores of venous thromboembolism in the Chinese Han population based on genetic and non-genetic factors for the case group and the control group. Between 0 and 1. Following the absolute voting method ensemble strategy of integrated machine learning, the risk prediction score of venous thromboembolism in the Chinese Han population based on genetic and non-genetic factors was calculated. C-score The following stratification is performed: 1) If the integrated model If the score is between 0 and 0.2, the individual's risk level for VTE is low. If the score is between 0.2 and 0.5, the individual's risk level for VTE is average risk. If the score is between 0.5 and 0.75, the individual's risk level for VTE is medium risk. If the score is above 0.75, the risk level of the individual suffering from VTE is high risk. The risk stratification scoring results of the integrated model are shown in Table 10.

[0159] Table 10 Risk stratification score table of integrated model

[0160]

[0161] In summary, in practical application, the implementation process of the method for predicting venous thromboembolism based on genetic and non-genetic characteristics provided in this application can be participated in Figure 7 .

[0162] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 8As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the prediction results of venous thromboembolism. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting venous thromboembolism based on genetic and non-genetic characteristics is implemented.

[0163] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0164] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0165] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0166] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0167] 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, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0168] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0169] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0170] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for predicting venous thromboembolism based on genetic and non-genetic characteristics, characterized in that: The method for predicting venous thromboembolism based on genetic and non-genetic characteristics comprises: Screening of genetic biological factor indicators and non-genetic factor indicators; Using LASSO regression algorithm to perform feature screening on the genetic biological factor index and the non-genetic factor index to obtain genetic and non-genetic characteristics; The genetic and non-genetic characteristics corresponding to patients who developed VTE were taken as the positive set, and the genetic and non-genetic characteristics corresponding to patients who did not develop VTE were taken as the negative set; The positive set and the negative set are randomly divided into a training set and a test set by using a stratified sampling method; The training set and the test set are used to train and test a plurality of machine learning models until the plurality of machine learning models meet the set conditions, thereby obtaining a plurality of trained machine learning models; Integrate multiple trained machine learning models to obtain a venous thromboembolism prediction model; The genetic and non-genetic characteristics of the test subject are obtained, and the genetic and non-genetic characteristics of the test subject are input into the venous thromboembolism prediction model to obtain a venous thromboembolism prediction result, and the venous thromboembolism prediction result is expressed as: ； In the formula, For the tester Prediction results of venous thromboembolism, For the j The detectors predicted by the machine learning model Prediction results of venous thromboembolism, For the j The weight coefficients of the machine learning model; The process of screening genetic biological factor indicators includes: m single nucleotide polymorphism sites are screened from a database based on keywords; the keywords are determined based on the name, clinical manifestations and existing research of venous thromboembolism; According to the distribution frequency of genotypes in the case group and the control group recorded in the public data corresponding to the m single nucleotide polymorphism sites, the random function is used to randomly allocate various genotypes in the case group and the control group; m single nucleotide polymorphism sites were added one by one according to the OR value from high to low, and the PRS value of different single nucleotide polymorphism site combinations of each individual in the case group and the control group was calculated using the genetic scoring method. The AUC of 1 to m combinations was obtained and compared, and the single nucleotide polymorphism sites included in the PRS-15 genetic scoring model were screened out, and these single nucleotide polymorphism sites were used as indicators of genetic biological factors. The process of screening for non-genetic factors includes: Obtain clinical data from patients with venous thromboembolism and healthy subjects; Relatively stable parameters are screened out from the clinical data; the relatively stable parameters include one or more of total cholesterol, D-dimer, fasting blood glucose, C-reactive protein, platelet count, age, fibrinogen, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, gender and triglycerides.

2. The method for predicting venous thromboembolism based on genetic and non-genetic characteristics according to claim 1, characterized in that: The method for predicting venous thromboembolism based on genetic and non-genetic characteristics also includes: The genetic biological factor index and the non-genetic factor index are subjected to univariate analysis.

3. The method for predicting venous thromboembolism based on genetic and non-genetic characteristics according to claim 1, characterized in that: The LASSO regression algorithm implements feature screening by adding an L1 regularization term to the linear regression model.

4. The method for predicting venous thromboembolism based on genetic and non-genetic characteristics according to claim 1, characterized in that: Various machine learning models include: linear regression analysis model, linear discriminant analysis model and random forest model.

5. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for predicting venous thromboembolism based on genetic and non-genetic characteristics as described in any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting venous thromboembolism based on genetic and non-genetic characteristics described in any one of claims 1 to 4 is implemented.

7. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for predicting venous thromboembolism based on genetic and non-genetic characteristics described in any one of claims 1 to 4 is implemented.

Images