Marker identification method and diagnosis model construction method for early gastric cancer detection

Abstract

The invention relates to the technical field of medicine, in particular to a marker identification method and a diagnosis model construction method for early gastric cancer detection, which comprises the following steps: screening out a plurality of robust difference miRNAs from a data set based on a linear model, an empirical Bayesian method and an RRA algorithm, and constructing a robust difference candidate marker set; constructing a co-expression network by adopting a WGCNA method; miRNA with GS and MM larger than corresponding thresholds in the co-expression network key module is screened out to serve as early gastric cancer related miRNA, and an early gastric cancer related candidate marker set is constructed; miRNAs with RNs scores and eigenvector centrality values exceeding corresponding thresholds in the gastric cancer specific regulation network are screened out to serve as early gastric cancer regulation miRNAs, and an early gastric cancer regulation candidate marker set is constructed; and integrating all candidate marker sets to obtain a final marker set. The identified marker is helpful for improving the early gastric cancer detection accuracy.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a marker identification method and a diagnostic model construction method for early gastric cancer detection. Background Art

[0002] Gastric cancer is a common malignancy worldwide and the fourth leading cause of cancer-related death. Gastric cancer is usually diagnosed at an advanced stage, and the prognosis of advanced gastric cancer (AGC) is poor, with a survival rate of only 20%; however, early treatment intervention can significantly improve the survival rate, potentially increasing it to 70%.

[0003] Currently, endoscopy is the gold standard for diagnosing gastric cancer; despite its high reliability, endoscopy also has limitations, including high cost, invasiveness, and associated risks. In the clinical diagnosis of gastric cancer, biomarkers such as carcinoembryonic antigen (CEA), carbohydrate antigen (CA) 19-9, CA 125, and pepsinogen are used. However, the limited specificity and sensitivity of these markers hinder their effectiveness in the early detection of gastric cancer. Therefore, the development of new diagnostic models for early gastric cancer is of great clinical significance for improving patient prognosis, reducing medical costs, and improving patients' quality of life. These models will facilitate early detection and treatment, which may significantly improve cure rates and reduce reliance on invasive diagnostic methods.

[0004] Liquid biopsy has emerged as a transformative approach in oncology, enabling noninvasive, real-time monitoring of tumor biomarkers such as circulating tumor cells, cell-free DNA, circulating tumor DNA, noncoding RNA (ncRNA), and exosomes from body fluids such as blood, urine, and cerebrospinal fluid. Compared with traditional tissue biopsies, liquid biopsies offer a less invasive and more practical means to track cancer progression, treatment efficacy, and drug resistance. With the rapid development of high-throughput sequencing and artificial intelligence-driven computing technologies, liquid biopsy has great clinical potential in the early diagnosis, prognosis assessment, and identification of chemotherapy resistance in patients with gastric cancer.

[0005] Before constructing a liquid biopsy-based diagnostic model for early gastric cancer, it is necessary to first identify microRNA markers for early gastric cancer detection. Most existing identification methods only consider the expression characteristics of microRNA (microRNA) in the blood, and lack consideration of the robustness of microRNA expression patterns and the regulatory function of microRNA itself, resulting in an inability to fully reflect the role of microRNA in gastric cancer, leading to reduced accuracy and reliability of the diagnostic model; failure to consider the robustness of microRNA expression patterns will lead to insufficient stability and anti-interference ability of the identified biomarkers in early gastric cancer; failure to consider the regulatory function of microRNA will lead to insufficient specificity and sensitivity of the identified biomarkers in early gastric cancer, thereby affecting the effect of early detection. Summary of the invention

[0006] To this end, the technical problem to be solved by the present invention is to overcome the problem that the prior art does not take into account the robustness of microRNA expression patterns and the regulatory function of microRNA, resulting in insufficient stability and anti-interference ability of the identified biomarkers in early gastric cancer, as well as insufficient specificity and sensitivity, which in turn leads to the inability to fully reflect the role of microRNA markers in gastric cancer, resulting in reduced accuracy and reliability of the diagnostic model, thereby affecting the detection effect of early gastric cancer.

[0007] In order to solve the above technical problems, the present invention provides a marker identification method for early gastric cancer detection, comprising:

[0008] Acquire a data set; wherein the data set includes a blood data set and a tissue data set; the data set includes patient data corresponding to multiple microRNAs;

[0009] Based on the linear model and empirical Bayes method, multiple differential microRNAs were screened from the data set to obtain a preliminary screening candidate marker set; based on the preliminary screening candidate marker set, a robust ranking aggregation algorithm was used to screen multiple robust differential microRNAs and construct a robust differential candidate marker set;

[0010] The weighted gene co-expression network analysis method was used to group the microRNAs in the data set according to the gene expression pattern to obtain multiple groups of modules, and a microRNA co-expression network was constructed; the P value and correlation coefficient of each group of modules were calculated; the modules with a P value less than their threshold and a correlation coefficient less than their threshold were taken as key modules; the gene significance and module membership of each microRNA in the key module were calculated; multiple microRNAs with gene significance greater than their threshold and module membership greater than their threshold in the key module were screened as multiple early gastric cancer-related microRNAs, and a set of candidate markers related to early gastric cancer was constructed;

[0011] Construct a gastric cancer-specific microRNA regulatory network and calculate the eigenvector centrality of each microRNA in the gastric cancer-specific microRNA regulatory network; calculate the RNs score of each microRNA in the gastric cancer-specific microRNA regulatory network based on an artificial intelligence network scoring system; screen out multiple microRNAs in the gastric cancer-specific microRNA regulatory network whose RNs score exceeds the threshold and whose eigenvector centrality value exceeds the threshold as multiple early gastric cancer regulatory microRNAs, and construct a set of candidate markers for early gastric cancer regulation;

[0012] The robust differential candidate marker set, the early gastric cancer-related candidate marker set, and the early gastric cancer regulatory candidate marker set were integrated to obtain the final marker set.

[0013] Preferably, the method of screening multiple differential microRNAs from a data set based on a linear model and an empirical Bayesian method to obtain a preliminary screening candidate marker set; based on the preliminary screening candidate marker set, screening multiple robust differential microRNAs using a robust ranking aggregation algorithm, and constructing a robust differential candidate marker set includes:

[0014] Based on the blood dataset and tissue dataset, a linear model is constructed, and its expression is:

[0015] ;

[0016] in, Indicates The microRNA in The expression level in each sample; Indicates samples, when the sample belongs to the blood dataset, , when the sample belongs to the organizational dataset, ; Indicates The average expression of microRNA in the blood dataset, i.e. the first parameter; Indicates The variation coefficient of the expression level of a microRNA in the tissue dataset relative to the expression level in the blood dataset, i.e., the second parameter;

[0017] The linear model was fitted using the limma function in the R language to obtain the estimated values ​​of the first parameter and the second parameter of each microRNA;

[0018] According to the empirical Bayesian method, the prior distribution of each microRNA is determined based on the estimated value of the first parameter and the estimated value of the second parameter of each microRNA; according to the Bayesian theorem, the posterior probability of each microRNA is calculated based on the prior distribution of each microRNA; the microRNAs with posterior probabilities greater than their thresholds are taken as differential microRNAs, and a preliminary screening candidate marker set is constructed;

[0019] The robust ranking aggregation algorithm was used to rank the stability of each differential microRNA in the preliminary screening candidate marker set, and the top 50 differential microRNAs were selected as robust differential microRNAs, and a robust differential candidate marker set was constructed.

[0020] Preferably, Correlation coefficient of modules The expression is:

[0021] ;

[0022] in, represents the Pearson correlation coefficient; Indicates the first The feature vector of each module; Indicates the disease status of early gastric cancer samples; the threshold of the module's correlation coefficient is 0.6.

[0023] Preferably, the threshold of the P value of the module is 0.05.

[0024] Preferably, Significant gene expression of microRNA The expression is:

[0025] ;

[0026] in, represents the Pearson correlation coefficient; Indicates The expression level of each microRNA; Indicates the disease status of early gastric cancer samples; the threshold of gene significance of microRNA is 0.45.

[0027] Preferably, Module membership of microRNAs The expression is:

[0028] ;

[0029] in, represents the Pearson correlation coefficient; Indicates The expression level of each microRNA; The feature values ​​representing key modules; the threshold for module membership of microRNAs is 0.6.

[0030] Preferably, the first The calculation formula of the eigenvector centrality of a microRNA is:

[0031] ;

[0032] in, Indicates The feature vector centrality of each microRNA; The adjacency matrix representing the gastric cancer-specific microRNA regulatory network; represents the eigenvalue; the threshold of the eigenvector centrality of microRNA is the median of the eigenvector centrality of all microRNAs in the gastric cancer-specific microRNA regulatory network.

[0033] Preferably, the first The expression of RNs score of microRNA is:

[0034] ;

[0035] in, Indicates the first The degree of the node, that is, The degree of each microRNA; represents the score of the support vector machine method based on recursive feature elimination; Indicates the first The average shortest path length from the node to other nodes, i.e., the shortest path length of the gastric cancer-specific microRNA regulatory network. The average shortest path length from one microRNA to other microRNAs; the threshold of the RNs score of a microRNA is the median of the RNs scores of all microRNAs in the gastric cancer-specific microRNA regulatory network;

[0036] The gastric cancer-specific microRNA regulatory network The degree of a node The expression is:

[0037] ;

[0038] in, Indicates The node and The connection weights between nodes; represents the number of all nodes in the gastric cancer-specific microRNA regulatory network;

[0039] The gastric cancer-specific microRNA regulatory network The average shortest path length from a node to other nodes The expression is:

[0040]

[0041] in, Indicates The node and The shortest path length between nodes.

[0042] Preferably, the blood dataset includes: GSE164174 dataset and GSE211692 dataset from the Gene Expression Comprehensive Database; the tissue dataset includes: GSE158315 dataset from the Gene Expression Comprehensive Database and a gastric cancer dataset from the Cancer Genome Atlas; the final marker set includes: miRNA miR-320b, miR-222-3p, miR-181a-5p, miR-103a-3p and miR-107.

[0043] The present invention also provides a method for constructing an early gastric cancer diagnosis model, comprising:

[0044] An early gastric cancer diagnosis model is constructed based on an artificial intelligence algorithm, using each microRNA marker in the final marker set obtained by the above-mentioned marker identification method for early gastric cancer detection as a model factor.

[0045] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0046] The present invention discloses a marker identification method for early gastric cancer detection. It uses an R language package and a robust ranking aggregation algorithm to screen out multiple microRNAs with robust differential expression. It uses a weighted gene expression network analysis method to divide multiple microRNAs in a data set into multiple groups of modules, and determines key modules by combining P values ​​with correlation coefficients and their corresponding thresholds. It screens out multiple microRNAs that are significantly associated with early gastric cancer based on the gene significance and module membership of all microRNAs in the key modules and their corresponding thresholds. On this basis, it screens out the key network indicators of each microRNA in the gastric cancer-specific microRNA regulatory network and the RNs scores of each microRNA in the network scoring system of artificial intelligence and their corresponding thresholds. Multiple microRNAs with strong regulatory capabilities; integrating all screened microRNAs to obtain multiple final microRNA markers; according to the above operations, the screened microRNA markers not only consider the expression characteristics of microRNA in the blood, but also consider the robustness of the microRNA expression pattern and the microRNA's own regulatory function, so that the biological and clinical functions of the identified microRNA markers are easier to explore and understand, so as to fully reflect the role of microRNA in gastric cancer, and then construct a diagnostic model with high accuracy and reliability for early gastric cancer detection, while reducing patients' dependence on invasive diagnostic methods, thereby reducing medical costs and improving patients' quality of life;

[0047] The method for constructing an early gastric cancer diagnostic model described in the present invention is based on the screened microRNA markers with robust differential expression patterns, high correlation with early gastric cancer, and significant regulatory capabilities, and according to an artificial intelligence algorithm, constructs an early gastric cancer diagnostic model, providing a new non-invasive detection, a more flexible and convenient, safer and more convenient detection method, which is helpful for the early detection of gastric cancer, thereby improving the cure rate and prognosis of patients; at the same time, due to the high stability and anti-interference ability, specificity and sensitivity of the identified early gastric cancer microRNA markers, the accuracy of early gastric cancer detection can be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0049] Figure 1 It is a flow chart of a marker identification method for early gastric cancer detection according to the present invention;

[0050] Figure 2 This is a schematic diagram of a diagnostic model based on five microRNA markers; Figure 2 A in the figure represents the ROC curve of the early gastric cancer diagnosis model based on random forest in the GSE164174 dataset; Figure 2 B in the figure represents the ROC curve of the early gastric cancer diagnosis model based on random forest in the GSE211692 dataset; Figure 2 C in the figure represents the ROC curve of the early gastric cancer diagnosis model based on random forest in the TCGA dataset; Figure 2 D in the figure represents the ROC curve of the early gastric cancer diagnosis model based on random forest in the serum cohort. DETAILED DESCRIPTION

[0051] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0052] Reference Figure 1 As shown, Figure 1 This is a flow chart of a marker identification method for early gastric cancer detection according to the present invention; specifically comprising:

[0053] S1: Obtain a data set; wherein the data set includes a blood data set and a tissue data set; the data set includes patient data corresponding to multiple microRNAs;

[0054] S2: Based on the linear model and empirical Bayes method, multiple differential microRNAs were screened from the data set to obtain a preliminary screening candidate marker set; based on the preliminary screening candidate marker set, a robust ranking aggregation algorithm was used to screen multiple robust differential microRNAs and construct a robust differential candidate marker set, including:

[0055] Based on the blood dataset and tissue dataset, a linear model is constructed, and its expression is:

[0056] ;

[0057] in, Indicates The microRNA in The expression level in each sample; Indicates samples, when the sample belongs to the blood dataset, , when the sample belongs to the organizational dataset, ; Indicates The average expression of microRNA in the blood dataset, i.e. the first parameter; Indicates The variation coefficient of the expression level of a microRNA in the tissue dataset relative to the expression level in the blood dataset, i.e., the second parameter;

[0058] The linear model was fitted using the limma function in the R language to obtain the estimated values ​​of the first parameter and the second parameter of each microRNA;

[0059] According to the empirical Bayesian method, the prior distribution of each microRNA is determined based on the estimated value of the first parameter and the estimated value of the second parameter of each microRNA; according to the Bayesian theorem, the posterior probability of each microRNA is calculated based on the prior distribution of each microRNA; the microRNAs with posterior probabilities greater than their thresholds are taken as differential microRNAs, and a preliminary screening candidate marker set is constructed;

[0060] The robust ranking aggregation algorithm was used to rank the stability of each differential microRNA in the preliminary screening candidate marker set, and the top 50 differential microRNAs were selected as robust differential microRNAs, and a robust differential candidate marker set was constructed;

[0061] S3: The weighted gene co-expression network analysis (WGCNA) method was used to group the microRNAs in the dataset according to their gene expression patterns to obtain multiple groups of modules and construct a microRNA co-expression network. The P value and correlation coefficient of each group of modules were calculated. Correlation coefficient of modules The expression is:

[0062] ;

[0063] in, represents the Pearson correlation coefficient; Indicates the first The feature vector of each module; Indicates the disease status of early gastric cancer samples;

[0064] The modules with P value less than the threshold and correlation coefficient less than the threshold are regarded as key modules; the threshold of P value of the module is 0.05; the threshold of correlation coefficient of the module is 0.6;

[0065] Calculate the gene significance (GS) and module membership (MM) of each microRNA in the key module, that is, Significant gene expression of microRNA The expression is:

[0066] ;

[0067] in, represents the Pearson correlation coefficient; Indicates The expression level of each microRNA; Indicates the disease status of early gastric cancer samples;

[0068] No. Module membership of microRNAs The expression is:

[0069] ;

[0070] in, represents the Pearson correlation coefficient; Indicates The expression level of each microRNA; Indicates the characteristic values ​​of key modules;

[0071] Multiple microRNAs with gene significance greater than the threshold and module membership greater than the threshold in the key module were screened as multiple early gastric cancer-related microRNAs, and a set of candidate markers related to early gastric cancer was constructed; the threshold of gene significance of microRNA was 0.45; the threshold of module membership of microRNA was 0.6;

[0072] S4: Construct a gastric cancer-specific microRNA regulatory network and calculate the eigenvector centrality of each microRNA in the gastric cancer-specific microRNA regulatory network. The calculation formula of the eigenvector centrality of a microRNA is:

[0073] ;

[0074] in, Indicates The feature vector centrality of each microRNA; The adjacency matrix representing the gastric cancer-specific microRNA regulatory network; represents the eigenvalue;

[0075] The artificial intelligence-based network scoring system calculates the RNs score of each microRNA in the gastric cancer-specific microRNA regulatory network, namely The expression of RNs score of microRNA is:

[0076] ;

[0077] in, Indicates the first The degree of the node, that is, The degree of each microRNA; represents the score of the support vector machine method based on recursive feature elimination; Indicates the first The average shortest path length from a node to other nodes, that is, The average shortest path length from one microRNA to other microRNAs;

[0078] The gastric cancer-specific microRNA regulatory network The degree of a node The expression is:

[0079] ;

[0080] in, Indicates The node and The connection weights between nodes; represents the number of all nodes in the gastric cancer-specific microRNA regulatory network;

[0081] The gastric cancer-specific microRNA regulatory network The average shortest path length from a node to other nodes The expression is:

[0082] ;

[0083] in, Indicates The node and The shortest path length between nodes;

[0084] A plurality of microRNAs whose RNs scores exceed their thresholds and whose feature vector centrality values ​​exceed their thresholds in the gastric cancer-specific microRNA regulatory network are screened out as a plurality of early gastric cancer regulatory microRNAs, and a set of candidate markers for early gastric cancer regulation is constructed; wherein the threshold of the feature vector centrality of the microRNA is the median of the feature vector centralities of all microRNAs in the gastric cancer-specific microRNA regulatory network; the threshold of the RNs score of the microRNA is the median of the RNs scores of all microRNAs in the gastric cancer-specific microRNA regulatory network; the early gastric cancer regulatory microRNA is a microRNA with high regulatory ability in early gastric cancer;

[0085] S5: Integrate the robust differential candidate marker set, the early gastric cancer-related candidate marker set, and the early gastric cancer regulatory candidate marker set to obtain the final marker set.

[0086] The present invention also provides a method for constructing an early gastric cancer diagnosis model, comprising:

[0087] An early gastric cancer diagnosis model was constructed based on artificial intelligence algorithms using each microRNA marker in the final marker set as a model factor; wherein the artificial intelligence algorithms included: decision tree, k-nearest neighbor (KNN), naive Bayes, random forest, support vector machine (SVM) and extreme gradient boosting (XGBoost).

[0088] In order to identify microRNA markers for early gastric cancer and verify the diagnostic model for early gastric cancer, the present invention retrieved microRNA expression profiles from 10,071 patient samples from the Gene Expression Omnibus (GEO) and The Cancer Genome Atlas (TCGA); the retrieved data sets included blood data sets and tissue data sets; wherein the blood data sets included: GSE164174 data sets and GSE211692 data sets from the Gene Expression Omnibus; the tissue data sets included: GSE158315 data sets from the Gene Expression Omnibus and gastric cancer data sets from The Cancer Genome Atlas (TCGA-STAD); wherein miRNA is an abbreviation of microRNA.

[0089] Based on S2, 501 and 669 differentially highly expressed miRNAs were detected in the blood datasets GSE164174 and GSE211692, respectively, and the intersection of the two was 471 miRNAs; similarly, 88 and 137 differentially highly expressed miRNAs were found in the tissue datasets GSE158315 and TCGA-STAD, respectively, and the number of differentially expressed miRNAs in common was 18; the Robust Rank Aggregation (RRA) algorithm was further used to optimize this selection, and finally 93 and 4 stably highly expressed miRNAs were obtained from the blood and tissue datasets, respectively.

[0090] Based on S3, the present invention uses the weighted gene co-expression network analysis (WGCNA) method to construct a miRNA co-expression network based on two blood data sets and two tissue data sets, which contains six modules; then, the correlation between the six modules and the disease state (early gastric cancer or normal control) is calculated respectively, and the module showing the strongest positive correlation with ESGC (early gastric cancer sample) and the strongest negative correlation with the normal sample is selected as the key module, the correlation coefficient between the key module and the early gastric cancer sample is COR=0.77, the correlation coefficient between the key module and the normal sample is COR=-0.77, and the p value of the key module is <0.001; in addition, all modules positively correlated with ESGC are also positively correlated with each other; based on these findings, the core miRNA of the key module and the miRNA significantly correlated with ESGC itself are selected as miRNAs highly correlated with ESGC, that is, early gastric cancer-related miRNA markers; wherein, the core miRNA is a microRNA with MM>0.6 and p value<0.05, and the miRNA significantly correlated with ESGC itself is GS>0.45 and p value< 0.05 of microRNA.

[0091] Expand the scope of the existing human miRNA regulatory network, so that the optimized human miRNA regulatory network covers 820 miRNAs and 12,735 mRNAs, integrate the microRNA and mRNA expression profiles in TCGA-STAD into the optimized human miRNA regulatory network, and construct a gastric cancer-specific microRNA regulatory network; Based on S4, calculate the key network indicators of each miRNA in the gastric cancer-specific microRNA regulatory network, including degree, closeness, betweenness, and eigenvector centrality;

[0092] The expression of the proximity is:

[0093] ;

[0094] The expression of the betweenness is:

[0095] ;

[0096] in, Indicates The proximity of nodes; Indicates The betweenness of nodes; Indicates The node and The number of shortest paths between nodes; Indicates The node and The shortest path between nodes goes through The number of paths to nodes;

[0097] At the same time, the RNs scoring function was applied to evaluate each miRNA in the network; through this comprehensive analysis, a total of 185 miRNA candidates (early gastric cancer-related microRNAs) were identified in the present invention, whose RNs and eigenvector centrality values ​​exceeded the median threshold of these parameters; these miRNAs are considered to have significant regulatory capabilities in early gastric cancer, namely, early gastric cancer regulatory microRNAs.

[0098] Combining the above three aspects, a total of five miRNAs (miR-320b, miR-222-3p, miR-181a-5p, miR-103a-3p, and miR-107) were obtained, and an early gastric cancer diagnostic model was constructed based on their expression levels.

[0099] Specifically, the GSE164174 microarray expression dataset was randomly divided into a 70% training set and a 30% test set. Using these datasets, the present invention constructed an early gastric cancer diagnosis model based on artificial intelligence through six different artificial intelligence methods based on the expression of the above five miRNAs. Among them, the random forest model performed best. The area under the curve (AUC) indicator was used to evaluate the performance of the early gastric cancer diagnosis model based on the random forest on the GSE164174 dataset, GSE211692, TCGA-STAD and serum cohorts. The ROC curves are shown in Figure 2. Figure 2 shown.

[0100] like Figure 2 As shown in A, the early gastric cancer diagnosis model based on random forest in GSE164174 achieved an AUC of 0.948 (95% confidence interval: 0.933-0.963; sensitivity: 0.902; specificity: 0.894); Figure 2 As shown in B, the AUC of the early gastric cancer diagnosis model based on random forest in GSE211692 is 0.977 (95% confidence interval: 0.973-0.980; sensitivity: 0.939; specificity: 0.925); Figure 2 As shown in Figure C, the AUC of the early gastric cancer diagnosis model based on random forest in TCGA was 0.815 (95% confidence interval: 0.750-0.881; sensitivity: 0.645; specificity: 0.889); Figure 2As shown in D in the figure, the AUC of the random forest-based early gastric cancer diagnosis model in the serum cohort was 0.802 (95% confidence interval: 0.723-0.882; sensitivity: 0.645; specificity: 0.889).

[0101] In summary, the present invention selects miRNAs for constructing an early gastric cancer diagnosis model from a large amount of miRNAs, combining the three characteristics of miRNA stable differential expression pattern, correlation with early gastric cancer, and regulatory ability. Since the method itself not only takes into account the expression level of microRNA itself, but also takes into account the regulatory function of microRNA, it is easier to explore the biological and clinical functions of the identified microRNA markers; then considering the accuracy, robustness and computational efficiency of the model, six artificial intelligence algorithms, namely decision tree, k-nearest neighbor, naive Bayes, random forest, support vector machine and extreme gradient boosting, are selected; through these six artificial intelligence algorithms, corresponding early gastric cancer diagnosis models are constructed respectively, and these six early gastric cancer diagnosis models are verified in multiple different data sets, and the best performance early gastric cancer diagnosis model based on random forest is selected, which has good reliability and wide applicability; in the process of diagnostic model construction and selection, the model is trained using high-quality data sets and verified in multiple data sets to ensure the stability and reliability of the model in different scenarios.

[0102] Obviously, the above embodiments are merely examples for the purpose of clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A marker identification method for early gastric cancer detection, characterized in that: include: Acquire a data set; wherein the data set includes a blood data set and a tissue data set; the data set includes patient data corresponding to multiple microRNAs; Based on the linear model and empirical Bayes method, multiple differential microRNAs were screened from the data set to obtain a preliminary screening candidate marker set; based on the preliminary screening candidate marker set, a robust ranking aggregation algorithm was used to screen multiple robust differential microRNAs and construct a robust differential candidate marker set; The weighted gene co-expression network analysis method was used to group the microRNAs in the data set according to the gene expression pattern to obtain multiple groups of modules, and a microRNA co-expression network was constructed; the P value and correlation coefficient of each group of modules were calculated; the modules with a P value less than their threshold and a correlation coefficient less than their threshold were taken as key modules; the gene significance and module membership of each microRNA in the key module were calculated; multiple microRNAs with gene significance greater than their threshold and module membership greater than their threshold in the key module were screened as multiple early gastric cancer-related microRNAs, and a set of candidate markers related to early gastric cancer was constructed; Construct a gastric cancer-specific microRNA regulatory network and calculate the eigenvector centrality of each microRNA in the gastric cancer-specific microRNA regulatory network; calculate the RNs score of each microRNA in the gastric cancer-specific microRNA regulatory network based on an artificial intelligence network scoring system; screen out multiple microRNAs in the gastric cancer-specific microRNA regulatory network whose RNs score exceeds the threshold and whose eigenvector centrality value exceeds the threshold as multiple early gastric cancer regulatory microRNAs, and construct a set of candidate markers for early gastric cancer regulation; The robust differential candidate marker set, the early gastric cancer-related candidate marker set, and the early gastric cancer regulatory candidate marker set were integrated to obtain the final marker set.

2. The marker identification method for early gastric cancer detection according to claim 1, characterized in that: Based on the linear model and empirical Bayes method, multiple differential microRNAs are screened from the data set to obtain a preliminary screening candidate marker set; Based on the preliminary screening of candidate marker sets, a robust ranking aggregation algorithm was used to screen out multiple robust differential microRNAs, and a robust differential candidate marker set was constructed, including: Based on the blood dataset and tissue dataset, a linear model is constructed, and its expression is: ； in, Indicates The microRNA in The expression level in each sample; Indicates samples, when the sample belongs to the blood dataset, , when the sample belongs to the organizational dataset, ; Indicates The average expression of microRNA in the blood dataset, i.e. the first parameter; Indicates The variation coefficient of the expression level of a microRNA in the tissue dataset relative to the expression level in the blood dataset, i.e., the second parameter; The linear model was fitted using the limma function in the R language to obtain the estimated values ​​of the first parameter and the second parameter of each microRNA; According to the empirical Bayesian method, the prior distribution of each microRNA is determined based on the estimated value of the first parameter and the estimated value of the second parameter of each microRNA; according to the Bayesian theorem, the posterior probability of each microRNA is calculated based on the prior distribution of each microRNA; the microRNAs with posterior probabilities greater than their thresholds are taken as differential microRNAs, and a preliminary screening candidate marker set is constructed; The robust ranking aggregation algorithm was used to rank the stability of each differential microRNA in the preliminary screening candidate marker set, and the top 50 differential microRNAs were selected as robust differential microRNAs, and a robust differential candidate marker set was constructed.

3. The marker identification method for early gastric cancer detection according to claim 1, characterized in that: No. Correlation coefficient of modules The expression is: ； in, represents the Pearson correlation coefficient; Indicates the first The feature vector of each module; Indicates the disease status of early gastric cancer samples; the threshold of the correlation coefficient of the module is 0.

6.

4. The marker identification method for early gastric cancer detection according to claim 1, characterized in that: The threshold of P value for the module was 0.

05.

5. The marker identification method for early gastric cancer detection according to claim 1, characterized in that: No. Significant gene expression of microRNA The expression is: ； in, represents the Pearson correlation coefficient; Indicates The expression level of each microRNA; Indicates the disease status of early gastric cancer samples; the threshold of gene significance of microRNA is 0.

45.

6. The marker identification method for early gastric cancer detection according to claim 1, characterized in that: No. Module membership of microRNAs The expression is: ； in, represents the Pearson correlation coefficient; Indicates The expression level of each microRNA; The feature values ​​representing key modules; the threshold for module membership of microRNAs is 0.

6.

7. The marker identification method for early gastric cancer detection according to claim 1, characterized in that: The first in the gastric cancer-specific microRNA regulatory network The calculation formula of the eigenvector centrality of a microRNA is: ； in, Indicates The eigenvector centrality of each microRNA; The adjacency matrix representing the gastric cancer-specific microRNA regulatory network; represents the eigenvalue; the threshold of the eigenvector centrality of microRNA is the median of the eigenvector centrality of all microRNAs in the gastric cancer-specific microRNA regulatory network.

8. The marker identification method for early gastric cancer detection according to claim 1, characterized in that: The first in the gastric cancer-specific microRNA regulatory network The expression of RNs score of microRNA is: ； in, Indicates the first The degree of the node, that is, The degree of each microRNA; represents the score of the support vector machine method based on recursive feature elimination; Indicates the first The average shortest path length from the node to other nodes, i.e., the shortest path length of the gastric cancer-specific microRNA regulatory network. The average shortest path length from one microRNA to other microRNAs; the threshold of the RNs score of a microRNA is the median of the RNs scores of all microRNAs in the gastric cancer-specific microRNA regulatory network; The gastric cancer-specific microRNA regulatory network The degree of a node The expression is: ； in, Indicates The node and The connection weights between nodes; represents the number of all nodes in the gastric cancer-specific microRNA regulatory network; The gastric cancer-specific microRNA regulatory network The average shortest path length from a node to other nodes The expression is: ； in, Indicates The node and The shortest path length between nodes.

9. The marker identification method for early gastric cancer detection according to claim 1, characterized in that: The blood dataset includes: GSE164174 dataset and GSE211692 dataset from the Gene Expression Comprehensive Database; the tissue dataset includes: GSE158315 dataset from the Gene Expression Comprehensive Database and a gastric cancer dataset from the Cancer Genome Atlas; the final marker set includes: miRNA miR-320b, miR-222-3p, miR-181a-5p, miR-103a-3p and miR-107.

10. A method for constructing an early gastric cancer diagnosis model, characterized in that: include: An early gastric cancer diagnosis model is constructed based on an artificial intelligence algorithm using each microRNA marker in the final marker set obtained by the marker identification method for early gastric cancer detection described in any one of claims 1-9 as a model factor.

Images