Machine learning-based super-enhancer related gastric cancer patient prognosis risk assessment system

Abstract

The application discloses a kind of based on machine learning's super enhancer related gastric cancer patient prognosis risk assessment system, including data acquisition module, data processing module, model training module and prognosis risk assessment module, wherein, data acquisition module is used to collect the historical data of gastric cancer patient, data processing module is used to carry out systemic quantitative analysis to the historical data of gastric cancer patient, model training module is used to train multiple machine learning models using the data after systemic quantitative analysis, prognosis risk assessment module is used to evaluate the prognosis risk of gastric cancer patient using the optimal model obtained after training.The application improves the precision of patient survival outcome prediction.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical technology, and in particular to a machine learning-based system for assessing the prognostic risk of patients with super-enhancer-associated gastric cancer. Background Technology

[0002] Gastric cancer (GC) is a malignant tumor originating from the epithelial cells of the gastric mucosa that can spread to surrounding tissues and organs. It is the fifth most common cancer worldwide and the third leading cause of death. According to data from the International Agency for Research on Cancer (IARC), 660,000 cases of gastric cancer died globally in 2022, accounting for 6.8% of all cancer deaths. my country is one of the countries with a high incidence of gastric cancer. In 2022, my country accounted for 7.4% of new gastric cancer cases and 7.3% of cancer deaths worldwide, higher than the world average. Furthermore, the overall prognosis for gastric cancer patients is poor, with a five-year survival rate of less than 30%, and a median survival time of less than 12 months for advanced-stage gastric cancer. Therefore, screening for gastric cancer-specific biomarkers to improve patient prognosis and provide a basis for developing personalized treatment plans has significant clinical importance and application value.

[0003] Super-enhancers (SEs) are a class of tandemly linked clusters of multiple transcriptionally active enhancers. Studies have shown that SEs are closely related to tumor progression and patient prognosis. SEs promote tumor development by driving the active transcription of key oncogenes and addictive oncogenes. Based on the expression profiles of SEs in tumors and normal tissues, gastric cancer-specific SEs can be screened, and their target genes can be further identified, thereby establishing prognostic risk models for high-risk populations of gastric cancer. With the rapid development of big data and information technology, artificial intelligence methods centered on machine learning are being more widely applied in clinical practice for early disease warning. Machine learning, with its advantages in high-throughput data processing, has been widely used and has shown great potential in constructing tumor prognostic models.

[0004] The current construction of gastric cancer risk scoring models has the following defects: (1) The traditional biomarker screening system does not integrate gastric cancer-specific super-enhancers regulatory network, resulting in the lack of characterization of key pathways in tumor evolution; (2) The selection of algorithms lacks bioinformatics basis, such as random forest, support vector machine and other algorithms showing significant performance differences in the same training set; (3) The heterogeneity of the tumor microenvironment causes feature drift, and single algorithm models are prone to overfitting. Summary of the Invention

[0005] To address at least one of the technical problems in the prior art, the present invention provides a machine learning-based prognostic risk assessment system for patients with super enhancer-related gastric cancer.

[0006] The purpose of this invention is to provide a machine learning-based prognostic risk assessment system for patients with super-enhancer-related gastric cancer, which can be achieved by adopting the following technical solutions:

[0007] A machine learning-based super-enhancer-related prognostic risk assessment system for gastric cancer patients includes a data acquisition module, a data processing module, a model training module, and a prognostic risk assessment module. The data acquisition module is used to collect historical data of gastric cancer patients, the data processing module is used to perform systematic quantitative analysis on the historical data of gastric cancer patients, the model training module is used to train multiple machine learning models using the data after systematic quantitative analysis, and the prognostic risk assessment module is used to assess the prognostic risk of gastric cancer patients using the optimal model obtained after training.

[0008] Furthermore, the historical data of the gastric cancer patients includes chromatin immunoprecipitation sequencing data of gastric cancer patients, RNA sequencing data of gastric cancer patients in a multicenter cohort, and survival data. The survival data includes survival status and survival time, with survival status including alive or dead.

[0009] Furthermore, the systematic quantitative analysis of historical data of gastric cancer patients includes:

[0010] Based on chromatin immunoprecipitation sequencing data, a candidate gene set of gastric cancer-specific super enhancer target genes was obtained.

[0011] Based on RNA sequencing data, survival data, and candidate gene sets, the patient's super enhancer prognostic genes were obtained.

[0012] Furthermore, the candidate gene set for gastric cancer-specific super enhancer target genes obtained based on chromatin immunoprecipitation sequencing data includes:

[0013] For chromatin immunoprecipitation sequencing data, peak calling was performed using MACS2 software to identify super enhancers (SEs).

[0014] For super enhancer SEs, the ROSE algorithm is used to identify super enhancer regions in order to eliminate interference from starter regions;

[0015] The super enhancer regions were compared with the GENCODE v43 gene annotation file to screen out target genes with H3K27ac peaks in the regions and whose distance from the gene transcription start site was less than a specified value. Target genes located in low-complexity regions of the genome were removed, and housekeeping genes co-localized with super enhancers were excluded. Finally, a candidate gene set of gastric cancer-specific super enhancer target genes was obtained.

[0016] Among them, gastric cancer-specific super enhancer target genes in the candidate gene set must meet the following conditions:

[0017] Gastric cancer-specific superenhancer target genes co-localize with their corresponding superenhancer regions;

[0018] The expression levels of gastric cancer-specific superenhancer target genes are significantly positively correlated with the corresponding superenhancer signal intensity.

[0019] Furthermore, based on RNA sequencing data, survival data, and candidate gene sets, the patient's super-enhancer prognostic genes are obtained, including:

[0020] Using RNA sequencing data and survival data, univariate Cox proportional hazards regression analysis was performed on gastric cancer-specific superenhancer target genes in the candidate gene set. The hazard ratio of gene expression level to survival time met the condition |log2(HR)|>0.5. After Benjamini-Hochberg correction, P<0.01 was used as the condition for statistical significance to screen superenhancer prognostic genes.

[0021] Furthermore, the expression profiles and survival data of super-enhancer prognostic genes are used as input data for the model, and the corresponding super-enhancer prognostic genes and their weights are used as output data for the model. The input data and output data constitute the sample.

[0022] The model training module includes a variable selection submodule and a model building submodule, both of which are pre-configured with various machine learning algorithms. The variable selection submodule is used to remove redundant features from the samples and retain the most predictive features using any machine learning algorithm. The model building submodule is used to train the model corresponding to any machine learning algorithm using the samples with retained features and select the optimal model from the trained models. The optimal model is the two models with the largest average C-index values ​​between the models in the variable selection submodule and the models in the model building submodule.

[0023] Furthermore, various ensemble machine learning algorithms include, but are not limited to, random forest, support vector machine, and logistic regression algorithms.

[0024] Furthermore, the assessment of prognostic risk for gastric cancer patients using the optimal model obtained after training includes:

[0025] We systematically quantified the gene expression data of patients with gastric cancer and the survival data of other patients, input the quantified data into the optimal model, and calculated the risk score; based on the risk score, we predicted the survival risk of patients.

[0026] Furthermore, the formula for the risk score is:

[0027]

[0028] Where RiskScore is the risk score, and i represents the data input to the optimal model. Let i be the expression profile of the i-th gene. Let be the weight of the i-th gene, and n represent the number of genes.

[0029] The present invention has the following advantages over the prior art:

[0030] (1) Traditional gastric cancer prognostic assessment models rely on a single data source or simple statistical analysis methods, resulting in poor accuracy and stability of prediction results. This invention improves the accuracy of survival outcome prediction by combining super-enhancer-related genes and multiple machine learning algorithms.

[0031] (2) There are differences in the collection of patient genetic information and survival data in multi-center cohort data, which are difficult to process uniformly by traditional models. This invention combines data standardization with multiple machine learning algorithms to effectively process gastric cancer data from different sources and ensure the generalization ability of the model. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0033] Figure 1 This is a structural block diagram of a machine learning-based super enhancer-related gastric cancer patient prognostic risk assessment system according to an embodiment of the present invention.

[0034] Figure 2 This refers to the C-index value of the machine learning algorithm combination model in this embodiment of the invention.

[0035] Figure 3 This refers to the AUC value of the prediction model in the training set of this invention.

[0036] Figure 4These are the AUC values ​​of the prediction models in the test set and external validation set of this embodiment of the invention.

[0037] Figure 5 Risk score calculation for embodiments of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be understood that the specific embodiments described are merely used to explain this application and are not intended to limit this application.

[0039] Example:

[0040] like Figure 1 As shown, this embodiment provides a machine learning-based super-enhancer-related prognostic risk assessment system for gastric cancer patients, mainly including a data acquisition module, a data processing module, a model training module, and a prognostic risk assessment module. The data acquisition module is used to collect historical data of gastric cancer patients, the data processing module is used to perform systematic quantitative analysis on the historical data of gastric cancer patients, the model training module is used to train multiple models using the data after systematic quantitative analysis, and the prognostic risk assessment module is used to assess the prognostic risk of gastric cancer patients using the optimal model obtained from the training of multiple models.

[0041] Specifically, the data acquisition module includes historical data of gastric cancer patients. This historical data includes chromatin immunoprecipitation sequencing (ChIP-seq) data, RNA sequencing (RNA-seq) data from a multicenter cohort of gastric cancer patients, and survival data.

[0042] ChIP-seq data were obtained through ChIP-seq experiments on gastric cancer tissues from patients, and consisted of H3K27ac histone peak data from gastric cancer patients. H3K27ac histone peak data is considered to be super-enhancer-specific modification and is used to identify gastric cancer-specific super-enhancers and their target genes, and to calculate weighted expression levels.

[0043] RNA-seq and survival data were obtained from the Cancer Genome Atlas Program (TCGA) database and the Gene Expression Omnibus (GEO) database, and were used to build prognostic models. Survival data included survival status and survival time. Survival status included survival or death; survival time was the duration of survival at the time of survival or death, in months.

[0044] Specifically, the data processing module is used for:

[0045] (1) Based on ChIP-seq data, a candidate gene set of gastric cancer-specific super enhancer target genes was obtained.

[0046] In this embodiment, the ChIP-seq data is H3K27ac histone peak data; the process of obtaining the candidate gene set of gastric cancer-specific super enhancer target genes based on the H3K27ac histone peak data is as follows:

[0047] For ChIP-seq data from gastric cancer tissue and adjacent normal tissue, peak calling was performed using MACS2 (v2.2.7) software to identify super-enhancers (SEs) (parameters set to: --qvalue 0.01 --broad --broad-cutoff 0.1). For SEs, super-enhancer regions were identified using the ROSE (Rank Ordering of Super-Enhancers) algorithm, with an association range of ±50kb to exclude interference from promoter regions. The super-enhancer regions were compared with the GENCODEv43 gene annotation file to screen for target genes with H3K27ac peaks within the region and a distance of <100kb from the transcription start site (TSS). Target genes located in low-complexity regions of the genome (such as telomeres and centromeres) were removed. Known housekeeping genes co-localized with super-enhancers were excluded, ultimately obtaining a candidate gene set of gastric cancer-specific super-enhancer target genes.

[0048] Gastric cancer-specific superenhancer target genes in the candidate gene set must meet the following two conditions: a) Gastric cancer-specific superenhancer target genes co-localize with the corresponding superenhancer regions (TSS±50kb); b) The expression level of gastric cancer-specific superenhancer target genes is significantly positively correlated with the signal intensity of the corresponding superenhancer (r>0.6, P<0.01).

[0049] (2) Based on RNA-seq data, survival data and candidate gene sets, the super enhancer prognostic genes of patients were obtained.

[0050] Survival status is categorized into 0 and 1 labels for survival and death, corresponding to the patient's survival outcome.

[0051] For STAD data in the TCGA database, univariate Cox proportional hazards regression analysis was performed on the candidate gene set using RNA-seq data and survival data. The hazard ratio (HR) of gene expression level to survival time must satisfy |log2(HR)|>0.5. After Benjamini-Hochberg correction, P < 0.01 was used as the condition for statistical significance to screen super enhancer prognostic genes.

[0052] This embodiment uses multivariate Cox proportional hazards regression analysis to remove the confounding effects of clinical information and screen for super enhancer prognostic genes that are independently associated with survival risk.

[0053] The expression profiles and survival data of super-enhancer prognostic genes were used as input data for the model, and the corresponding super-enhancer prognostic genes and their weights were used as output data. The input and output data constituted the sample, where the weights of the super-enhancer prognostic genes were... It is an unknown quantity.

[0054] In this embodiment, RNA-seq data from seven cohorts (GSE15459, GSE26901, GSE26899, GSE26253, GSE62254, and GSE84437) were randomly selected from the TCGA and GEO databases, with a sample size of n=1,972 (n being the number of patients). The samples corresponding to the STAD data in the TCGA database constituted the training set, while the samples corresponding to the cohorts in the GEO database constituted the validation and test sets.

[0055] Specifically, the model training module includes a variable selection submodule and a model building submodule. Both submodules are pre-loaded with a variety of existing ensemble machine learning algorithms, including random forest, support vector machine, logistic regression, etc.

[0056] The machine learning algorithm in the variable selection submodule is used to select samples, removing redundant features and retaining the most predictive samples.

[0057] The model is built using machine learning algorithms in the model building sub-module, and then optimized using reinforcement learning algorithms to improve the model's prediction accuracy.

[0058] That is, based on machine learning algorithms, it uses variable screening sub-modules and model building sub-modules to find the optimal model, builds machine learning models based on samples in the screened training set, and improves prediction accuracy through training and optimization.

[0059] In the variable selection submodule, based on the super enhancer prognostic genes and survival data in the samples, machine learning algorithms are used to select genes with certain predictive capabilities. In the model building submodule, based on the selected genes, machine learning algorithms are used to train the model: each machine learning algorithm corresponds to a certain number of selected genes from different groups.

[0060] The output of the model training module is the optimal model with the highest predictive power. The model training module optimizes the survival prediction objective function to find the optimal gene weight risk parameters, minimizing the overall prediction error while ensuring the model's effectiveness in screening prognostic variables. Survival prediction represents the predicted relative risk of the patient; a higher risk score indicates a greater risk of adverse prognosis.

[0061] The objective function is:

[0062]

[0063] in, is the loss term, used to measure the difference between the predicted survival and the actual survival outcome in each model; y represents the survival data in the sample. Let X be the feature mapping function, X be the expression profile of super-enhancer prognostic genes in the sample, n be the number of samples, and p be the number of variables (i.e., the number of super-enhancer prognostic genes in the input model). To control the parameters of the feature mapping, The sample gene weights to be estimated; This indicates that model complexity is controlled by a combination of L1 and L2 regularization to prevent overfitting. Here, λ=0: no regularization, the model relies entirely on the loss term to fit the data (potentially overfitting); λ>0: regularization is in effect, the larger λ is, the lower the model complexity (coefficients are compressed); α represents a real number between 0 and 1 (α∈[0,1]). This represents the sum of the absolute values ​​of gene weights; Let L1 norm be denoted as L1 norm, and let SUM be the sum of squared gene weights. γ is used to introduce additional constraints (such as variable selection rules, tree complexity, orthogonality). k Indicates the weight of the extended term. This refers to the regularization or constraint function defined according to the specific requirements of the machine learning algorithm, and is usually related to the model parameters. or feature mapping parameters Related.

[0064] The patient's predictive scoring formula is:

[0065]

[0066] Where RSi is the predicted score of sample i, a linear combination of the mapped features and gene weights; Let represent the original feature vector (dimension p×1) of the i-th sample; T represents the transpose of a matrix or vector to ensure that the dimensions of the multiplication between vectors match. This represents the optimized gene weights output by the model.

[0067] By adjusting parameters and function forms, a unified formula can cover a variety of machine learning methods:

[0068] ,

[0069] For each algorithm combination, the following process is executed:

[0070] Parameter tuning: Using n-1 samples in the training set, determine the optimal combination of hyperparameters through LOOCV;

[0071] Model training: using optimal parameters Reconstruct the model on the complete training set;

[0072] Prediction validation: Calculate the prediction score for the sample, repeat n times until all samples are covered.

[0073] The C-index is used to evaluate the predictive performance of a model and is suitable for survival analysis. It compares all possible sample pairs based on the predicted risk score to determine whether the model's predicted survival time and survival status are consistent with actual survival information.

[0074] The formula for calculating C-index is:

[0075]

[0076] in, This represents the predicted score for the i-th sample, generated by a built-in machine learning algorithm (such as Elastic Net, Lasso, etc.). This represents the survival time of the i-th sample.

[0077] II is an indicator function that indicates whether the condition is true or false:

[0078] II ( > ) indicates whether the predicted score of the i-th sample is greater than the predicted score of the j-th sample;

[0079] II ( > The expression indicates whether the survival time of the i-th sample is greater than that of the j-th sample.

[0080] II ( = ) indicates whether the predicted score of the i-th sample is equal to the predicted score of the j-th sample;

[0081] II ( = The expression indicates whether the survival time of the i-th sample is equal to the survival time of the j-th sample.

[0082] The summation of all unequal survival time pairs ensures that the compared sample pairs are meaningful (i.e., the survival times are unequal). The C-index reflects the consistency between the model's predicted ranking and the actual ranking; a higher C-index indicates better model performance.

[0083] The model with the highest average C-index value in the training and testing sets was selected as the optimal model. The input data of the optimal model were the super enhancer prognostic gene expression data and survival data of gastric cancer patients, and the output data were the target gene expression weights under the training of the optimal model, i.e., the risk genes.

[0084] The prognostic risk scoring system includes an optimal model, which is used to calculate the final risk score for a given gastric cancer patient and to predict survival.

[0085] Specifically, performance evaluations were performed to calculate time-dependent AUC (1 / 2 / 3-year survival rate).

[0086] The prognostic risk assessment module loads the final model parameters, including the selected gene variables with the greatest predictive power. The survival outcome prediction module takes as input the quantified target gene expression data of the patient to be predicted and outputs the predicted survival outcome of the patient to be predicted.

[0087] In one embodiment, all machine learning combinations are ranked and the C-index value of each model is calculated. CoxBoost is used for variable selection, and Lasso is used in the model building submodule. The average C-index value of the combination of these two algorithms is 0.6295, meaning CoxBoost+Lasso is considered the optimal combination. (See [link to relevant documentation]). Figure 2 .

[0088] Gene weight of each gene ( , , , ..., The gene weights are obtained through the training process of the machine learning model, and represent the contribution of each gene to the patient's prognostic risk.

[0089]

[0090] Where RiskScore represents the calculated final risk score, and i represents the risk gene input to the optimal model. This represents the expression level of the i-th gene in the patient to be predicted. represents the gene weight of the i-th gene, obtained from training the optimal model, and n represents the number of variables.

[0091] The risk score is calculated through the following steps:

[0092] Input data: Input the gene expression data of the patient to be predicted, including the expression levels of all prognostic-related variables. , , , ..., ), where n is the number of prognostic genes included in the super enhancer.

[0093] In this embodiment, n=6.

[0094] The optimal model is used to predict the survival outcome of a patient. The specific process is as follows: quantified relevant data of the patient to be predicted are input into the optimal model, and the model outputs the prediction result based on the patient's risk gene data.

[0095] The relevant data of the patients to be predicted include the expression of risk genes in the patients. The risk score is numerically quantified according to the rules. Furthermore, the calculation process of the risk score can be automated by a pre-trained machine learning model. In different datasets and patient groups, the risk score can be used to assess the risk level of the patients, reflecting the severity of the disease or the risk of death. That is, the lower the patient's risk score, the lower the risk of death, and the higher the patient's risk score, the higher the risk of death.

[0096] The following example illustrates the functionality implemented by the system described above:

[0097] In multiple gastric cancer cohorts in the TCGG and GEO databases (GSE15459, GSE26901, GSE26899, GSE26253, GSE62254, and GSE84437), the optimal model was used for prognostic evaluation. Based on the optimal model, the expression of risk genes (FAAH, RAMP1, SERPINE1, MAP4K4, FOXS1, and NGF) was extracted from patients in each cohort. The final risk score for each patient was calculated by combining the corresponding gene weights, and the patient's survival was predicted and compared with their actual survival.

[0098] Receiver operating characteristic (ROC) tests were performed on gastric cancer patients, and the area under the curve (AUC) was used to evaluate the model's performance in predicting 1-, 2-, and 3-year survival rates. AUC quantifies the model's discriminative power, representing the probability that the model correctly distinguishes between a randomly selected positive sample and a negative sample. An AUC value closer to 1.0 indicates stronger predictive performance. The optimal model in this embodiment demonstrated good prognostic predictive ability in multiple multicenter gastric cancer cohorts (see reference...). Figure 3 , Figure 4 It has a good and robust predictive level in terms of gastric cancer prognosis.

[0099] After quantifying and encoding the corresponding data of the patients to be assessed, the corresponding survival risk values ​​are input into the prognostic risk assessment module to obtain the patients' survival risk values. These values ​​are then compared with a preset threshold T. If RiskScore > T, the survival risk assessment result is "probable death"; if RiskScore ≤ T, the assessment result is "probable survival". Test results show that the patient survival risk values ​​calculated by this system can clearly distinguish between patients who are likely to die or those who are likely to survive. Taking Figure 2 as an example, the risk score T can be used as a cutoff value to distinguish between high-risk and low-risk patients. To improve prediction accuracy, the threshold can be adjusted according to actual needs (see Figure 5).

[0100] The system provided in this embodiment effectively improves the accuracy and generalization ability of gastric cancer survival outcome prediction. The data processing module constructs a standardized dataset based on multi-dimensional data of gastric cancer patients; the model training module ensures the efficient performance of selected variables in the model through variable selection and parameter optimization; and the optimal model is selected using the C-index value as an evaluation index, which effectively avoids overfitting and enhances the stability and robustness of the model.

[0101] The risk score calculation formula in the prognostic risk assessment module reflects the linear correlation between variable characteristics and survival risk. Through the training results of the machine learning model, gene weights can reflect the contribution of each variable to survival risk, and combined with the individual gene expression of the patient to be predicted, a precise survival risk value is quantified.

[0102] The system provided in this embodiment has high scalability and adaptability, can be optimized and adjusted for different gastric cancer patient groups, and demonstrates good applicability in multi-center cohorts. In particular, the prognostic risk assessment module can provide reliable probability assessments of patient survival at different time periods, which has significant clinical application value.

[0103] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A machine learning-based system for assessing the prognostic risk of gastric cancer patients with super-enhancer associations, characterized in that, It includes a data acquisition module, a data processing module, a model training module, and a prognostic risk assessment module, among which, The data acquisition module is used to collect historical data of gastric cancer patients, including chromatin immunoprecipitation sequencing data, RNA sequencing data and survival data of gastric cancer patients in a multicenter cohort. The data processing module is used to perform systematic quantitative analysis of historical data from gastric cancer patients. The model training module is used to train multiple machine learning models using data after systematic quantitative analysis. The prognostic risk assessment module is used to assess the prognostic risk of gastric cancer patients using the optimal model obtained after training. It performs systematic quantitative analysis on the gene expression data of the gastric cancer patients to be predicted and the survival data of other patients. The quantitatively analyzed data is input into the optimal model to calculate the risk score and predict the patient's survival risk based on the risk score. Specifically, the data processing module is used to obtain a candidate gene set of gastric cancer-specific super enhancer target genes based on chromatin immunoprecipitation sequencing data, and to obtain the patient's super enhancer prognostic genes based on RNA sequencing data, survival data, and the candidate gene set. The process of obtaining super-enhancer prognostic genes for patients based on RNA sequencing data, survival data, and candidate gene sets includes: using RNA sequencing data and survival data, performing univariate Cox proportional hazards regression analysis on gastric cancer-specific super-enhancer target genes in the candidate gene set; the hazard ratio of gene expression level to survival time must satisfy |log2(HR)|>0.5; after Benjamini-Hochberg correction, P<0.01 is used as the condition for statistical significance to screen super-enhancer prognostic genes. The model training module is based on machine learning algorithms. It uses a variable selection submodule combined with a model building submodule to find the optimal model and builds a machine learning model based on the samples in the selected training set. The model training module optimizes the survival prediction of the objective function to find the optimal gene weight risk parameters, so as to minimize the overall prediction error and ensure that the model has a good screening effect on prognosis-related variables.

2. The super-enhancer-associated gastric cancer patient prognostic risk assessment system according to claim 1, characterized in that, Survival data includes survival status and survival time. Survival status includes whether the individual is alive or dead.

3. The super-enhancer-associated gastric cancer patient prognostic risk assessment system according to claim 2, characterized in that, The candidate gene set of gastric cancer-specific super enhancer target genes obtained based on chromatin immunoprecipitation sequencing data includes: For chromatin immunoprecipitation sequencing data, peak calling was performed using MACS2 software to identify super enhancers (SEs). For super enhancer SEs, the ROSE algorithm is used to identify super enhancer regions in order to eliminate interference from starter regions; The super enhancer regions were compared with the GENCODE v43 gene annotation file to screen out target genes with H3K27ac peaks in the regions and whose distance from the gene transcription start site was less than a specified value. Target genes located in low-complexity regions of the genome were removed, and housekeeping genes co-localized with super enhancers were excluded. Finally, a candidate gene set of gastric cancer-specific super enhancer target genes was obtained. Among them, gastric cancer-specific super enhancer target genes in the candidate gene set must meet the following conditions: Gastric cancer-specific superenhancer target genes co-localize with their corresponding superenhancer regions; The expression levels of gastric cancer-specific superenhancer target genes are significantly positively correlated with the corresponding superenhancer signal intensity.

4. The super-enhancer-associated gastric cancer patient prognostic risk assessment system according to any one of claims 2 to 3, characterized in that, The expression profiles and survival data of super-enhancer prognostic genes are used as input data for the model, and the corresponding super-enhancer prognostic genes and their weights are used as output data for the model. The input data and output data constitute the sample. The model training module includes a variable selection submodule and a model building submodule. Both submodules are pre-configured with a variety of machine learning algorithms. The variable selection submodule is used to remove redundant features from the samples and retain the most predictive features using any machine learning algorithm. The model building submodule is used to train the model corresponding to any machine learning algorithm using samples with preserved features, and select the optimal model from the trained models; wherein, the optimal model is the two models with the largest average C-index values ​​of the models in the variable selection submodule and the models in the model building submodule.

5. The super-enhancer-associated gastric cancer patient prognostic risk assessment system according to claim 4, characterized in that, Various ensemble machine learning algorithms include random forest, support vector machine, and logistic regression.

6. The super-enhancer-associated gastric cancer patient prognostic risk assessment system according to claim 4, characterized in that, The method of using the optimal model obtained after training to assess the prognostic risk of gastric cancer patients includes: A systematic quantitative analysis of gene expression data for predicting gastric cancer patients was conducted. The quantitatively analyzed data was then input into the optimal model to calculate a risk score. Based on the risk score, the patient's survival risk was predicted.

7. The super-enhancer-associated gastric cancer patient prognostic risk assessment system according to claim 6, characterized in that, The formula for the risk score is: Where RiskScore is the risk score, and i represents the data input to the optimal model. Let i be the expression profile of the i-th gene. Let be the weight of the i-th gene, and n represent the number of genes.

Images