A screening method for biomarkers for predicting the prognosis of esophageal squamous cell carcinoma

By obtaining candidate molecules from bioinformatics databases, constructing interaction networks, and using prognostic prediction models for multiple screenings, the problem of low reproducibility in esophageal squamous cell carcinoma prognostic prediction models was solved, and the prediction accuracy was improved.

CN112397153BActive Publication Date: 2024-11-08THE FIRST AFFILIATED HOSPITAL OF HENAN UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011295497.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-11-18
Publication Date
2024-11-08
Estimated Expiration
2040-11-18

AI Technical Summary

Technical Problem

The molecular heterogeneity, specimen source, tissue processing and detection technology differences in existing esophageal squamous cell carcinoma prognostic prediction models lead to low reproducibility of results, affecting prediction accuracy.

Method used

Candidate molecules were obtained from bioinformatics databases, and an interaction network was constructed for the first screening. A prognostic prediction model was used for the second screening to identify biomarkers for predicting the prognosis of esophageal squamous cell carcinoma. These biomarkers were then classified and screened using multiple models.

Benefits of technology

This improved the reliability and accuracy of the screening results, ensuring that the selected biomarkers could more accurately predict the prognosis of esophageal squamous cell carcinoma.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112397153B_ABST
    Figure CN112397153B_ABST
Patent Text Reader

Abstract

The present application relates to the field of biotechnology, and specifically discloses a method for screening biomarkers for predicting the prognosis of esophageal squamous cell carcinoma, the method comprising: searching a bioinformation database according to the identification information of esophageal squamous cell carcinoma to obtain a candidate molecules associated with esophageal squamous cell carcinoma; obtaining the gene expression values ​​of a candidate molecules; constructing an interaction network of a candidate molecules according to the gene expression values; performing a first screening on multiple candidate molecules based on the interaction network to filter some candidate molecules, and retaining b candidate molecules that meet the preset conditions; performing a second screening on the b candidate molecules retained after the first screening based on the prognosis prediction model to obtain c preferred molecules; and screening out biomarkers for predicting the prognosis of esophageal squamous cell carcinoma from the c preferred molecules. Through the above method, the credibility of the screening results is higher, and the accuracy of the screening results is higher.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of biotechnology, and in particular to a method for screening a biomarker for predicting the prognosis of esophageal squamous cell carcinoma, a computer-readable storage medium, and a biomarker for predicting the prognosis of esophageal squamous cell carcinoma. Background Art

[0002] Cancer is a disease that is extremely harmful to humans. According to relevant epidemiological data, there are approximately 11 million new cases of cancer each year. Among many cancers, the incidence of esophageal cancer ranks eighth among all malignant tumors, and the mortality rate ranks sixth. About 45,000 people suffer from the disease each year worldwide. Unlike Western countries where most cases are esophageal adenocarcinoma, esophageal squamous cell carcinoma (ESCC, hereinafter referred to as esophageal squamous cell carcinoma or ESCC) accounts for more than 95% of the overall incidence of esophageal cancer in my country. The main treatments for esophageal squamous cell carcinoma are currently surgery, radiotherapy, and chemotherapy. Although the comprehensive treatment of esophageal squamous cell carcinoma has been continuously improved in the past decade, its prognosis is still not ideal. Currently, its five-year overall survival rate is only 15%-25%.

[0003] During the long-term research, the inventors of this application found that the low reproducibility between different clinical studies is the biggest challenge faced by existing research due to differences in tumor molecular heterogeneity, specimen sources, tissue processing, detection technology, data analysis, etc. Therefore, exploring and studying biomarkers for the prognosis of esophageal squamous cell carcinoma and establishing a relatively accurate prognosis prediction model have extremely special clinical significance in my country. Summary of the invention

[0004] In view of the differences in the molecular heterogeneity of tumors, specimen sources, tissue processing, detection technology, data analysis, etc. in the prior art, as well as the low reproducibility of different clinical experimental results, the present application provides a screening method for biomarkers for predicting the prognosis of esophageal squamous cell carcinoma, a computer-readable storage medium, and a biomarker for predicting the prognosis of esophageal squamous cell carcinoma.

[0005] In a first aspect, the present application provides a method for screening biomarkers for predicting the prognosis of esophageal squamous cell carcinoma, the method comprising: searching a bioinformation database according to the identification information of the esophageal squamous cell carcinoma to obtain a candidate molecules associated with the esophageal squamous cell carcinoma; obtaining gene expression values ​​of the a candidate molecules; constructing an interaction network of the a candidate molecules according to the gene expression values; performing a first screening on multiple candidate molecules based on the interaction network to filter out some of the candidate molecules, and retaining b candidate molecules that meet preset conditions, wherein b<a; performing a second screening on the b candidate molecules retained after the first screening based on a prognosis prediction model to obtain c preferred molecules, wherein c<b; and screening out the biomarker for predicting the prognosis of esophageal squamous cell carcinoma from the c preferred molecules.

[0006] In a second aspect, the present application provides a computer-readable storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions can be executed by a processor to implement the aforementioned method.

[0007] In a third aspect, the present application provides a biomarker for predicting the prognosis of esophageal squamous cell carcinoma, wherein the biomarker is screened by the aforementioned method.

[0008] Advantages and beneficial effects of the present application: Different from the prior art, the present application obtains candidate molecules associated with esophageal squamous cell carcinoma from the bioinformatics database, which basically covers all candidate molecules associated with esophageal squamous cell carcinoma. At the same time, the present application performs a first screening of multiple candidate molecules based on the interaction network of the candidate molecules to filter out some candidate molecules, and performs a second screening of the candidate molecules retained after the first screening based on the prognosis prediction model to obtain preferred molecules, and then screens out biomarkers for predicting the prognosis of esophageal squamous cell carcinoma from the preferred molecules, and performs classification screening through multiple models, so that the screening results are more credible and the screening results are more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. 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 creative work. Among them:

[0010] Figure 1 This is a flow chart of the first embodiment of the method for screening biomarkers for predicting the prognosis of esophageal squamous cell carcinoma of the present application;

[0011] Figure 2 This is a flow chart of the second embodiment of the method for screening biomarkers for predicting the prognosis of esophageal squamous cell carcinoma of the present application;

[0012] Figure 3 This is a flow chart of the third embodiment of the method for screening biomarkers for predicting the prognosis of esophageal squamous cell carcinoma of the present application;

[0013] Figure 4 This is a flow chart of a fourth embodiment of the method for screening biomarkers for predicting the prognosis of esophageal squamous cell carcinoma of the present application;

[0014] Figure 5 This is a flow chart of the fifth embodiment of the method for screening biomarkers for predicting the prognosis of esophageal squamous cell carcinoma of the present application;

[0015] Figure 6 It is the interaction network diagram of 16 candidate molecules;

[0016] Figure 7 is the probability that the 17 candidate molecules are included in the candidate molecule combination in the five classifier algorithms;

[0017] Figure 8 This is the survival analysis result of SFN in 179 gene expression profile samples (GEO number GSE53625);

[0018] Fig. 9 This is the survival analysis result of SFN in 37 cases of TCGA-ESCC data set;

[0019] Fig.10 This is the survival analysis result of SFN in 89 independent experimental samples. DETAILED DESCRIPTION

[0020] 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.

[0021] In the description of the present application, it should be noted that the terms "upper", "lower", "inner" and "outer" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific position, be constructed and operated in a specific position, and therefore cannot be understood as limiting the present application. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0022] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0023] The present application proposes a method for screening biomarkers for predicting the prognosis of esophageal squamous cell carcinoma. It should be noted that biomarkers refer to molecules related to the disease. For example, biomarkers can be molecules related to the presence, stage, prognosis, etc. of esophageal squamous cell carcinoma. Specifically, esophageal squamous cell carcinoma biomarkers can include proteins that are differentially expressed in esophageal squamous cell carcinoma subjects (such as full-length polypeptides, splice variants, post-translationally modified polypeptides, etc.) as well as fragments of gene products and corresponding polynucleotide sequences, such as mRNA, DNA, etc.

[0024] like Figure 1 As shown, the method comprises the following steps:

[0025] S10: searching a biological information database according to the identification information of esophageal squamous cell carcinoma to obtain a candidate molecules associated with esophageal squamous cell carcinoma.

[0026] Specifically, the bioinformatics database includes Chinese and English academic databases, which may include China National Knowledge Infrastructure database, Wanfang database, Web of Science NCBI database. Literature related to ESCC was retrieved from the above Chinese and English academic databases, and 48 candidate molecules associated with ESCC were sorted out, as shown in Table 1.

[0027] Table 1 48 candidate molecules associated with ESCC

[0028]

[0029]

[0030] S20: Obtain gene expression values ​​of a candidate molecules.

[0031] Specifically, the gene expression values ​​of the above 48 candidate molecules associated with ESCC were obtained.

[0032] S30: Construct an interaction network of a candidate molecules based on the gene expression values.

[0033] Specifically, the software tool NetBox was used to construct an interaction network of a candidate molecules.

[0034] NetBox is a Java-based software tool that can be used to perform human protein interaction network analysis. The tool is based on the human interaction network (HIN), which consists of four data sources: HPRD, Reactome, NCI-Nature PID database, and MSKCC Cancer Cell Map. It can connect genes into a network through some calculations, identify significant "linker" genes, and divide the network into modules.

[0035] S40: Performing a first screening on the plurality of candidate molecules based on the interaction network to filter out some of the candidate molecules and retaining b candidate molecules that meet the preset conditions, wherein b<a.

[0036] The interaction network of multiple candidate molecules was constructed based on NetBox software, wherein the parameters of the interaction network included: shortest path threshold = 1, P value threshold < 0.05, and the candidate molecules with interactions less than the preset value among the 48 candidate molecules were filtered out, and then 17 candidate molecules that met the interaction network parameters were obtained, as shown in Table 2 for details.

[0037] Table 2 17 candidate molecules with interactions equal to or less than the preset value

[0038] CCNA2 CD44 MDM2 TRAM1 RRM2B CCND1 EGFR MLH1 RAC3 SFN BRCA1 CDKN2A PTGS2 PIK3CA TP53 VEGFA RAD51

[0039] It should be noted that when the 17 candidate molecules are input into STRING( https: / / string-db.org / ) website, and after setting the molecular relationship score to 0.7, the candidate molecules that are not strongly related to other candidate molecules are filtered out, and the following is obtained: Figure 6 The 16 candidate molecules are shown.

[0040] S50: Based on the prognosis prediction model, a second screening is performed on the b candidate molecules retained after the first screening to obtain c preferred molecules, where c<b.

[0041] Specifically, 17 candidate molecules were randomly combined to obtain 2 17 -1=131071 candidate molecule combinations. The above 131071 candidate molecule combinations are screened based on the prognosis prediction model to obtain several preferred candidate molecule combinations. The occurrence frequencies of the 17 candidate molecules in several candidate molecule combinations are counted, and c preferred molecules with the highest occurrence frequencies are selected.

[0042] S60: Screen out biomarkers for predicting the prognosis of esophageal squamous cell carcinoma from c priority molecules.

[0043] Specifically, the c priority molecules screened in the above step S50 are further screened to obtain biomarkers for predicting the prognosis of esophageal squamous cell carcinoma.

[0044] Furthermore, based on independent validation experiments, independent validation was performed in several gene expression profile samples. Precise quantitative PCR experiments were performed on the biomarkers to detect the gene expression values ​​of the biomarkers in esophageal squamous cell carcinoma tissue samples and adjacent tissue samples, and the difference in gene expression values ​​or the multiple of the difference in gene expression values ​​between the biomarkers in esophageal squamous cell carcinoma tissue samples and adjacent tissue samples were calculated. Survival analysis was performed using gene expression values ​​as factors, and the above samples were divided into a high-risk (High) group and a low-risk (Low) group according to the difference in SFN gene expression values. The specific survival analysis results are shown below. Figure 8-10 shown.

[0045] Different from the prior art, the present application obtains candidate molecules associated with esophageal squamous cell carcinoma from the bioinformatics database, which basically covers all candidate molecules associated with esophageal squamous cell carcinoma. At the same time, the present application performs a first screening of multiple candidate molecules based on the interaction network of the candidate molecules to filter out some candidate molecules, and performs a second screening of the candidate molecules retained after the first screening based on the prognosis prediction model to obtain preferred molecules, and then screens out biomarkers for predicting the prognosis of esophageal squamous cell carcinoma from the preferred molecules, and performs classification screening through multiple models, so that the screening results are more credible and the screening results are more accurate.

[0046] like Figure 2 As shown, in one embodiment, step S20 includes the following steps:

[0047] S21: obtaining several pairs of gene expression profiles from the GEO database (Gene Expression Omnibus database), each pair of gene expression profiles including a gene expression profile of an esophageal squamous cell carcinoma tissue sample and a gene expression profile of an adjacent cancer tissue sample.

[0048] Specifically, in this step, the molecular chip data information of the same esophageal squamous cell carcinoma patient cohort (GEO number GSE53625) was retrieved from the GEO database (https: / / www.ncbi.nlm.nih.gov / geo / ), and 179 pairs of gene expression profiles of esophageal squamous cell carcinoma tissue samples and adjacent tissue samples from the same esophageal squamous cell carcinoma patient were extracted from the molecular chip data information, that is, 179 gene expression profile samples.

[0049] S22: Align the gene probe sequences of the chip platform to the human reference genome (GRCh37) to re-annotate the chip platform.

[0050] Specifically, the chip platform is the microarray chip platform GPL18109. The gene annotation method based on sequence similarity is adopted. The human reference genome (GRCh37) in the Gencode database is used as the genome annotation file. Gencode and SeqMap are used to align the gene probe sequences of the chip platform to the human reference genome (GRCh37), thereby completing the re-annotation of the chip platform.

[0051] S23: Extracting gene expression values ​​of corresponding a candidate molecules from a number of pairs of gene expression profiles according to the re-annotated chip platform.

[0052] Specifically, the gene expression profile of the GEO database (GEO number is GSE53625) in step S21 is searched according to the re-annotated chip platform, and then the corresponding gene expression values ​​of the above 48 candidate factors are extracted.

[0053] In one embodiment, the gene expression value of each candidate molecule in the above step S23 is the difference between the gene expression value of the corresponding esophageal squamous cell carcinoma tissue sample and the gene expression value of the adjacent cancer tissue sample.

[0054] Specifically, the difference between the gene expression value of the esophageal squamous cell carcinoma tissue sample and the gene expression value of the adjacent cancer tissue sample is used as the gene expression value of the candidate molecule, and the gene expression value of the candidate molecule is used as the input data for all subsequent calculations.

[0055] like Figure 3 As shown, in one embodiment, before step S50, the method further includes:

[0056] S70: Randomly combine the b candidate molecules retained after the first screening to obtain m candidate molecule combinations.

[0057] Randomly combine the 17 candidate molecules in Table 2 to obtain 2 17 -1=131071 candidate molecule combinations.

[0058] S80: Use multiple classifier algorithms to construct m models corresponding to m candidate molecular combinations respectively.

[0059] Among them, the classifier algorithms include at least: logistic regression algorithm (LR), support vector machine algorithm (SVM), artificial neural network algorithm (ANN), random forest algorithm (RF) and extreme gradient boosting algorithm (XGBoost).

[0060] Among them, LR, SVM and ANN are weak classifier algorithms, and RF and XGBoost are strong classifier algorithms. The above five classifier algorithms are used to construct 131,071 models corresponding to the above 131,071 candidate molecular combinations.

[0061] S90: Obtain a training sample set and a test sample set, wherein the training sample set includes e pairs of gene expression profiles, and the test sample set includes the remaining f pairs of gene expression profiles.

[0062] Specifically, the 179 gene expression profile samples in step S21 are divided into a training sample set and a test sample set, wherein the training sample set includes 134 gene expression profile samples, and the test sample set includes the remaining 45 gene expression profile samples.

[0063] S100: Taking whether the survival period is greater than a preset time as a label, and using cross-validation and parameter optimization methods, m models are trained according to the training sample set to obtain n candidate models with the best model quality parameters in various classifier algorithms.

[0064] Specifically, for the above 131071 models, the cross-validation method was used to optimize the parameters and establish the models in the training sample set (134 gene expression profile samples). Among them, when the model quality parameter is the area under the receiver operating characteristic curve (AUC), if the AUC of a certain model is greater than the average value of the AUC, the model is determined as the candidate model with the best model quality parameter, thereby obtaining n candidate models in various classifier algorithms, where 1000<n<131071, and n is an integer.

[0065] S110: For each candidate model, each pair of gene expression profiles in the test sample set is input into the candidate model to obtain a prognosis prediction model with the best p model quality parameters among various classifier algorithms.

[0066] When the model quality parameter is the area under the receiver operating characteristic curve (AUC), the candidate models ranked in the top 1000 by AUC are determined as the prognosis prediction models in various classifier algorithms in descending order of the quality parameters, that is, the prognosis prediction models with the best 1000 quality parameters are obtained.

[0067] like Figure 4 As shown, in one embodiment, step S50 includes:

[0068] S51: Counting the number of occurrences of each candidate molecule in the p candidate molecule combinations corresponding to the p prognosis prediction models, and selecting c candidate molecules with the largest number of occurrences as preferred molecules.

[0069] Specifically, the occurrence probability of each candidate molecule in the 1000 feature combinations corresponding to the above 1000 prognostic prediction models with the best quality parameters was counted, and the five candidate molecules with the highest occurrence probability in each classifier algorithm were selected as the preferred molecules of the corresponding classifier algorithm. For details, see Table 3 and Figure 7 .

[0070] Table 3 Preferred molecules of each classifier algorithm

[0071]

[0072] Step S60 includes:

[0073] S61: Take the intersection of c priority molecules in all classifier algorithms to screen out biomarkers for predicting the prognosis of esophageal squamous cell carcinoma.

[0074] Specifically, Figure 6 As shown in the figure, among the five classifier algorithms (LR, SVM, ANN, RF, and XGBoost), LR screened out two preferred molecules that the other four classifier algorithms did not screen out: TP53 and EGFR; SVM screened out one preferred molecule that the other four classifier algorithms did not screen out: TRAM1; ANN screened out one preferred molecule that the other four classifier algorithms did not screen out: CCND1; LR and RF both screened out one preferred molecule that the other three classifier algorithms did not screen out: RAC3; RF and ANN both screened out one preferred molecule that the other three classifier algorithms did not screen out Selected the preferred molecule: MDM2; RF and XGBoost both screened out one preferred molecule that the other three classifier algorithms did not screen out: PIK3CA; XGBoost and SVM both screened out one preferred molecule that the other three classifier algorithms did not screen out: VEGFA; SVM, ANN, RF, and XGBoost all screened out LR but did not screen out the preferred molecule: PTGS2; LR, SVM, ANN, and XGBoost all screened out RF but did not screen out the preferred molecule: CD44; and the above five classifier algorithms all screened out the same preferred molecule: SFN. Therefore, SFN was determined to be a biomarker for predicting the prognosis of esophageal squamous cell carcinoma.

[0075] After the above steps, SFN was finally determined as a biomarker for predicting the prognosis of esophageal squamous cell carcinoma.

[0076] Furthermore, independent verification was performed in several gene expression profile samples based on independent verification experiments. Precise quantitative PCR experiments were performed on SFN to detect the gene expression values ​​of SFN in esophageal squamous cell carcinoma tissue samples and paracancerous tissue samples, and the difference or multiple of the gene expression values ​​of SFN in esophageal squamous cell carcinoma tissue samples and paracancerous tissue samples were calculated. Survival analysis was performed using SFN as a factor.

[0077] In addition, the embodiment of the present application also proposes a biomarker for predicting the prognosis of esophageal squamous cell carcinoma, and the biomarker is screened by the screening method for biomarkers for predicting the prognosis of esophageal squamous cell carcinoma as described above.

[0078] Specifically, the biomarker is SFN, and SFN is the mRNA expression level of the gene or the protein expression level thereof.

[0079] It should be noted that the model construction in this application is carried out in R version 3.6.3 (http: / / www.r-project.org). Among them, the weak classifier uses the following software packages: bestglm, e1071 and nnet; the ensemble learning algorithm uses the following software packages: randomForest and xgboost; the survival analysis uses the following software package: survminer.

[0080] In addition, an embodiment of the present application also proposes a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the steps of the biomarker screening method for predicting the prognosis of esophageal squamous cell carcinoma in the above embodiment are implemented.

[0081] In conclusion, the present application obtains candidate molecules associated with esophageal squamous cell carcinoma from the bioinformatics database, which basically covers all candidate molecules associated with esophageal squamous cell carcinoma. At the same time, the present application performs a first screening of multiple candidate molecules based on the interaction network of the candidate molecules to filter some candidate proteins, and performs a second screening of the candidate molecules retained after the first screening based on the prognosis prediction model to obtain preferred molecules, and then screens out biomarkers for predicting the prognosis of esophageal squamous cell carcinoma from the preferred molecules, and performs classification screening through multiple models, so that the screening results are more credible and the screening results are more accurate.

[0082] The above description is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for screening biomarkers for predicting the prognosis of esophageal squamous cell carcinoma, characterized in that: The method comprises: Searching a biological information database according to the identification information of the esophageal squamous cell carcinoma to obtain a candidate molecules associated with the esophageal squamous cell carcinoma; Obtaining gene expression values ​​of a candidate molecules; According to the gene expression values, construct an interaction network of a candidate molecules ; Perform a first screening of multiple candidate molecules based on the interaction network , To filter part of the candidate molecules and retain b candidate molecules that meet the preset conditions, wherein b < a; Randomly combining the b candidate molecules retained after the first screening to obtain m candidate molecule combinations; Using a plurality of classifier algorithms to construct m models corresponding to the m candidate molecule combinations respectively, wherein the classifier algorithms at least include: a logistic regression algorithm, a support vector machine algorithm, an artificial neural network algorithm, a random forest algorithm, and an extreme gradient boosting algorithm; Obtaining a training sample set and a test sample set, wherein the training sample set includes e pairs of the gene expression profiles, and the test sample set includes the remaining f pairs of the gene expression profiles; Taking whether the survival period is greater than a preset time as a label, and adopting a cross-validation and parameter optimization method, the m models are trained according to the training sample set to obtain the candidate models with the best n model quality parameters in various classifier algorithms; For each candidate model, each pair of the gene expression profiles in the test sample set is input into the candidate model to obtain the prognosis prediction model with the best p model quality parameters among various classifier algorithms; Based on the prognosis prediction model, the b candidate molecules retained after the first screening are screened for a second time to obtain c preferred molecules, wherein c < b; The biomarkers for predicting the prognosis of esophageal squamous cell carcinoma are screened out from the c priority molecules.

2. The screening method according to claim 1, characterized in that The step of obtaining the gene expression values ​​of the plurality of candidate molecules comprises: Acquire several pairs of gene expression profiles in the GEO database, each pair of gene expression profiles comprising a gene expression profile of an esophageal squamous cell carcinoma tissue sample and a gene expression profile of an adjacent tissue sample; Aligning the gene probe sequences of the chip platform to the human reference genome to re-annotate the chip platform; The gene expression values ​​of the corresponding a candidate molecules are extracted from several pairs of the gene expression profiles according to the re-annotated chip platform.

3. The screening method according to claim 2, characterized in that The gene expression value of each candidate molecule is the difference between the gene expression value of the corresponding esophageal squamous cell carcinoma tissue sample and the gene expression value of the adjacent cancer tissue sample.

4. The screening method according to claim 1, characterized in that The step of constructing an interaction network of a plurality of candidate molecules according to the gene expression values ​​comprises: An interaction network of the plurality of candidate molecules is constructed based on NetBox software, wherein the parameters of the interaction network include: shortest path threshold = 1, P value threshold < 0.

05.

5. The screening method according to claim 1, characterized in that When the model quality parameter is the area under the receiver operating characteristic curve (AUC), the n models whose AUC is greater than the average value of the AUC are determined as the candidate models with the best model quality parameters among the various classifier algorithms; When the model quality parameter is the area under the receiver operating characteristic curve (AUC), the p candidate models with the largest AUC are arranged in descending order and determined as the prognosis prediction model with the best model quality parameter among the various classifier algorithms.

6. The screening method according to claim 5, characterized in that Based on the prognosis prediction model, the b candidate molecules retained after the first screening are screened for the second time, and c preferred molecules are obtained, including: Counting the number of occurrences of each of the candidate molecules in the p candidate molecule combinations corresponding to the p prognosis prediction models in the various classifier algorithms, and selecting c candidate molecules with the largest number of occurrences as the preferred molecules in the various classifier algorithms; The biomarkers for predicting the prognosis of esophageal squamous cell carcinoma screened out from the c priority molecules include: The intersection of the c preferred molecules in all the classifier algorithms is taken to screen out the biomarker for predicting the prognosis of esophageal squamous cell carcinoma.

7. A computer-readable storage medium, characterized in that: Computer-readable instructions are stored thereon, and the computer-readable instructions can be executed by a processor to implement the method according to any one of claims 1 to 6.