Biological marker, model for predicting recurrence of esophageal squamous cell carcinoma and construction method thereof

Abstract

The application relates to the technical field of biological medicine, and discloses a biomarker, a model and a construction method for predicting recurrence of esophageal squamous cell carcinoma, wherein the biomarker comprises TRIM37, LLPH, SOX17, LINC02694, SERPINH1, TP53TG5, ANGPT2, ACTRT2, HECTD4, P2RY1 and LINC01102. The biomarker can effectively predict recurrence of the esophageal squamous cell carcinoma, and the sensitivity and specificity values are both high. According to the biomarker, the recurrence can be predicted, and the advantages of safety, non-invasiveness, easy-to-obtain sample, high accuracy and convenient operation are achieved, and accurate judgment is provided for recurrence prediction of the esophageal squamous cell carcinoma.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, specifically to a biomarker, model, and construction method for predicting recurrence of esophageal squamous cell carcinoma. Background Technology

[0002] Esophageal squamous cell carcinoma (ESCC) is the main pathological type of esophageal cancer, accounting for approximately 90% of esophageal cancer patients in my country. Recurrence is a significant factor affecting patient prognosis. The recurrence rate after radical resection of esophageal squamous cell carcinoma is high, ranging from 30.3% to 58.0%, resulting in a poor overall prognosis for patients with this type of cancer, with a 5-year overall survival rate of approximately 20%. Therefore, early and accurate prediction of recurrence is crucial.

[0003] Currently, methods for predicting esophageal cancer recurrence mainly include clinical parameters, biomarkers, genomics, and imaging techniques. Clinical parameters, such as tumor size, lymph node metastasis, and tumor stage, form the basis for assessing recurrence risk. Biomarkers, such as CEA and SCC, can serve as auxiliary indicators of tumor progression and recurrence. Imaging techniques, such as CT, MRI, and PET-CT, can monitor changes in tumor volume and metabolic activity. However, existing methods for predicting esophageal squamous cell carcinoma recurrence have some significant drawbacks and limitations. First, while clinical parameters are readily available, they often lack sufficient sensitivity and specificity, making it difficult to accurately predict the recurrence risk of individual patients. Second, although imaging techniques can provide intuitive tumor information, they are often limited by resolution and contrast, making it difficult to detect small recurrent lesions. Furthermore, the invasiveness and cost of imaging examinations also limit their application in recurrence monitoring. Therefore, it is essential to discover an accurate, reliable, and clinically applicable biomarker to predict the recurrence of esophageal squamous cell carcinoma in advance. Summary of the Invention

[0004] To address the aforementioned technical shortcomings, the purpose of this invention is to provide biomarkers, models, and construction methods for predicting the recurrence of esophageal squamous cell carcinoma. These biomarkers can effectively predict the recurrence of esophageal squamous cell carcinoma with high sensitivity and specificity. Recurrence prediction based on these biomarkers has the advantages of being safe and non-invasive, having readily available samples, high accuracy, and convenient operation, thus providing accurate judgment for the prediction of esophageal squamous cell carcinoma recurrence.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides a biomarker for predicting recurrence of esophageal squamous cell carcinoma, the biomarker including TRIM37, LLPH, SOX17, LINC02694, SERPINH1, TP53TG5, ANGPT2, ACTRT2, HECTD4, P2RY1 and LINC01102.

[0006] Secondly, the present invention provides uses of the biomarker, the uses including: constructing a model for predicting recurrence of esophageal squamous cell carcinoma, and preparing a product for predicting recurrence of esophageal squamous cell carcinoma, the product being based on the model for predicting recurrence of esophageal squamous cell carcinoma to predict recurrence of esophageal squamous cell carcinoma.

[0007] Thirdly, the present invention provides a model for predicting the recurrence of esophageal squamous cell carcinoma, wherein the input variable of the model is the content of the biomarker.

[0008] Furthermore, the method for determining the content of the biomarker is 5hmC high-throughput detection.

[0009] Fourthly, the present invention provides a method for constructing a model to predict the recurrence of esophageal squamous cell carcinoma, comprising the following steps: Step 1. Samples from multiple patients with esophageal squamous cell carcinoma were analyzed to obtain 5hmC sequencing data of DNA; Step 2. Perform the sequencing data obtained in Step 1 through first filtering, screening, and second filtering to obtain biomarkers for predicting recurrence of esophageal squamous cell carcinoma. Step 3. Input the biomarker data obtained in Step 2 into the machine learning model, train the model, store the trained model, and obtain a model for predicting the recurrence of esophageal squamous cell carcinoma.

[0010] Furthermore, the first filtering includes: removing peak information that appears only in 10 or fewer samples; The screening process includes: comparing sequencing data from different samples using DEseq2 software, retaining 5hmC peak regions with read counts greater than 50, and identifying differentially regulated biomarkers of 5hmC based on fold change >= 0.5 and p-value < 0.01. The second filtering includes: filtering the differentially expressed biomarkers obtained through the screening using the recursive feature elimination algorithm in Scikit-Learn to obtain biomarkers for predicting recurrence of esophageal squamous cell carcinoma: TRIM37, LLPH, SOX17, LINC02694, SERPINH1, TP53TG5, ANGPT2, ACTRT2, HECTD4, P2RY1, and LINC01102; Furthermore, the sample is plasma.

[0011] Furthermore, the second filter employs a logistic regression cross-validator, wherein the class weights of the logistic regression cross-validator are set to balanced mode, the number of cross-validation folds is set to 2, the maximum number of iterations is set to 1000, and the scoring metric is set to accuracy.

[0012] Furthermore, in step 3, the machine learning model includes: training a logistic regression CV model.

[0013] The beneficial effects of this invention are as follows: (1) The biomarkers of the present invention can effectively predict the recurrence of esophageal squamous cell carcinoma in advance.

[0014] (2) The prediction model of the present invention has the advantages of high specificity and high sensitivity. By applying the biomarkers and / or models described in the present invention, safe, non-invasive and highly accurate prediction of esophageal squamous cell carcinoma recurrence in patients can be achieved.

[0015] (3) According to the biomarkers and models provided by the present invention, the sample for relapse prediction is peripheral blood, which is easy to obtain; the prediction method relies on high-throughput sequencing, which has high detection efficiency; the prediction model is reliable and the prediction results are highly accurate. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the results of the prediction model of this invention using sklearn to calculate the area under the curve (AUC); Figure 2 This is a schematic diagram of the recurrence prediction results of 61 patients with esophageal squamous cell carcinoma, with a sensitivity of 0.93 and a specificity of 0.89. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention provides a biomarker for predicting recurrence of esophageal squamous cell carcinoma, the biomarker including TRIM37, LLPH, SOX17, LINC02694, SERPINH1, TP53TG5, ANGPT2, ACTRT2, HECTD4, P2RY1 and LINC01102.

[0020] The TRIM37 gene encodes an E3 ubiquitin ligase belonging to the triple motif family. It is primarily involved in intracellular protein ubiquitination, histone ubiquitination regulation, and gene expression silencing. It also plays a role in physiological activities such as peroxisome homeostasis and cell cycle regulation. Abnormal expression or functional mutations of this gene are associated with the development and progression of various diseases and tumors.

[0021] The LLPH gene encodes a neurologically related functional protein that participates in the regulation of synaptic plasticity, neural signal transmission, and long-term synaptic facilitation. It plays an important role in the physiological mechanisms related to neural development and learning and memory, and its abnormal expression may affect the functional homeostasis of the nervous system.

[0022] SOX17 belongs to the SOX transcription factor family and is a key transcriptional regulator involved in embryonic endoderm differentiation, angiogenesis, and tissue development. It can affect cell proliferation and differentiation by regulating the Wnt / β-catenin signaling pathway and plays an important role in maintaining normal cell growth and inhibiting abnormal proliferation.

[0023] LINC02694 is a long intergenic non-coding RNA that does not encode functional proteins. It mainly participates in gene expression regulation through epigenetic regulation and post-transcriptional regulation, affecting cell proliferation, apoptosis, and signaling pathway activity, and plays a regulatory role in various physiological and pathological processes.

[0024] The SERPINH1 gene encodes heat shock protein 47 (HSP47), which, as a collagen-specific molecular chaperone, mediates the correct folding, processing, and secretion of collagen in the endoplasmic reticulum. It participates in extracellular matrix construction, tissue repair, and fibrosis-related processes, and plays an important role in maintaining connective tissue homeostasis.

[0025] TP53TG5 is a downstream target gene of the p53 signaling pathway. Its expression is regulated by p53 protein and it participates in the regulation of DNA damage response, cell cycle arrest and apoptosis. It plays a role in maintaining genome stability and inhibiting abnormal cell proliferation.

[0026] ANGPT2 encodes angiopoietin 2, a secretory protein related to angiogenesis regulation. It participates in the regulation of vascular stability, angiogenesis, and vascular permeability by binding to its receptor, and plays an important role in vascular remodeling, inflammatory response, and tissue microenvironment construction.

[0027] ACTRT2 belongs to the actin-related protein family and is specifically expressed in testicular tissue. It participates in spermatogenesis and cytoskeleton assembly, and plays a role in germ cell development and the maintenance of normal reproductive function.

[0028] HECTD4 encodes an E3 ubiquitin ligase containing a HECT domain. It participates in signaling pathway regulation by mediating the ubiquitination and degradation of substrate proteins, and can affect the levels of cell proliferation, migration, and inflammation-related proteins, playing a role in cell physiological homeostasis and the regulation of abnormal proliferation.

[0029] P2RY1 encodes the P2Y1 purinergic receptor, a G protein-coupled receptor that primarily mediates ADP signaling and participates in processes such as platelet aggregation, calcium ion release, and vascular smooth muscle regulation. It plays an important role in coagulation function and cardiovascular-related physiological activities.

[0030] LINC01102 is a long non-coding RNA that participates in cell biological processes by regulating target gene transcription, mRNA stability, and signaling pathway activity. It can affect cell proliferation, apoptosis, and differentiation, and plays a regulatory role in expression under various physiological and pathological conditions.

[0031] The present invention provides the use of the biomarker, the use including: for constructing a model for predicting the recurrence of esophageal squamous cell carcinoma, and for preparing a product for predicting the recurrence of esophageal squamous cell carcinoma, the product being based on the model for predicting the recurrence of esophageal squamous cell carcinoma to predict the recurrence of esophageal squamous cell carcinoma.

[0032] Specifically, "the model for constructing a predictive model for esophageal squamous cell carcinoma recurrence" refers to using data from the aforementioned biomarkers (a combination of TRIM37, LLPH, SOX17, LINC02694, SERPINH1, TP53TG5, ANGPT2, ACTRT2, HECTD4, P2RY1, and LINC01102) and specific algorithms and computational methods to establish a mathematical or computational structure capable of assessing an individual's risk of esophageal squamous cell carcinoma recurrence. The core of this model lies in transforming biomarker content information into clinically meaningful predictive results.

[0033] This invention provides a model for predicting the recurrence of esophageal squamous cell carcinoma, wherein the input variable of the model is the content of the biomarker.

[0034] Biomarker levels refer to the data obtained by quantitatively detecting specific biomarkers in biological samples. Changes in the levels of these biomarkers, such as TRIM37, LLPH, SOX17, LINC02694, SERPINH1, TP53TG5, ANGPT2, ACTRT2, HECTD4, P2RY1, and LINC01102, may be closely related to the recurrence of esophageal squamous cell carcinoma. Various methods exist for measuring biomarker levels. For example, quantitative polymerase chain reaction (qPCR) can be used to detect gene mRNA expression levels; enzyme-linked immunosorbent assay (ELISA) or mass spectrometry can be used to detect protein expression levels; or high-throughput sequencing technologies (such as RNA-seq or specific epigenetic sequencing methods) can be used to obtain quantitative data at the genomic or epigenomic level. This quantitative data provides objective and reproducible numerical inputs for models, enabling them to learn and identify patterns associated with esophageal squamous cell carcinoma recurrence based on these numerical changes.

[0035] Furthermore, the method for determining the content of the biomarker is 5hmC high-throughput detection. Preferably, it is 5hmC-Seal.

[0036] High-throughput 5hmC detection is a technique for large-scale, efficient quantitative analysis of 5-hydroxymethylcytosine (5hmC) modifications in biological samples. As an important epigenetic marker, the distribution and abundance of 5hmC in the genome are closely related to the occurrence and development of various diseases. High-throughput detection allows for the simultaneous analysis of 5hmC levels in a large number of samples or genomic regions, thus obtaining comprehensive and detailed epigenetic information. 5hmC-Seal (Selective Chemical Labeling and Enrichment for 5-Hydroxymethylcytosine) is a highly specific and sensitive high-throughput 5hmC detection method. The core of this technology lies in its unique chemical labeling and enrichment strategy, which can accurately capture and quantify 5hmC sites in the genome. Specifically, 5hmC-Seal technology typically utilizes T4 phage β-glucosyltransferase (T4-βGT) to specifically add glucose groups to 5hmC, forming glucosyl-5hmC. Subsequently, DNA fragments containing 5hmC were efficiently enriched using a biotin-labeled anti-glucosyl-5hmC antibody to separate them from the complex genomic background. Finally, the enriched DNA fragments were subjected to high-throughput sequencing to achieve precise identification and quantification of the 5hmC site.

[0037] Through the above technical solution, this application addresses the inaccuracies and inefficiencies in existing biomarker determination methods by specifying a particular method for measuring biomarker content, thereby improving the reliability of model predictions. Specifically, the biomarker content determination method is 5hmC high-throughput detection. This technology can efficiently process a large number of samples simultaneously, providing high-precision 5hmC modification level data, avoiding errors that may be introduced by traditional detection methods, and ensuring the quality of data input to the model. Preferably, 5hmC-Seal further optimizes the detection process. Through its high specificity and sensitivity, it more accurately identifies and quantifies 5hmC modification, enhancing the reliability of the measurement results and providing a more stable input basis for the model. This enables the model for predicting esophageal squamous cell carcinoma recurrence, constructed based on a combination of biomarkers TRIM37, LLPH, SOX17, LINC02694, SERPINH1, TP53TG5, ANGPT2, ACTRT2, HECTD4, P2RY1, and LINC01102, to obtain more accurate and reliable prediction results, thus providing more effective guidance for the early diagnosis and treatment of esophageal squamous cell carcinoma.

[0038] This invention provides a method for constructing a model to predict the recurrence of esophageal squamous cell carcinoma, comprising the following steps: Step 1. Samples from multiple esophageal squamous cell carcinoma patients were analyzed to obtain 5hmC sequencing data of DNA; this step aims to obtain raw data for model training. 5hmC (5-hydroxymethylcytosine) is an important epigenetic modification that plays a crucial role in gene expression regulation and disease development. Analyzing its sequencing data can reflect the epigenetic state of DNA in esophageal squamous cell carcinoma patient samples.

[0039] The sample used is plasma, preferably peripheral blood. Using peripheral blood effectively solves the problem of unstable and unreliable biomarker detection results caused by the lack of sample type limitation. Plasma provides pure circulating cell-free DNA, avoiding interference from cellular components, allowing 5hmC sequencing data to more accurately reflect the epigenetic characteristics of esophageal squamous cell carcinoma, thereby improving the accuracy and stability of biomarker detection. Simultaneously, using standardized and readily available peripheral blood as the plasma source ensures the consistency and reproducibility of the sample source, which is crucial for constructing a model to predict esophageal squamous cell carcinoma recurrence, enhancing the generalization ability and predictive performance of the trained model across different patient populations. Furthermore, this non-invasive sample collection method significantly improves patient compliance, facilitating the clinical promotion and application of this predictive method. Overall, this application, by optimizing sample selection, provides a high-quality and reliable data foundation for subsequent 5hmC sequencing, biomarker screening, and machine learning model construction, thereby improving the overall accuracy and practicality of esophageal squamous cell carcinoma recurrence prediction. Step 2. The sequencing data obtained in Step 1 are sequentially subjected to first filtering, screening, and second filtering to obtain biomarkers for predicting recurrence of esophageal squamous cell carcinoma. This phased processing aims to extract biomarkers that are highly correlated with the prediction of recurrence of esophageal squamous cell carcinoma from the raw sequencing data, gradually remove noise and redundant information, and focus on differential features with statistical and biological significance, thereby improving the reliability and predictive ability of the finally selected biomarkers.

[0040] The first filtering step includes removing peak information that appears only in 10 or fewer samples. This filtering step aims to remove low-frequency or sample-specific noise data, improving data reliability and the efficiency of subsequent analysis. If a 5hmC peak appears only in a very small number of samples, it is likely to represent random noise, sequencing errors, or individual differences that do not have universal biological significance, rather than a stable biomarker. Specifically, this can be achieved by writing scripts or using bioinformatics tools to count the frequency of each 5hmC peak across all samples and setting a threshold (e.g., 10 samples), removing peaks below this threshold from the dataset. Another approach is to set a minimum sample coverage during the data preprocessing stage, when merging and quantizing the original peak files, ensuring that only peaks detected in a sufficient number of samples are included in subsequent analysis.

[0041] The screening process includes: comparing sequencing data from different samples using DEseq2 software, retaining 5hmC peak regions with read counts greater than 50, and identifying differentially expressed biomarkers of 5hmC upregulation or downregulation based on FoldChange >= 0.5 and pvalue < 0.01. This screening step aims to identify 5hmC peak regions with significant differential expression between the recurrent and non-recurrent esophageal squamous cell carcinoma groups, i.e., differential biomarkers. DEseq2 is an R package widely used for differential expression analysis of count data, which can assess the significant differences in regional expression levels between different groups using statistical methods. In DEseq2 analysis, the read count of each 5hmC peak region is used as input to define sample groups and run differential expression analysis. Simultaneously, a threshold of read counts greater than 50 is set to ensure that the analyzed peaks have sufficient sequencing depth and avoid statistical bias caused by low-abundance peaks. The criteria for differential biomarkers, FoldChange >= 0.5 and pvalue < 0.01, are commonly used statistical thresholds used to screen for differentially expressed peaks with statistical significance and certain biological effects. Besides DEseq2, other similar differential expression analysis tools, such as edgeR or limma, can also be used. They can also handle count data and perform differential analysis, and can be combined with similar statistical thresholds for screening.

[0042] The second filtering includes: filtering the differentially expressed biomarkers obtained through the screening using the recursive feature elimination algorithm in Scikit-Learn to obtain biomarkers for predicting recurrence of esophageal squamous cell carcinoma: TRIM37, LLPH, SOX17, LINC02694, SERPINH1, TP53TG5, ANGPT2, ACTRT2, HECTD4, P2RY1, and LINC01102.

[0043] The second filter uses a logistic regression cross-validator, with the class weights of the logistic regression cross-validator set to balanced mode, the number of cross-validation folds set to 2, the maximum number of iterations set to 1000, and the scoring metric set to accuracy.

[0044] This filtering step aims to further refine the set of differentially expressed biomarkers, removing redundant or low-contribution features to obtain an optimal combination of biomarkers with high predictive power. Recursive Feature Elimination (RFE) is a feature selection algorithm that iteratively builds a model and removes the least important features until the desired number of features or performance metric is reached. RFE algorithms are typically used in conjunction with an evaluator, such as a linear model or support vector machine. In each iteration, the model is trained on the current feature set, and then the lowest-ranking features are removed based on their importance. This process is repeated until a predetermined number of features remain, or the cross-validation score reaches its optimum. Besides RFE, other feature selection methods, such as L1 regularization-based feature selection or tree-based feature importance evaluation, can be used to identify and select the most predictive subset of biomarkers. The final combination of specific biomarkers (TRIM37, LLPH, SOX17, LINC02694, SERPINH1, TP53TG5, ANGPT2, ACTRT2, HECTD4, P2RY1, and LINC01102) was the result of rigorous screening and optimization, and together they constitute the core feature set for predicting recurrence of esophageal squamous cell carcinoma.

[0045] Step 3. Input the biomarker data obtained in Step 2 into the machine learning model, train the model, and store the trained model to obtain a model for predicting esophageal squamous cell carcinoma recurrence. The machine learning model includes: training a logistic regression CV model. Logistic regression, also known as logistic regression analysis, is a generalized linear regression analysis model. This embodiment uses logistic regression analysis for data mining to explore factors related to esophageal squamous cell carcinoma recurrence and predict recurrence based on these factors. Through logistic regression analysis, the weights of independent variables can be obtained, thus revealing which factors are closely related to esophageal squamous cell carcinoma recurrence. Furthermore, based on these weights, esophageal squamous cell carcinoma recurrence can be predicted.

[0046] This step is crucial for building a predictive model. It aims to utilize selected biomarker data, learn patterns in the data through machine learning algorithms, and generate a model capable of predicting new samples. Machine learning models can automatically learn patterns and regularities from large amounts of data and apply them to predict or classify new, unknown data. Training the model involves adjusting its parameters using a dataset with known labels (e.g., whether recurrence has occurred) to accurately identify the relationship between biomarkers and recurrence status. Various machine learning algorithms can be used to train the model, such as logistic regression, support vector machines, random forests, gradient boosting trees, and neural networks. The training process typically includes data preprocessing, model selection, parameter tuning, and model evaluation. Storing the trained model means preserving its parameters and structure so that it can be directly loaded and used to predict the recurrence risk of new esophageal squamous cell carcinoma patient samples without retraining.

[0047] Through the above technical solution, this application provides a systematic, multi-stage optimized model construction method, effectively addressing the challenge of screening highly relevant and reliable combinations from massive biomarkers. The first filter removes low-frequency noise, ensuring data purity; the DEseq2 software is used for differential screening to identify potential biomarkers with statistical and biological significance; a second filter, employing a recursive feature elimination algorithm, further simplifies and optimizes the feature set, obtaining the most predictive biomarker combinations. This progressive screening strategy significantly improves the accuracy and robustness of the constructed model, enabling it to more effectively identify the risk of esophageal squamous cell carcinoma recurrence, providing a more reliable tool for early clinical diagnosis and intervention. Example

[0048] This invention provides a model for predicting the recurrence of esophageal squamous cell carcinoma. The relevant parameters of 11 5hmC characteristic biomarkers in the model are shown in Table 1. Table 1

[0049] In the embodiments of the present invention, the 5hmC-Seal high-throughput sequencing method is used to sequence the samples. The 5hmC-Seal high-throughput sequencing method used in the following embodiments is explained as follows: 5hmC-Seal is a high-throughput sequencing method based on 5hmC. This method uses an improved chemical glycosylation marker combined with next-generation high-throughput sequencing technology to obtain the distribution information of 5hmC on genomic DNA.

[0050] Due to the high sensitivity of chemical labeling, the input DNA can be as low as 1-10 ng. The DNA can be fragmented genomic DNA or small fragments such as cfDNA. According to the requirements of next-generation sequencing, the DNA fragments are padded at both ends, and then an A tail is ligated to the 3' end. Sequencing Y-shaped adapters are ligated to both ends of each DNA fragment using AT specific ligation. These adapters contain index information that distinguishes samples and sequences such as amplification primers. Next, the 5hmC labeling step was performed. First, UDP-6-N3-Glc was added, and under specific conditions, all 5hmC on the DNA reacted to become N3-5ghmC. Then, DBCO-PEG4-Biotin was added, and all N3-5ghmC were linked to biotin. Finally, through efficient and specific binding of biotin-magnetic beads, all DNA fragments containing 5hmC sites were screened. After PCR amplification and purification, the 5hmC-based DNA library was constructed. The size and distribution of DNA bands in each sample were analyzed using Fragment Analyzer for quality control. After precise quantification of the library by qPCR, high-throughput sequencing was performed using an Illumina Nextseq 500 sequencer to obtain the base sequences of all DNA fragments in the library.

[0051] Application Example 1 122 patients with esophageal squamous cell carcinoma, including both recurrent and non-recurrent patients, were included in the study. Peripheral blood samples (3-4 mL) were collected from these 122 patients, and cfDNA was extracted from the plasma for 5hmC-Seal high-throughput sequencing. The sequencing throughput for each sample was 1.5 Gb, and the sequencing band size was 38 bp.

[0052] The 122 samples were randomly divided into a training cohort and a validation cohort at a 1:1 ratio. Peak information of plasma cfDNA from different patients in the training cohort (61 cases) was compared and filtered, with each peak appearing in at least 10 samples. Using DEseq2 software, DNA sequencing data from different samples were compared to identify 5hmC peak regions with more than 50 reads. Differentially regulated biomarkers of 5hmC were identified based on FoldChage >= 0.5 and pvalue < 0.01. Then, the recursive feature elimination algorithm (RFECV) in Scikit-Learn was used to further filter these differentially regulated biomarkers (using parameters: estimator = LogisticRegressionCV (class_weight = 'balanced', cv = 2, max_iter = 1000), scoring = 'accuracy'). Biomarkers obtained: TRIM37, LLPH, SOX17, LINC02694, SERPINH1, TP53TG5, ANGPT2, ACTRT2, HECTD4, P2RY1 and LINC01102.

[0053] Finally, a model was built using the 11 selected biomarkers. A logistic regression (LR) model was trained, and receiver operating characteristic (ROC) analysis was used to evaluate the model's performance. The area under the curve (AUC), sensitivity, and specificity were calculated using sklearn. The results of calculating the AUC using sklearn are shown below. Figure 1 As shown, the constructed LR prediction model was used in Example 2 to predict relapse in patients in the validation group.

[0054] Application Example 2 Peripheral blood sequencing results were obtained from the validation group (61 patients, including 15 relapsed and 46 non-relapsed patients) in Application Example 1. The sequencing data were processed and analyzed to screen for 5hmC-enriched sites. Then, relapse prediction was performed using the prediction model obtained in Application Example 1 based on the screened 5hmC-enriched sites. The prediction results are as follows: Figure 2 As shown, the results indicated that 11 out of 15 relapsed patients were predicted to be relapsed, with a sensitivity of 0.93; and 41 out of 46 non-relapsed patients were predicted to be non-relapsed, with a specificity of 0.89.

[0055] Therefore, it can be concluded that the prediction model constructed based on the above 11 5hmC characteristic biomarkers provided by the present invention can effectively predict the recurrence of esophageal squamous cell carcinoma in advance.

[0056] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A biomarker for predicting recurrence of esophageal squamous cell carcinoma, characterized in that, The biomarkers include TRIM37, LLPH, SOX17, LINC02694, SERPINH1, TP53TG5, ANGPT2, ACTRT2, HECTD4, P2RY1, and LINC01102.

2. The use of the biomarker as described in claim 1, characterized in that, The uses include: constructing a model for predicting the recurrence of esophageal squamous cell carcinoma, and preparing a product for predicting the recurrence of esophageal squamous cell carcinoma, the product being based on the model for predicting the recurrence of esophageal squamous cell carcinoma.

3. A model for predicting recurrence of esophageal squamous cell carcinoma, characterized in that, The input variable of the model is the content of the biomarker.

4. The model as described in claim 3, characterized in that, The method for determining the content of the biomarker is 5 hmC high-throughput detection.

5. A method for constructing a model to predict the recurrence of esophageal squamous cell carcinoma, characterized in that, Includes the following steps: Step 1. Samples from multiple patients with esophageal squamous cell carcinoma were analyzed to obtain 5hmC sequencing data of DNA; Step 2. Perform the sequencing data obtained in Step 1 through first filtering, screening, and second filtering to obtain biomarkers for predicting recurrence of esophageal squamous cell carcinoma. Step 3. Input the biomarker data obtained in Step 2 into the machine learning model, train the model, store the trained model, and obtain a model for predicting the recurrence of esophageal squamous cell carcinoma.

6. The construction method as described in claim 5, characterized in that, The first filtering includes: removing peak information that appears only in 10 or fewer samples; The screening process includes: comparing sequencing data from different samples using DEseq2 software, retaining 5hmC peak regions with read counts greater than 50, and identifying differentially regulated biomarkers of 5hmC based on fold change >= 0.5 and p-value < 0.

01. The second filtering includes: filtering the differentially expressed biomarkers obtained through the screening using the recursive feature elimination algorithm in Scikit-Learn to obtain biomarkers for predicting recurrence of esophageal squamous cell carcinoma: TRIM37, LLPH, SOX17, LINC02694, SERPINH1, TP53TG5, ANGPT2, ACTRT2, HECTD4, P2RY1, and LINC01102.

7. The construction method as described in claim 5, characterized in that, The sample in step 1 is plasma.

8. The construction method as described in claim 6, characterized in that, The second filter uses a logistic regression cross-validator, with the class weights of the logistic regression cross-validator set to balanced mode, the number of cross-validation folds set to 2, the maximum number of iterations set to 1000, and the scoring metric set to accuracy.

9. The construction method as described in claim 5, characterized in that, The machine learning model in step 3 includes: training a logistic regression CV model.

Images