DNA methylation site marker for detecting hepatocellular carcinoma, multiplex dPCR kit and diagnostic model construction method

Abstract

The application discloses a DNA methylation site marker for detecting hepatocellular carcinoma, a multiplex dPCR kit and a diagnostic model construction method. The DNA methylation site marker comprises cg02829688, cg13080379, cg03760839, cg10703826, cg12664119, cg16990168 and cg23371746 sites. The application designs a primer probe group for the above sites, constructs a multiplex dPCR kit of a double reaction system, realizes quantification of target DNA methylation sites, takes the methylation levels of the 7 sites as characteristic variables, adopts an XGBoost algorithm and / or an LR algorithm to construct a diagnostic model. Experimental verification shows that the model can effectively distinguish hepatocellular carcinoma patients from liver cirrhosis, chronic hepatitis B, metabolic dysfunction-related fatty liver disease patients and healthy individuals, especially shows good efficiency in auxiliary diagnosis of early hepatocellular carcinoma, and provides a new technical scheme for clinical diagnosis and screening of hepatocellular carcinoma.

Description

Technical Field

[0001] This invention belongs to the field of molecular biology detection and clinical medical diagnosis, specifically involving DNA methylation site markers for detecting hepatocellular carcinoma, multiplex dPCR kits, and methods for constructing diagnostic models. Background Technology

[0002] Hepatocellular carcinoma (HCC) is a highly fatal malignant tumor worldwide. Because early HCC symptoms are often subtle, and small lesions can easily be confused with cirrhotic nodules on imaging, approximately 70% of patients are diagnosed at an advanced stage, missing the opportunity for radical surgery. Currently, the clinically accepted screening method combining alpha-fetoprotein (AFP) and abdominal ultrasound has a high rate of missed diagnoses for early HCC (sensitivity of only about 63%).

[0003] In recent years, liquid biopsy-based detection of circulating tumor DNA (ctDNA) methylation has shown great potential. However, existing technologies still face challenges: 1) Insufficient biomarker specificity: some biomarkers have background signals in non-hepatic tissues; 2) Limited detection sensitivity: the abundance of ctDNA in the peripheral blood of early-stage patients is extremely low (often below 0.1%), making it difficult for traditional qPCR to achieve accurate quantification, while NGS sequencing is expensive and time-consuming.

[0004] Therefore, there is an urgent need to develop a non-invasive diagnostic method for early liver cancer that is highly specific, highly sensitive, and suitable for large-scale clinical application. Summary of the Invention

[0005] The purpose of this invention is to provide DNA methylation site markers for detecting hepatocellular carcinoma, multiplex dPCR kits, and methods for constructing diagnostic models.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A DNA methylation site marker for detecting hepatocellular carcinoma, the DNA methylation site marker comprising the sites cg02829688, cg13080379, cg03760839, cg10703826, cg12664119, cg16990168, and cg23371746.

[0007] A primer-probe set for detecting the aforementioned DNA methylation site markers, comprising primers and probes for cg02829688, cg13080379, cg03760839, cg10703826, cg12664119, cg16990168, and cg23371746 sites, and primers and probes for the internal reference gene ACTB; the sequences of the primer-probe set are shown below: cg16990168-F:5'-GGGTYGGAGTTTTATAGAATGTTTTTATAAT-3', cg16990168-R:5'-TCTACRCACTAATCRAATCTTACTAATT-3', cg13080379-F:5'-GGTGGAGTAGGGGTTGAG-3', cg13080379-R:5'-CTTTCRAACTAAAATCRACTAAAACAACT-3', cg02829688-F:5'-GGAGGAGGAATTTTTTTATAGAAGGT-3', cg02829688-R:5'-CCRCCAAAATCTTCTACTAAATACTAC-3', cg03760839-F:5'-GGGYGAGTTTTTTAGGYGT-3', cg03760839-R:5'-CTCCATAAATTCTCTACAAAAAACC-3', cg12664119-F:5'-YGYGAGGTTYGGGAGA-3', cg12664119-R:5'-CTCCRCAATCCRACAAAAAATTC-3', cg10703826-F:5'-AGTTAAGTTGAAAAAAYGTTTYGGA-3', cg10703826-R:5'-ATAAAACCRTCTCRACCAAAACA-3', cg23371746-F:5'-GTTGGAGATAAGGTAGGGGA-3', cg23371746-R:5'-CRCCRAACRACACCAA-3', ACTB-NCF:5'-TGGTGTTTGTYTYTYTGAYTAGGT-3', ACTB-NCR:5'-CCARCCTCATRRCCTTRTCACA-3', cg16990168-FAM:5'-FAM-GGGAATTTA+GTTTYG+CGTAT-BHQ-3', cg13080379-Cy5:5'-Cy5-GGT+CGTAG+TYGGAG-BHQ-3', cg02829688-HEX:5'-HEX-GTT+TYG+CGGG+AGA-BHQ-3', cg03760839-ROX: 5'-ROX-GGGTTG+CGTT+TYGTT-BHQ-3', cg12664119-HEX:5'-HEX-TTT+TG+C+GGTAAGAAAAGA-BHQ-3', cg10703826-FAM: 5'-FAM-AGYGGTTTAGATATAGTT+CGATTT-BHQ-3', cg23371746-ROX: 5'-ROX-TTTTTAG+CGGTTGGGGTTTT-BHQ-3', ACTB-Cy5: 5'-Cy5-AAGA+CAGTGTTGTGGGTGTAGG-BHQ-3'.

[0008] The application of the aforementioned DNA methylation site markers and the aforementioned primer-probe set in the preparation of hepatocellular carcinoma diagnostic products; Furthermore, the diagnosis of hepatocellular carcinoma includes distinguishing hepatocellular carcinoma from cirrhosis, chronic hepatitis B, metabolic dysfunction-related fatty liver disease, and / or healthy individuals. Furthermore, the product is a multiplex dPCR kit.

[0009] A multiplex dPCR kit for the diagnosis of hepatocellular carcinoma, the multiplex dPCR kit comprising the primer and probe set described above; Furthermore, the multiplex dPCR kit includes two reaction systems: the first reaction system is used to detect cg16990168, cg02829688, cg03760839 and cg13080379 sites, and the second reaction system is used to detect cg10703826, cg12664119 and cg23371746 sites and the internal reference gene ACTB; Furthermore, the reaction procedure of the multiplex dPCR kit includes: pre-incubation at 50°C for 3 min, pre-denaturation at 95°C for 2 min; followed by 40 cycles, each cycle including denaturation at 95°C for 15 s and annealing at 56°C for 30 s.

[0010] A method for constructing a diagnostic model for hepatocellular carcinoma includes the following steps: 1) Collect plasma samples from the training cohort used as a diagnostic model for hepatocellular carcinoma; 2) Extract cfDNA from plasma samples and perform sulfite conversion treatment. Then, use the multiplex dPCR kit described in any one of claims 6 to 8 to detect cg02829688, cg13080379, cg03760839, cg10703826, cg12664119, cg16990168 and cg23371746 sites on the converted cfDNA to obtain the methylation level of DNA methylation site markers for each sample. 3) Using methylation level as a feature, a hepatocellular carcinoma diagnostic model is obtained by training with XGBoost or LR machine learning classification algorithms.

[0011] A diagnostic model for hepatocellular carcinoma is obtained by the above-described construction method.

[0012] The beneficial effects of this invention are as follows: This invention, through deep integration of public big data mining and clinical tissue sample validation, screened a set of specific DNA methylation sites located on the liver-enriched gene TBX15 and constructed a machine learning-based diagnostic model. Compared with existing technologies, this invention has the following significant advantages: (1) Extremely high tissue specificity and signal-to-noise ratio: The biomarkers selected in this invention are derived from "liver-enriched genes", which are expressed at low levels or not at all in non-liver tissues. This characteristic effectively filters out background noise interference from non-target tissues such as leukocytes in peripheral blood, making the detection signal highly focused on liver lesions and significantly improving the accuracy of diagnosis.

[0013] (2) Excellent early diagnostic efficacy: Clinical validation data show that the model of this invention has good ability to distinguish early hepatocellular carcinoma (BCLC0 / A stage), and can effectively differentiate patients with early hepatocellular carcinoma from patients with benign liver diseases such as cirrhosis. In the "gray zone" scenario of clinical screening, this invention can still maintain high sensitivity, effectively making up for the shortcomings of existing serological screening in missing diagnosis, and providing strong technical support for the "early detection and early treatment" of liver cancer.

[0014] (3) High sensitivity and stability of the detection system: Combining multiplex dPCR technology, this invention can achieve absolute quantification of extremely low abundance ctDNA signals. Compared with traditional qPCR, this system is more resistant to inhibitors in plasma samples, amplifies more robustly, and greatly improves the reliability of clinical complex sample detection.

[0015] (4) Model intelligence and clinical applicability: By introducing XGBoost and logistic regression machine learning algorithms, the complex nonlinear features between multiple methylation sites can be captured, overcoming the defect of unstable performance of single markers. The constructed model can effectively distinguish hepatocellular carcinoma from a variety of other liver diseases and healthy controls. Attached Figure Description

[0016] Figure 1 The process of discovering DNA methylation sites.

[0017] Figure 2 Differences in DNA methylation levels between HCC cancer tissue and adjacent normal samples.

[0018] Figure 3 : Multiplex dPCR results of DNA methylation sites.

[0019] Figure 4 Comparison of performance metrics of 11 machine learning algorithms.

[0020] Figure 5 The diagnostic efficacy of the HCC diagnostic model in distinguishing between HCC patients and non-HCC populations (including LC, CHB, MASLD, and HC).

[0021] Figure 6 The diagnostic efficacy of HCC diagnostic models and traditional serological markers AFP and DCP in distinguishing HCC patients from LC patients.

[0022] Figure 7 The diagnostic efficacy of HCC diagnostic models compared with traditional serological markers AFP and DCP in distinguishing early HCC patients from LC patients. Detailed Implementation

[0023] This invention provides DNA methylation site markers for detecting hepatocellular carcinoma, a multiplex dPCR kit, and a method for constructing a diagnostic model. Those skilled in the art can refer to the content of this invention to make reasonable adjustments and optimizations to the relevant technical solutions. It should be particularly noted that all such equivalent substitutions, modifications, and variations are obvious to those skilled in the art and should be included within the scope of protection of this invention. The methods and applications described in this invention have been described through preferred embodiments. Those skilled in the art can make appropriate modifications, alterations, or combinations to the methods and applications of this invention without departing from the technical solutions, spirit, and scope of protection of this invention, thereby realizing and applying the technical solutions of this invention.

[0024] Example 1: Screening and Validation of DNA Methylation Site Markers 1.1 Bioinformatics screening of candidate biomarkers This embodiment first uses a large-scale screening of a multi-center public dataset to identify liver-specific DNA methylation site markers with high diagnostic potential. The specific screening process is as follows ( Figure 1 ): (1) Data Acquisition and Preprocessing: Three independent whole-genome methylation profile datasets for hepatocellular carcinoma (HCC) were downloaded from public databases: GSE89852 dataset (containing 37 pairs of paired HCC cancer tissues and adjacent normal tissues), GSE157341 dataset (containing 239 HCC tissues and 35 normal liver tissues), and TCGA-LIHC cohort (containing 377 HCC tissues and 50 normal liver tissues). The raw methylation data were standardized using ChAMP and TCGAbiolinks software packages to eliminate batch effects and experimental errors.

[0025] (2) Analysis and screening of differentially methylated sites (DMPs): A significance threshold of P≤10 was set. -15 In the GSE89852 dataset, GSE157341 dataset, and TCGA-LIHC cohort, 5515, 7378, and 10128 significant DMPs were identified, respectively. The intersection of the DMPs identified in the three datasets yielded 1851 candidate methylation sites, corresponding to 517 genes, that showed significant consistent differences across these three independent HCC whole-genome methylation profile datasets.

[0026] (3) Tissue-specific filtering: The 517 candidate genes were cross-referenced with 978 liver-enriched expression genes in the Human Protein Atlas database to exclude interference from non-hepatic signals. After screening, 10 DNA methylation sites were finally obtained on 4 liver-enriched expression genes (LRP5, NLRP11, PKLR and TBX15).

[0027] 1.2 Validation of clinical samples by multiplex targeted sequencing To verify the clinical reliability of the bioinformatics screening results, this embodiment used independent clinical samples for experimental verification. The specific verification process is as follows: (1) Sample preparation: 29 pairs of HCC cancer tissues and paired adjacent normal tissue samples were collected, genomic DNA was extracted and bisulfite conversion was performed.

[0028] (2) Multiplex targeted sequencing: MethylTarget™ high-throughput sequencing technology was used to quantitatively detect the methylation level of the 10 DNA methylation sites obtained by the above bioinformatics screening.

[0029] (3) Experimental results: The results showed that there were statistically significant differences in 10 DNA methylation sites on 4 liver-enriched expression genes between HCC cancer tissue and adjacent normal tissue. Among them, seven loci (cg02829688, cg13080379, cg03760839, cg10703826, cg12664119, cg16990168, cg23371746) on the TBX15 gene (Ensembl: ENSG00000092607, Gene ID: 6913) and cg15302350 on the LRP5 gene showed highly significant differences (P<0.0001); cg25651783 on the PKLR gene and cg25730098 on the NLRP11 gene also showed significant differences (P=0.0076, P=0.001, respectively). Figure 2 A).

[0030] (4) Core target identification: By calculating the methylation ratio of each DNA methylation site in HCC cancer tissue and adjacent normal tissue, it was found that 7 sites on the TBX15 gene showed higher signal-to-noise ratios, and their methylation difference folds were all greater than 2. Figure 2 B). Based on the statistical significance and fold change characteristics of each DNA methylation site, seven DNA methylation sites located on the TBX15 gene (cg02829688, cg13080379, cg03760839, cg10703826, cg12664119, cg16990168, cg23371746) were finally selected as core detection targets for the subsequent construction of a multiplex dPCR diagnostic system.

[0031] Example 2: Construction and performance evaluation of a dPCR-based multiple methylation detection system 2.1 Construction and optimization of multiplex dPCR reaction system Using the seven DNA methylation sites on the TBX15 gene identified in Example 1 as detection targets, a multiplex digital polymerase chain reaction (dPCR) detection system was constructed. This detection system consisted of two independent reactions: Reaction 1 detected four sites: cg16990168, cg02829688, cg03760839, and cg13080379, corresponding to the FAM, HEX, ROX, and Cy5 fluorescence channels, respectively; Reaction 2 detected three sites: cg10703826, cg12664119, and cg23371746, as well as the internal control ACTB, corresponding to the FAM, HEX, ROX, and Cy5 fluorescence channels, respectively. The sequences of the primers and probes used are shown in Table 1. The experimental instrument used was a BioDigital Qing digital PCR instrument (Shanghai Xiaohaigui Technology Co., Ltd., China).

[0032] Table 1. Specific nucleotide sequences of primers and probes used in Example 2

[0033] Note: The bases marked with "+" in the sequence are locked nucleic acid (LNA) modified bases; Y and R are degenerate bases, where Y represents C / T and R represents A / G.

[0034] Both dPCR reactions 1 and 2 used a 30 μL reaction system for droplet generation, PCR amplification, and fluorescence signal detection. The final concentrations / volumes of each component in the reaction system were as follows: 1× Taq PCR buffer (TaKaRa), 6 mM MgCl2, 0.3 mM dNTPs, 1 U Taq DNA polymerase (TaKaRa), 0.3 μM per probe, 0.3 μM per primer, 5 μL template DNA, and the remaining volume was brought to a total volume of 30 μL with ultrapure water.

[0035] The specific experimental procedure for dPCR was as follows: 30 μL of the dPCR reaction system was placed in a droplet generator to generate droplets, followed by PCR amplification. The amplification reaction conditions were set as follows: pre-incubation at 50℃ for 3 min, pre-denaturation at 95℃ for 2 min; then 40 cycles were performed, each cycle including denaturation at 95℃ for 15 s and annealing at 56℃ for 30 s; after amplification, signal data for each fluorescence channel were collected using a chip reader. For each positive sample, the proportion of hypermethylated DNA was calculated as the ratio of the number of droplets containing the target methylated sequence to the number of droplets containing the internal control ACTB sequence. The experimental results are as follows: Figure 3 As shown, in the two independent reactions, the seven DNA methylation sites and the internal reference gene ACTB all exhibited stable and clear fluorescence signal distributions. Furthermore, the positive droplets from each methylation site and the internal reference gene were effectively identified, and no significant cross-interference of fluorescence signals was observed between the detection channels, indicating that the system has good target specificity.

[0036] 2.2 Performance Analysis (LOB, LOD, LOQ) Evaluation The performance of the multiplex dPCR detection system constructed in 2.1 was systematically analyzed and evaluated, and the indicators are as follows (Table 2): (1) Limit of blank (LOB): By repeating the independent experiment 30 times for the template-free control, the LOB of all target sites was determined to be less than 1 copy / µL by statistical analysis of the detection results of each target site.

[0037] (2) Limit of Detection (LOD) and Linearity: Using 10 4 The methylated plasmid standards (synthesized by Xiamen Newtech Biotechnology Co., Ltd.) were tested at concentration gradients from copies / µL to 0.1 copies / µL. Specifically, the methylated plasmid at position cg02829688 was obtained by inserting the nucleotide sequence shown in SEQ ID NO.25 into the EcoRV site of the pUC57 plasmid; the methylated plasmid at position cg13080379 was obtained by inserting the nucleotide sequence shown in SEQ ID NO.26 into the EcoRV site of the pUC57 plasmid; the methylated plasmid at position cg03760839 was obtained by inserting the nucleotide sequence shown in SEQ ID NO.27 into the EcoRV site of the pUC57 plasmid; the methylated plasmid at position cg10703826 was obtained by inserting the nucleotide sequence shown in SEQ ID NO.28 into the EcoRV site of the pUC57 plasmid; the methylated plasmid at position cg12664119 was obtained by inserting the nucleotide sequence shown in SEQ ID NO.29 into the EcoRV site of the pUC57 plasmid; and the methylated plasmid at position cg16990168 was obtained by inserting the nucleotide sequence shown in SEQ ID NO.25 into the EcoRV site of the pUC57 plasmid. The nucleotide sequence shown in SEQ ID NO. 30 was inserted into the EcoRV site of the pUC57 plasmid; the methylated plasmid at the cg23371746 site was obtained by inserting the nucleotide sequence shown in SEQ ID NO. 31 into the EcoRV site of the pUC57 plasmid. The results showed that the LOD for all targets was 1 copy / µL; within the concentration range of 1–10,000 copies / µL, the observed values ​​for each target showed a high linear correlation with the theoretical concentration values, with a linear correlation coefficient R0. 2 >0.999.

[0038] (3) Limit of Quantification (LOQ): Methylation plasmids and internal control plasmids (synthesized by Xiamen Newtec Biotechnology Co., Ltd.) were mixed in different proportions to prepare a series of standards with methylation abundances of 25%, 5%, 1%, 0.2%, and 0.04%, respectively, for testing. The internal control plasmid was obtained by inserting the nucleotide sequence shown in SEQ ID NO.32 into the EcoRV site of the pUC57 plasmid. Experiments confirmed that both multiplex dPCR detection systems constructed in this invention can stably detect trace methylation signals as low as 0.2%.

[0039] Table 2 Multiplex dPCR system performance

[0040] The above results demonstrate that the multiplex dPCR detection system constructed in this invention possesses excellent sensitivity, linearity, and quantitative accuracy, enabling precise quantitative detection of trace amounts of tumor DNA in complex contexts, fully meeting the detection needs of low-abundance clinical samples.

[0041] Example 3: Construction and Clinical Efficacy Validation of a Hepatocellular Carcinoma Diagnostic Model Based on Machine Learning Algorithms 3.1 Sample set construction and model selection This embodiment uses a large-scale clinical sample to train and validate a hepatocellular carcinoma (HCC) diagnostic model constructed based on the seven DNA methylation sites (cg02829688, cg13080379, cg03760839, cg10703826, cg12664119, cg16990168, cg23371746) of this invention. All clinical samples included in this embodiment underwent strict pretreatment and detection according to the following procedure: First, cell-free DNA (cfDNA) was extracted from the samples using a magnetic bead method; the extracted cfDNA was then subjected to sulfite conversion to achieve specific labeling of DNA methylation status; subsequently, the converted cfDNA was used to detect the methylation levels of the seven DNA methylation sites according to the multiplex dPCR detection system described in Example 2, providing detection data support for the training and validation of the HCC diagnostic model.

[0042] (1) Sample grouping: Plasma samples from 171 HCC patients and 112 LC patients were included and randomly assigned to the training set (HCC: 120 cases, LC: 79 cases) and the test set (HCC: 51 cases, LC: 33 cases) in a ratio of approximately 7:3.

[0043] (2) Algorithm selection: Eleven machine learning algorithms were selected for comparative analysis. The focus was on the predictive robustness and clinical applicability of each algorithm after model construction on the training and test sets. Finally, the XGBoost algorithm and the logistic regression (LR) algorithm were selected to construct the HCC diagnostic model. Figure 4 ).

[0044] (3) Generalization ability assessment: To further verify the specificity of the model, plasma samples from 119 patients with chronic hepatitis B (CHB), 86 patients with metabolic dysfunction-associated fatty liver disease (MASLD), and 123 healthy controls (HC) were additionally included. These samples were randomly assigned to the training and test sets at a ratio of approximately 7:3 to evaluate the diagnostic efficacy of the model in distinguishing between HCC patients and non-HCC individuals (including LC, CHB, MASLD, and HC). Ultimately, the training set included 120 HCC cases, 79 LC cases, 83 CHB cases, 60 MASLD cases, and 74 HC cases; the test set included 51 HCC cases, 33 LC cases, 36 CHB cases, 26 MASLD cases, and 49 HC cases. The Control group was defined as a pooled group of all non-HCC patients mentioned above. The training set Control group consisted of 79 LC patients, 83 CHB patients, 60 MASLD patients, and 74 HC patients, totaling 296 patients. The test set Control group consisted of 33 LC patients, 36 CHB patients, 26 MASLD patients, and 49 HC patients, totaling 144 patients. This group was used to evaluate the overall diagnostic efficacy of the model in distinguishing HCC patients from all types of non-HCC patients.

[0045] 3.2 Diagnostic Model Performance Evaluation The diagnostic capabilities of the HCC diagnostic models constructed using the XGBoost algorithm and the LR algorithm in differentiating between HCC patients and non-HCC populations were evaluated using ROC curves, and their efficacy was compared with commonly used traditional serological markers. Specifically, the levels of alpha-fetoprotein (AFP) in patient serum samples were detected using Roche electrochemiluminescence immunoassay, and the levels of abnormal prothrombin (DCP) were detected using enzyme-linked immunosorbent assay (ELISA).

[0046] (1) XGBoost diagnostic model: In the training set, the area under the curve (AUC) distinguishing HCC patients from non-HCC patients were: HCC vs LC: 0.939 (95% CI: 0.908–0.969), HCC vs CHB: 0.898 (95% CI: 0.856–0.940), HCC vs MASLD: 0.911 (95% CI: 0.869–0.952), HCC vs HC: 0.909 (95% CI: 0.869–0.949), HCC vs Control: 0.914 (95% CI: 0.879–0.948); In the test set, the AUC distinguishing HCC patients from non-HCC patients were: HCC vs LC: 0.881 (95% CI: 0.809–0.954), HCC vs CHB: 0.876 (95% CI: 0.804–0.948), HCC vs MASLD: 0.831 (95% CI: 0.737–0.926), HCC vs HC: 0.916 (95% CI: 0.858–0.975), HCC vs Control: 0.880 (95% CI: 0.822–0.938) ( Figure 5 A).

[0047] (2) LR diagnostic model: In the training set, the AUCs distinguishing HCC patients from non-HCC patients were: HCC vs LC: 0.915 (95% CI: 0.878–0.952), HCC vs CHB: 0.877 (95% CI: 0.831–0.923), HCC vs MASLD: 0.900 (95% CI: 0.856–0.943), HCC vs HC: 0.907 (95% CI: 0.868–0.946), HCC vs Control: 0.900 (95% CI: 0.863–0.936); In the test set, the AUCs distinguishing HCC patients from non-HCC patients were: HCC vs LC: 0.862 (95% CI: 0.779–0.944), HCC vs CHB: 0.877 (95% CI: 0.878–0.952 ... 0.808–0.947), HCC vs MASLD: 0.856 (95% CI: 0.773–0.940), HCC vs HC: 0.924 (95% CI: 0.872–0.975), HCC vs Control: 0.882 (95% CI: 0.828–0.936) ( Figure 5 B).

[0048] (3) Comparison with existing technologies: The diagnostic model based on methylation sites constructed in this invention has significantly better diagnostic efficacy than commonly used traditional serological markers. Experimental data show that the AUC of alpha-fetoprotein (AFP) in distinguishing HCC patients from LC patients in the training set and test set is only 0.813 (95% CI: 0.755-0.870) and 0.763 (95% CI: 0.663-0.863), respectively. Figure 6 A); Abnormal prothrombin (DCP) showed AUCs of 0.828 (95% CI: 0.773–0.883) and 0.841 (95% CI: 0.757–0.924) in the training and test sets, respectively, distinguishing HCC patients from LC patients. Figure 6 B).

[0049] (4) A probabilistic weighted fusion strategy was adopted to fuse the predicted probability of the XGBoost diagnostic model with the predicted probability of serum biomarkers (AFP / DCP) levels to construct a joint diagnostic model to distinguish HCC patients from LC patients: the AUC of the XGBoost+AFP joint model was 0.954 (95% CI: 0.928-0.980) and 0.902 (95% CI: 0.837-0.966) in the training set and 0.902 (95% CI: 0.837-0.966) in the test set; the AUC of the XGBoost+DCP joint model was 0.958 (95% CI: 0.934-0.982) and 0.916 (95% CI: 0.856-0.975) in the training set and 0.965 (95% CI: 0.943-0.986) in the training set and 0.921 (95% CI: 0.943-0.986) in the test set; and the AUC of the XGBoost+AFP+DCP joint model reached 0.921 (95% CI: 0.943-0.986) in the training set and 0.921 (95% CI: 0.943-0.986) in the test set. 0.864-0.978).

[0050] (4) Using a probabilistic weighted fusion strategy, the predicted probability of the LR diagnostic model was fused with the predicted probability of serum biomarkers (AFP / DCP) levels to construct a combined diagnostic model to distinguish HCC patients from LC patients: The AUC of the LR+AFP combined model was 0.935 (95% CI: 0.904-0.967) and 0.894 (95% CI: 0.826-0.963) in the training set and 0.936 (95% CI: 0.905-0.967) and 0.909 (95% CI: 0.848-0.971) in the training set and 0.945 (95% CI: 0.916-0.974) and 0.918 (95% CI: 0.861-0.975) in the training set and 0.945 (95% CI: 0.916-0.974) and 0.918 (95% CI: 0.861-0.975) in the training set and 0.945 (95% CI: 0.916-0.974) in the training set and 0.918 (95% CI: 0.861-0.975) in the testing set.

[0051] The results of this embodiment demonstrate that the HCC diagnostic model constructed based on the DNA methylation sites of this invention has extremely high clinical application value.

[0052] Example 4: Performance verification of the model of the present invention in early HCC diagnosis 4.1 Construction of early HCC subgroups To address the limitations of ultrasound combined with AFP screening in early-stage HCC in clinical practice, this embodiment further validates the diagnostic value of the HCC diagnostic model constructed in Embodiment 3 for early-stage HCC (BCLC stage 0 / A or CNLC stage I). Experimental data were derived from the clinical cohort in Embodiment 3, from which early-stage HCC subgroup data were extracted: early-stage HCC subgroups were analyzed from HCC patients in the training and test sets according to the BCLC stage 0 / A or CNLC stage I staging criteria, while LC patients in the corresponding cohorts were matched as controls. The specific sample distribution is as follows: the training set included 39 early-stage HCC patients and 79 LC patients; the test set included 15 early-stage HCC patients and 33 LC patients. Based on these subgroup samples, the diagnostic efficacy of the XGBoost and LR models for differentiating between early-stage HCC and LC patients was systematically evaluated in both the training and test sets, thereby validating the practical value of the models for core early diagnosis scenarios in clinical practice.

[0053] 4.2 Early Diagnostic Efficacy Assessment Experimental results show that the diagnostic model based on DNA methylation sites constructed in this invention performs outstandingly in the early diagnosis of HCC, and its diagnostic efficacy is significantly better than that of traditional serological markers (AFP, DCP). (See attached figures.) Figure 7 : (1) XGBoost diagnostic model: In the training set, the differential diagnostic AUC for early HCC patients and LC patients reached 0.866 (95% CI: 0.795-0.937), significantly higher than that of AFP (0.718, 95% CI: 0.620-0.816) and DCP (0.786, 95% CI: 0.694-0.878) in the same group; in the test set, the differential diagnostic AUC for early HCC patients and LC patients remained at 0.842 (95% CI: 0.733-0.951), consistently outperforming the diagnostic efficacy of AFP (AUC=0.701, 95% CI: 0.508-0.894) and DCP (AUC=0.802, 95% CI: 0.633-0.971) in the same period. Figure 7 A).

[0054] (2) The LR diagnostic model also demonstrated robustness in the diagnosis of early HCC. Its differential diagnostic AUCs for early HCC patients and LC patients in the training and test sets were 0.839 (95% CI: 0.763-0.915) and 0.812 (95% CI: 0.693-0.931), respectively, both superior to the single-item test results of AFP (training set: AUC=0.718, 95% CI: 0.620-0.816; test set: AUC=0.701, 95% CI: 0.508-0.894) and DCP (training set: AUC=0.786, 95% CI: 0.694-0.878; test set: AUC=0.802, 95% CI: 0.633-0.971). Figure 7 B).

[0055] (3) A probabilistic weighted fusion strategy was adopted to fuse the predicted probability of the XGBoost diagnostic model with the predicted probability of serum biomarkers (AFP / DCP) levels to construct a joint diagnostic model for differentiating early HCC patients from LC patients: the AUC of the XGBoost+AFP joint model was 0.885 (95% CI: 0.819-0.951) and 0.859 (95% CI: 0.755-0.963) in the training set and 0.859 (95% CI: 0.755-0.963) in the test set; the AUC of the XGBoost+DCP joint model was 0.909 (95% CI: 0.852-0.966) and 0.865 (95% CI: 0.761-0.968) in the training set and 0.865 (95% CI: 0.761-0.968) in the test set; and the AUC of the XGBoost+AFP+DCP joint model reached 0.916 (95% CI: 0.860-0.971) and 0.879 (95% CI: 0.860-0.971) in the training set and 0.879 (95% CI: 0.860-0.971) in the test set. 0.780-0.978).

[0056] (4) Using a probabilistic weighted fusion strategy, the predicted probability of the LR diagnostic model was fused with the predicted probability of serum biomarkers (AFP / DCP) levels to construct a combined diagnostic model for differentiating between early HCC and LC patients: the AUC of the LR+AFP combined model was 0.858 (95% CI: 0.786-0.930) and 0.836 (95% CI: 0.726-0.947) in the training set and 0.836 (95% CI: 0.726-0.947) in the test set; the AUC of the LR+DCP combined model was 0.881 (95% CI: 0.816-0.947) and 0.857 (95% CI: 0.753-0.961) in the training set and 0.885 (95% CI: 0.820-0.951) and 0.873 (95% CI: 0.775-0.971) in the training set and 0.885 (95% CI: 0.820-0.951) in the test set and 0.873 (95% CI: 0.775-0.971) in the test set.

[0057] The experimental data in this embodiment show that the diagnostic model based on DNA methylation sites provided by the present invention can effectively make up for the sensitivity deficiency of traditional serological markers in early HCC screening, significantly improve the detection rate of early liver cancer, and has important application value in early warning, screening and clinical diagnosis of liver cancer.

Claims

1. A DNA methylation site marker for detecting hepatocellular carcinoma, characterized in that: The DNA methylation site markers include cg02829688, cg13080379, cg03760839, cg10703826, cg12664119, cg16990168, and cg23371746.

2. A primer-probe set for detecting DNA methylation site markers as described in claim 1, characterized in that: The primer and probe set includes primers and probes targeting the cg02829688, cg13080379, cg03760839, cg10703826, cg12664119, cg16990168, and cg23371746 sites, as well as primers and probes targeting the internal reference gene ACTB; the sequences of the primer and probe set are shown below: cg16990168-F:5'-GGGTYGGAGTTTTATATAGAATGTTTTTATAAT-3', cg16990168-R: 5'-TCTACRCACTAATCRAATCTTACTAAATT-3', cg13080379-F: 5'-GGTGGAGTAGGGGTTGAG-3', cg13080379-R: 5'-CTTTCRAACTAAAATCRACTAAAACAACT-3', cg02829688-F: 5'-GGAGGAGGAAAATTTTTTTATAGAAGGT-3', cg02829688-R: 5'-CCRCCAAAATCTTCTACTAAATACTAC-3', cg03760839-F:5'-GGGYGAGTTTTTTAGGYGT-3', cg03760839-R: 5'-CTCCATAAATTCTCTACAACAAAAACC-3', cg12664119-F:5'-YGYGAGGTTYGGGAGA-3', cg12664119-R: 5'-CTCCRCAATCCRACAAAAAATTC-3', cg10703826-F: 5'-AGTTAAGTTGAAAAAAYGTTTYGGA-3', cg10703826-R: 5'-ATAAAAACCRTCTCRACCAAAACA-3', cg23371746-F:5'-GTTGGAGATAAGGTAGAYGGGA-3', cg23371746-R: 5'-CRCCRAACRACACCAAA-3', ACTB-NCF: 5'-TGGTTGTTTGTYTYTYTGAYTAGGT-3', ACTB-NCR: 5'-CCARCCTCATRRCCTTTRTCACA-3', cg16990168-FAM: 5'-FAM-GGGAATTTA+GTTTYG+CGTAT-BHQ-3', cg13080379-Cy5:5'-Cy5-GGT+CGTAG+TYGGAG-BHQ-3', cg02829688-HEX:5'-HEX-GTT+TYG+CGGG+AGA-BHQ-3', cg03760839-ROX: 5'-ROX-GGGTTG+CGTT+TYGTT-BHQ-3', cg12664119-HEX:5'-HEX-TTT+TG+C+GGTAAGAAAAGA-BHQ-3', cg10703826-FAM: 5'-FAM-AGYGGTTTAGATATAGTT+CGATTT-BHQ-3', cg23371746-ROX: 5'-ROX-TTTTTAG+CGGTTGGGGTTTT-BHQ-3', ACTB-Cy5: 5'-Cy5-AAGA+CAGTGTTGTGGGTGTAGG-BHQ-3'.

3. The application of the DNA methylation site markers of claim 1 and the primer-probe set of claim 2 in the preparation of hepatocellular carcinoma diagnostic products.

4. The application according to claim 3, characterized in that: The diagnosis of hepatocellular carcinoma includes distinguishing hepatocellular carcinoma from cirrhosis, chronic hepatitis B, metabolic dysfunction-related fatty liver disease, and / or healthy individuals.

5. The application according to claim 3, characterized in that: The product is a multiplex dPCR kit.

6. A multiplex dPCR kit for the diagnosis of hepatocellular carcinoma, characterized in that: The multiplex dPCR kit includes the primer and probe set as described in claim 2.

7. The multiplex dPCR kit according to claim 6, characterized in that: The multiplex dPCR kit includes two reaction systems: the first reaction system is used to detect cg16990168, cg02829688, cg03760839 and cg13080379 sites, and the second reaction system is used to detect cg10703826, cg12664119 and cg23371746 sites and the internal reference gene ACTB.

8. The multiplex dPCR kit according to claim 6, characterized in that: The reaction program of the multiplex dPCR kit includes: pre-incubation at 50°C for 3 min, pre-denaturation at 95°C for 2 min; followed by 40 cycles, each cycle including denaturation at 95°C for 15 s and annealing at 56°C for 30 s.

9. A method for constructing a diagnostic model for hepatocellular carcinoma, characterized in that: Includes the following steps: 1) Collect plasma samples from the training cohort used as a diagnostic model for hepatocellular carcinoma; 2) Extract cfDNA from plasma samples and perform sulfite conversion treatment. Then, use the multiplex dPCR kit described in any one of claims 6 to 8 to detect cg02829688, cg13080379, cg03760839, cg10703826, cg12664119, cg16990168 and cg23371746 sites on the converted cfDNA to obtain the methylation level of DNA methylation site markers for each sample. 3) Using methylation level as a feature, a hepatocellular carcinoma diagnostic model is obtained by training with XGBoost or LR machine learning classification algorithms.

10. A diagnostic model for hepatocellular carcinoma, characterized in that: It is constructed by the method described in claim 9.

Images