Construction method and application of DNA methylation site identification model
By constructing a two-step model based on DNA methylation sites, the problem of distinguishing pathological types of primary liver cancer is solved, and the accurate identification of hepatocellular carcinoma and intrahepatic cholangiocarcinoma is achieved, providing an efficient biomarker to support treatment choices.
Patent Information
- Application Number
- CN202411451080.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-17
- Publication Date
- 2025-03-07
AI Technical Summary
The prior art is difficult to accurately distinguish the pathological types of primary liver cancer, especially hepatocellular carcinoma and intrahepatic cholangiocarcinoma, resulting in risks in treatment choice and follow-up plan.
By screening DNA methylation sites, a two-step model based on differentially methylated CpG sites was constructed to distinguish primary liver cancer and distinguish hepatocellular carcinoma from intrahepatic cholangiocarcinoma.
Accurate identification of primary liver cancer is achieved, and can distinguish hepatocellular carcinoma from intrahepatic cholangiocarcinoma with high sensitivity and specificity, providing valuable biomarkers to support therapeutic decisions.
Smart Images
Figure BDA0005089144390000031 
Figure BDA0005089144390000041 
Figure HDA0005089144400000011
Abstract
Description
Technical Field
[0001] The invention belongs to the field of biomedicine, and specifically relates to a method for constructing a DNA methylation site identification model and an application thereof. Background Art
[0002] Primary liver cancer (PLC) is the fourth leading cause of cancer-related death. The number of PLC patients is increasing worldwide, which will lead to serious health problems and heavy economic burden. In addition to the rare hepatocellular cholangiocarcinoma (HChC), the most common pathological types of PLC are hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC). The clinical symptoms and imaging manifestations of the two are similar, such as round masses in the liver, abdominal pain, abdominal distension, etc.; but they are completely different in etiology, pathogenesis and treatment.
[0003] Hepatocellular carcinoma accounts for 70%-80% of PLC, and hepatocellular carcinoma originates from hepatocytes. Hepatitis virus infection, excessive drinking, autoimmune diseases, aflatoxin, etc. are risk factors for its occurrence and development. ICC is the second most common PLC, accounting for 8%-15% of liver malignancies. It originates from intrahepatic bile duct epithelial cells and is mainly related to biliary diseases such as sclerosing cholangitis and intrahepatic bile duct stones. Since the treatment methods of HCC and ICC are completely different, inaccurate identification may have catastrophic consequences for patients. Accurate identification of the pathological type of PLC is of great significance for selecting appropriate treatment methods and formulating follow-up plans. Previous studies have shown that serum biomarkers carbohydrate antigen 19-9 (CA19-9) and alpha-fetoprotein (AFP) can be used to distinguish HCC from ICC, but their sensitivity and specificity are not ideal. Therefore, some strategies are urgently needed to improve the accuracy of distinguishing HCC from ICC.
[0004] DNA methylation of promoter CpG islands (CGIs) of tumor suppressor genes (TSGs) is an important mechanism that often induces the occurrence and development of various cancers. Abnormal methylation of promoter CpG sites of TSGs can change the spatial structure of chromatin, resulting in low or no expression of tumor suppressor genes. Recent studies have shown that abnormal gene methylation is closely related to the occurrence of HCC and ICC, and has a potential role in screening biomarkers and targets. Previous studies have shown that circulating tumor DNA methylation biomarkers can be used to distinguish HCC from normal tissues with a sensitivity of 85.7% and a specificity of 94.3%. In addition, DNA methylation of 10 CpG sites can be used to distinguish tumors from normal tissues in patients with liver cancer with a sensitivity of 86% and a specificity of 100%. However, these studies have only focused on differential markers for HCC. Few studies have focused on differential markers for distinguishing HCC from ICC.
[0005] In this context, the present invention aims to develop useful biomarkers by screening DNA methylation sites, which can be used not only to identify PLC but also to distinguish HCC from ICC, which is crucial for the selection of subsequent treatment options. Summary of the invention
[0006] The purpose of the present invention is to develop biomarkers for distinguishing hepatocellular carcinoma and intrahepatic cholangiocarcinoma by screening DNA methylation sites and constructing a differentiation model.
[0007] The technical solution of the present invention is:
[0008] A method for constructing a DNA methylation site identification model, characterized in that the identification model is constructed by screening DNA methylation sites, and the identification model is based on DNA methylation sites.
[0009] The discrimination model independently contains several differentially methylated CpG (DMC) sites.
[0010] A two-step model based on differentially methylated CpG (DMC) sites was established; the first step was to identify primary liver cancer (PLC); the second step was to distinguish hepatocellular carcinoma (HCC) from intrahepatic cholangiocarcinoma (ICC).
[0011] The two-step model, the first step is to identify primary liver cancer (PLC), the differentially methylated CpG (DMC) sites are: cg24035245, cg21072795, cg00261162, ch.7.135065R, cg17569842, cg26361533, cg14373727.
[0012] The number of differentially methylated CpG (DMC) sites is 2, 3, 4, 5, 6, and 7, among which the first 3 are preferred.
[0013] The second step was to differentiate hepatocellular carcinoma (HCC) from intrahepatic cholangiocarcinoma (ICC), and the differentially methylated CpG (DMC) sites were: cg17769836, cg17591574, cg07823562, and cg05663031.
[0014] The number of differentially methylated CpG (DMC) sites is 2, 3, or 4, with the first 3 being preferred.
[0015] The application of the identification model constructed by the method described above in the identification of biomarkers for hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC).
[0016] The key discovery of the present invention is the identification of several specific methylation sites as potential biomarkers that can not only distinguish PLC from normal, but also distinguish HCC from ICC.
[0017] The process of the present invention is to screen candidate sites with the highest Gini value from differentially methylated CpG (DMC) sites by using a random forest method, and then confirm the combination with the highest area under the curve (AUC) and the least number of required methylation sites in the training data set and the validation data set. The present invention proposes a method to prevent model overfitting. We calculated the AUC of the two models respectively, and the AUC will increase with the increase of the number of sites included in the training set; and after the AUC of different site combinations in the validation set reaches a peak, it will decrease with the increase of the included sites. In order to reduce overfitting, we selected the minimum site combination that can reach the AUC peak to achieve its cost-effectiveness.
[0018] The most preferred technical solution of the present invention is: a two-step model based on differential methylation sites is established. First, PLC is identified using cg24035245, cg21072795 and cg00261162. If PLC is considered, HCC and ICC are further distinguished using cg17769836, cg17591574 and cg07823562. By detecting these 6 methylation sites, we can identify the pathological subtypes of PLC with high sensitivity and specificity. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 The distribution of differentially methylated sites and genes in HCC and ICC; the distribution of differentially methylated sites of CpG islands in HCC (A) and ICC (E); the distribution of differentially methylated genes of CpG islands in HCC (B) and ICC (F); the distribution of the distances between differentially methylated sites and transcription start sites (TSS) in HCC (C) and ICC (G); the distribution of the distances between differentially methylated genes and transcription start sites in HCC (D) and ICC (H);
[0020] Figure 2is the site screening of the model; Gini values of the top 10 differentially methylated sites for distinguishing PLC samples from normal samples (A); AUC curves of the model (PLC vs. normal) based on the combination of indicator sites in the training set and validation set (B); evaluation of the model (PLC vs. normal) based on three sites (C); receiver operating characteristic (ROC) curves of the model (PLC vs. normal) based on the training set (D) and validation set (E) of the three sites; Gini values of the top 10 differentially methylated sites for distinguishing HCC from ICC (F); AUC curves of the prediction model (HCC vs. ICC) based on the combination of indicator sites in the training set and validation set (G); evaluation of the model (HCC vs. ICC) based on three sites (H); receiver operating characteristic (ROC) curves of the model (HCC vs. ICC) based on three sites in the training set (I) and validation set (J);
[0021] Figure 3 is the effectiveness evaluation of the model; the efficiency of the model in distinguishing PLC and normal samples in the training set (A) and the validation set (C); the heat map (PLC vs. normal) of the actual status, predicted status, data source, and methylation indicator sites in the training set (B) and the validation set (D); the efficiency of the model in distinguishing HCC and ICC in the training set (E) and the validation set (G); the heat map (HCC vs. ICC) of the actual status, predicted status, data source, and methylation indicator sites in the training set (F) and the validation set (H). DETAILED DESCRIPTION
[0022] Those skilled in the art will understand that the following examples are used to further illustrate the present invention but are not intended to limit the present invention. If no specific techniques or conditions are specified in the examples, the techniques or conditions described in the literature in the art are used.
[0023] Abbreviations
[0024] PLC: primary liver cancer; HCC: hepatocellular carcinoma; ICC: intrahepatic cholangiocarcinoma; DMC: differentially methylated CpG; AUC: area under the curve; HChC: hepatocellular cholangiocarcinoma; CGI: CpG island; TSG: tumor suppressor gene.
[0025] Example 1. Identification of differentially methylated CpG sites
[0026] The HCC and ICC-related DNA methylation array datasets in Table 1 were downloaded from the UCSC Cancer Browser (https: / / xenabrowser.net / datapages / ) and the Gene Expression Omnibus (GEO) database (https: / / www.ncbi.nlm.nih). The probe removal criteria were as follows: (1) methylation site missing β value > 30%; (2) methylation sites on sex chromosomes; only methylation sites shared in the training dataset were retained. The Wilcoxon rank sum test was performed on the three groups of datasets (96 normal samples and 402 HCC samples, 96 normal samples and 108 ICC samples, and 108 ICC samples and 402 HCC samples), and the remaining missing values were filled using the Bioconductor package. Unless otherwise specified, sites with adjusted p-value < 0.05 and |log2FoldChange|> 1 (log2FC) were considered DMC sites.
[0027] Table 1. DNA methylation dataset
[0028]
[0029]
[0030] From the training dataset in Table 1, methylation data of 96 normal samples and 402 HCC samples were compared, and 8177 hypermethylated sites and 3152 hypomethylated sites were identified in HCC samples. In the comparison of 96 normal samples and 108 ICC samples, 33449 hypermethylated sites and 1049 hypomethylated sites were identified in ICC samples. Then, according to the genomic region, the distribution of these DMC sites and their corresponding genes was visualized, and we can see that regardless of HCC (attached Figure 1 A, 1B) or ICC (with Figure 1 E, 1F), hypermethylation mainly occurs in CpG islands. Figure 1 Compared with 1H, the proportion of hypomethylation in the HCC genome is higher ( Figure 1 C, 1D). In the promoter region, hypermethylation was predominant in both cancer types (Appendix Figure 1 C, 1D, 1G, 1H). This promoter hypermethylation and gene body hypomethylation are characteristic of solid tumors. The difference in gene body methylation levels also indicates heterogeneity between HCC and ICC, demonstrating the feasibility of using DMC sites as potential biomarkers.
[0031] Example 2. Construction and optimization of the model
[0032] The present invention uses the random forest method to screen 10 candidate sites with the highest Gini values from the DMC sites (common DMC sites of 96 normal samples and 108 ICC samples; common DMC sites of 96 normal samples and 402 HCC samples; 108 ICC samples and 402 HCC samples), and then confirms the combination with the highest AUC and the least number of required methylation sites in the training data set and the validation data set to avoid overfitting.
[0033] First, to distinguish PLC from normal samples, we selected 6565 common hypermethylated sites and 187 common hypomethylated sites in HCC and ICC samples as the PLC-specific site library. Then, based on the Gini coefficient of DMC sites calculated by the random forest method, we selected different combinations of DMC sites from the 1st to the 10th position (cg24035245, cg21072795, cg00261162, ch.7.135065R, cg17569842, cg26361533, cg14373727, cg11408493, ch.19.50335620F, cg16711650) as the discrimination model (Appendix Figure 2 A). As attached Figure 2 As shown in Figure B, the AUC value of the training set increases with the increase in the number of loci. After adding two loci, the AUC value is close to the peak; however, after adding seven loci, the AUC value of the validation set decreases. Therefore, in order to ensure the performance of the model and avoid overfitting, and considering economy and simplicity, we selected the first three loci (cg24035245, cg21072795, cg00261162) for model evaluation. After evaluation, the error rate of PLC and normal identification is less than 6% (see Appendix Figure 2 C), the AUC of the training set and the validation set were 0.991 and 0.979 respectively (Appendix Figure 2 D, 2E).
[0034] Next, we used 11759 DMC loci in ICC and HCC samples to construct a discrimination model for HCC and ICC according to the above method. The top 10 loci with the highest Gini coefficient (cg17769836, cg17591574, cg07823562, cg05663031, cg16366607, cg19485539, cg10686044, cg08417728, cg20392615, cg10446401) are shown in the attached figure. Figure 2 As shown in F, as the number of loci increases, the AUC of the training set increases significantly, but the AUC of the validation set drops sharply after more than 4 loci (see Appendix Figure 2G). Similarly, we selected the first three sites (cg17769836, cg17591574, cg07823562) for model evaluation. The results showed that the ICC identification error rate was about 10% (see Appendix Figure 2 H), slightly higher than HCC, the AUCs of the training set and validation set based on these three sites were 0.954 and 0.972, respectively (Appendix Figure 2 I, 2J).
[0035] Example 3. Evaluation of the effectiveness of the model
[0036] We still used the first three sites (cg24035245, cg21072795, and cg00261162) as examples to evaluate the PLC identification model. In the training set consisting of 96 normal samples and 510 PLC samples (108 ICC samples and 402 HCC samples), the true positive rate (sensitivity) and true negative rate (specificity) were 98.8% (504 / 510) and 94.8% (91 / 96), respectively (Appendix Figure 3 A). A validation set consisting of 252 normal samples and 600 PLC samples (96 ICC samples and 504 HCC samples) was combined. Because one PLC sample had missing values, its sensitivity and specificity were 97.3% (583 / 599) and 81% (204 / 252), respectively (Appendix Figure 3 C). Compared with the normal sample group, in both the training set and the validation set, cg24035245 was hypermethylated, and cg21072795 and cg00261162 were hypomethylated in the PLC sample group (Appendix Figure 3 B, 3D).
[0037] For the first three loci (cg17769836, cg17591574, and cg07823562) for distinguishing HCC from ICC, four HCC samples with missing values were deleted. The accuracy of distinguishing HCC in the training set and validation set was 95.8% (385 / 402) and 96.8% (484 / 500), respectively (Appendix Figure 3 E, 3G). Consistent with previous results, ICC has a low accuracy in the training set (Appendix Figure 3 E) was 89.8% (97 / 108), and in the validation set (Appendix Figure 3 Compared with HCC, the methylation levels of these three sites (cg17769836, cg17591574, cg07823562) were higher in ICC samples in both the training and validation sets (Appendix Figure 3 F, 3H).
[0038] The results showed that the discrimination model based on the random forest algorithm performed well in distinguishing PLC from normal and HCC from ICC.
[0039] In summary, in the present invention, we established a two-step model based on differential methylation sites. First, cg24035245, cg21072795 and cg00261162 were used to distinguish PLC from normal. If PLC is considered, cg17769836, cg17591574 and cg07823562 are used to further distinguish HCC from ICC. By detecting these 6 methylation sites, we can identify the pathological subtypes of PLC with high sensitivity and specificity.
[0040] The above general description of the present invention and the description of its specific implementation methods should not be understood as limiting the technical solution of the invention. According to the present invention, those skilled in the art can add, reduce or combine the technical features disclosed in the above general description and / or specific implementation methods (including examples) without violating the constituent elements of the invention involved, to form other technical solutions belonging to the invention, which are also within the protection scope of the present invention. The full scope of the present invention is given by the attached claims and any equivalents thereof.
Claims
1. A method for constructing a DNA methylation site identification model, characterized in that: By screening DNA methylation sites, an identification model is constructed, wherein the identification model is based on DNA methylation sites.
2. The method for constructing a DNA methylation site identification model according to claim 1, characterized in that: The discrimination model independently contains several differentially methylated CpG sites.
3. The method for constructing a DNA methylation site identification model according to claim 1 or 2, characterized in that: A two-step model based on differentially methylated CpG sites was established; the first step was to identify primary liver cancer; the second step was to distinguish hepatocellular carcinoma from intrahepatic cholangiocarcinoma.
4. The method for constructing a DNA methylation site identification model according to claim 3, characterized in that: The two-step model, the first step is to identify primary liver cancer, the differentially methylated CpG sites are: cg24035245, cg21072795, cg00261162, ch.7.135065R, cg17569842, cg26361533, cg14373727.
5. The method for constructing a DNA methylation site identification model according to claim 4, characterized in that: The number of differentially methylated CpG (DMC) sites is 2, 3, 4, 5, 6, and 7, among which the first 3 are preferred.
6. The method for constructing a DNA methylation site identification model according to claim 3, wherein: The second step was to distinguish hepatocellular carcinoma from intrahepatic cholangiocarcinoma. The differentially methylated CpG sites were: cg17769836, cg17591574, cg07823562, and cg05663031.
7. The method for constructing a DNA methylation site identification model according to claim 6, wherein: The number of differentially methylated CpG sites is 2, 3, or 4, with the first 3 being preferred.
8. Application of the identification model constructed by the method according to any one of claims 1 to 7 in the identification of biomarkers for hepatocellular carcinoma and intrahepatic cholangiocarcinoma.