Construction method and verification method of melanoma prognosis model based on lactate gene methylation characteristics
By constructing a melanoma prognostic model based on lactation gene methylation characteristics, and using hierarchical clustering and Cox regression models to screen differentially methylated sites, the problem of insufficient stability of existing models across different datasets was solved, achieving a more accurate and robust prognostic assessment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GANNAN MEDICAL UNIV
- Filing Date
- 2026-02-28
- Publication Date
- 2026-06-05
AI Technical Summary
Existing melanoma prognostic risk models rely on absolute methylation levels, which are susceptible to batch effects, resulting in insufficient stability and applicability when applied across different datasets.
By constructing a prognostic model based on the methylation characteristics of lactic acid genes, hierarchical clustering and Cox regression models were used to screen differentially methylated site pairs, a methylated site pair matrix was constructed, and the technical bias was reduced by utilizing the relative methylation order relationship, thus constructing a robust prognostic biomarker model.
It improves the accuracy and robustness of melanoma prognosis prediction, can be effectively applied to samples from different sources, has good external applicability and clinical applicability, and supports personalized prognostic assessment.
Smart Images
Figure CN122157752A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of melanoma prognostic technology, and in particular to a method for constructing and validating a melanoma prognostic model based on the methylation characteristics of lactation genes. Background Technology
[0002] Melanoma is a malignant tumor caused by melanocytes in the skin. Over the past few decades, the global incidence of melanoma has shown a significant upward trend, and the five-year survival rate for patients with advanced melanoma is less than 20%. Prognostic biomarkers can help clinicians predict patient outcomes and develop individualized treatment plans. Therefore, identifying molecular markers associated with melanoma prognosis is of significant clinical importance.
[0003] Previous studies have shown that abnormal lactation modifications are closely related to tumor development and progression. DNA methylation is an important epigenetic modification that can affect lactate production and metabolism by regulating key glycolytic enzymes. Therefore, studying the methylation status of lactation genes is beneficial for understanding the relationship between lactation and the development and progression of melanoma.
[0004] In recent years, researchers have constructed many prognostic risk models for melanoma based on DNA methylation. However, these models mainly rely on absolute methylation levels, which are highly susceptible to batch effects across different datasets, thus affecting the translation and application of biomarkers in clinical practice. Summary of the Invention
[0005] Therefore, it is necessary to provide a method for constructing and validating a melanoma prognostic model based on the methylation characteristics of lactation genes to address the aforementioned technical problems.
[0006] A method for constructing a melanoma prognostic model based on lactation gene methylation characteristics includes the following steps:
[0007] DNA methylation data and clinical information of melanoma were obtained from the Cancer Genome Atlas Database and the Gene Expression Comprehensive Database, respectively. The DNA methylation data were preprocessed, and the DNA methylation data from the Cancer Genome Atlas Database was used as the training set and the DNA methylation data from the Gene Expression Comprehensive Database was used as the validation set. Based on the methylation sites corresponding to the lactation-related gene set, a hierarchical clustering method was used to cluster the training set samples. Differential methylation sites were screened according to the subtypes obtained from the clustering, candidate methylation site pairs were constructed, and a methylation site pair matrix was constructed based on the relative methylation order relationship of the methylation sites. The methylation site pair matrix was screened using a Cox regression model to obtain methylation site pairs that were significantly associated with the prognosis of melanoma, and the C-index value, P-value, and hazard ratio of the obtained methylation site pairs were evaluated. Using the methylation site pair with the highest C-index value obtained from the forward selection order as the seed, a greedy algorithm is used to screen methylation site pairs that are significantly associated with melanoma prognosis. If adding a methylation site pair increases the C-index value, the current methylation site is retained; otherwise, the current methylation site is not added. This process continues until the C-index value no longer increases, resulting in a methylation site pair combination. The methylation site pair combination is then validated using a validation set, and the validated methylation site pair combination is used as a prognostic biomarker model for lactation-related gene methylation features in melanoma.
[0008] In one embodiment, the preprocessing includes: Remove methylation sites with a missing value of more than 70% in DNA methylation data, non-CpG methylation sites, methylation sites that interact with SNPs, methylation sites with cross-reactivity, and methylation sites located on sex chromosomes. The k-nearest neighbor method was used to complete the removed methylation sites in the DNA methylation data, and the methylation sites detected by both the training and validation sets were extracted after completion.
[0009] In one embodiment, the training set is melanoma DNA methylation data from the Cancer Genome Atlas database; the validation set is melanoma DNA methylation data of GSE51547 and GSE144487 from the Gene Expression Synthesis Database.
[0010] In one embodiment, the lactation-related gene set is derived from the MSigDB database, which consists of the union of genes from five entries retrieved using the keyword "lactic". The five entries are GOBP_LACTATE_METABOLIC_PROCESS, HP_INCREASED_CIRCULATING_LACTATE_CONCENTRATION, HP_LACTIC_ACIDOSIS, HP_LACTICACIDURIA, and HP_SEVERE_LACTIC_ACIDOSIS.
[0011] In one embodiment, differentially methylated sites are screened based on subtypes obtained from clustering, and candidate methylation site pairs are constructed, including: Differential methylation analysis was performed on the subtypes obtained from clustering. The p-values were corrected using the Benjamini-Hochberg method. Methylation sites with a corrected p-value less than 0.05 and |Δβ|>0.1 were selected as differential methylation sites. Differentially methylated sites are combined in pairs to obtain candidate methylation site pairs.
[0012] In one embodiment, the methylation site pair matrix is screened using a Cox regression model to obtain methylation site pairs that are significantly associated with melanoma prognosis, including: Methylation sites with a p-value less than 0.05 after Benjamini-Hochberg correction were identified as methylation sites significantly associated with melanoma prognosis. The methylation site pair matrix was screened using a Cox regression model to obtain methylation site pairs significantly associated with melanoma prognosis.
[0013] In one embodiment, the methylation site pair matrix is screened using a Cox regression model to obtain methylation site pairs that are significantly associated with melanoma prognosis, and the method further includes: If the risk ratio corresponding to a relative methylation order value of 1 for a methylation site pair is less than 1, then the site positions of the current methylation site pair are swapped.
[0014] A method for validating a melanoma prognostic model based on lactation gene methylation features includes the following steps: DNA methylation data and clinical information of melanoma were obtained from the Cancer Genome Atlas Database and the Gene Expression Comprehensive Database, respectively. The DNA methylation data were preprocessed, and the DNA methylation data from the Cancer Genome Atlas Database was used as the training set and the DNA methylation data from the Gene Expression Comprehensive Database was used as the validation set. Based on the methylation sites corresponding to the lactation-related gene set, a hierarchical clustering method was used to cluster the training set samples. Differential methylation sites were screened according to the subtypes obtained from the clustering, candidate methylation site pairs were constructed, and a methylation site pair matrix was constructed based on the relative methylation order relationship of the methylation sites. The methylation site pair matrix was screened using a Cox regression model to obtain methylation site pairs that were significantly associated with the prognosis of melanoma, and the C-index value, P-value, and hazard ratio of the obtained methylation site pairs were evaluated. Using the methylation site pair with the highest C-index value obtained through forward selection as the seed, a greedy algorithm is used to screen methylation site pairs that are significantly associated with melanoma prognosis. If adding a methylation site pair increases the C-index value, the current methylation site is retained; otherwise, it is not added. This process continues until the C-index value no longer increases, resulting in a methylation site pair combination. The methylation site pair combination is then validated using a validation set, and the validated methylation site pair combination is used as a prognostic biomarker model for lactation-related gene methylation features in melanoma. The prognostic biomarker model for lactation-related gene methylation characteristics of melanoma was validated on a validation set, including: survival analysis, multivariate Cox regression analysis, functional enrichment analysis, immune infiltration analysis, immunotherapy efficacy analysis, and chemotherapy drug sensitivity analysis.
[0015] The aforementioned method for constructing and validating a melanoma prognostic model based on lactation gene methylation features utilizes DNA methylation data from the Cancer Genome Atlas database as the training set and DNA methylation data from the Gene Expression Comprehensive Database as the validation set. This ensures the constructed melanoma lactation-related gene methylation prognostic biomarker model has good external applicability and can be used for prognostic assessment of melanoma samples from different sources. By constructing a methylation site pair matrix based on the relative methylation order of methylation sites, the technical bias caused by data from different platforms, batches, and laboratories is significantly reduced, effectively avoiding batch effects between different datasets. It can be directly applied to independent sample sets without additional correction, exhibiting excellent robustness and clinical applicability. By using methylation sites corresponding to lactation-related gene sets as anchors, hierarchical clustering and differential methylation site screening are performed to identify prognostic-related functional sites, improving the reliability and accuracy of the constructed prognostic biomarker model. The prognostic biomarker model constructed in this invention can achieve personalized prognostic prediction for patients, providing potential technical support for prognostic assessment of melanoma patients. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating a method for constructing a melanoma prognostic model based on lactation gene methylation features in one embodiment. Figure 2This is a schematic diagram illustrating the establishment of a methylation site pair matrix in one embodiment. Detailed Implementation
[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0018] In one embodiment, such as Figure 1 As shown, a method for constructing a melanoma prognostic model based on lactation gene methylation features is provided, including the following steps: Step 201: Obtain DNA methylation data and clinical information of melanoma from the Cancer Genome Atlas Database and the Gene Expression Comprehensive Database, respectively. Preprocess the DNA methylation data, using the DNA methylation data from the Cancer Genome Atlas Database as the training set and the DNA methylation data from the Gene Expression Comprehensive Database as the validation set.
[0019] It should be noted that, along with the DNA methylation data, corresponding RNA-seq and somatic mutation data were also collected simultaneously. RNA-seq was used to calculate the cytolytic activity score, and somatic mutation data was used to calculate the tumor mutation burden. Clinical information was used for multivariate Cox regression.
[0020] Step 202: Based on the methylation sites corresponding to the lactation-related gene set, hierarchical clustering is used to cluster the training set samples. Differential methylation sites are screened according to the subtypes obtained from the clustering, candidate methylation site pairs are constructed, and a methylation site pair matrix is constructed based on the relative methylation order relationship of the methylation sites.
[0021] Step 203: The methylation site pair matrix is screened using a Cox regression model to obtain methylation site pairs that are significantly associated with the prognosis of melanoma, and the C-index value, P-value, and hazard ratio of the obtained methylation site pairs are evaluated.
[0022] Step 204: Using the methylation site pair with the largest C-index value obtained from the forward selection order as the seed, a greedy algorithm is used to screen methylation site pairs that are significantly related to the prognosis of melanoma. If adding a methylation site pair increases the C-index value, the current methylation site is retained; if the C-index value does not increase, the current methylation site is not added. This process continues until the C-index value no longer increases, resulting in a combination of methylation site pairs. The combination of methylation site pairs is then validated using a validation set. The validated combination of methylation site pairs is used as a prognostic biomarker model for the methylation characteristics of lactation-related genes in melanoma.
[0023] The aforementioned method for constructing a melanoma prognostic model based on lactation gene methylation features utilizes DNA methylation data from the Cancer Genome Atlas database as the training set and DNA methylation data from the Gene Expression Comprehensive Database as the validation set. This ensures the constructed melanoma lactation-related gene methylation prognostic biomarker model has good external applicability and can be used for prognostic assessment of melanoma samples from different sources. By constructing a methylation site pair matrix based on the relative methylation order of methylation sites, the technical bias caused by data from different platforms, batches, and laboratories is significantly reduced, effectively avoiding batch effects between different datasets. It can be directly applied to independent sample sets without additional correction, exhibiting excellent robustness and clinical applicability. By using methylation sites corresponding to lactation-related gene sets as anchors, hierarchical clustering and differential methylation site screening are performed to identify prognostic-related functional sites, improving the reliability of the constructed prognostic biomarker model and enhancing the accuracy of prognostic prediction. The prognostic biomarker model constructed in this invention can achieve personalized prognostic prediction for patients, providing potential technical support for prognostic assessment of melanoma patients.
[0024] In one embodiment, the DNA methylation data obtained included 699 samples.
[0025] In one embodiment, the training set is melanoma DNA methylation data from the Cancer Genome Atlas database; the validation set is melanoma DNA methylation data of GSE51547 and GSE144487 from the Gene Expression Synthesis Database.
[0026] In one embodiment, the preprocessing includes: Remove methylation sites with a missing value of more than 70% in DNA methylation data, non-CpG methylation sites, methylation sites that interact with SNPs, methylation sites with cross-reactivity, and methylation sites located on sex chromosomes. The k-nearest neighbor method was used to complete the removed methylation sites in the DNA methylation data, and the methylation sites detected by both the training and validation sets were extracted after completion.
[0027] In this embodiment, methylation sites with a missing value ratio exceeding 70% are removed to reduce the interference of highly missing data on the stability of the statistical model. Non-CpG sites, SNP-related sites, and cross-reactive probes are eliminated to reduce the impact of technical bias and genotype differences on the methylation signal, thereby improving the specificity and accuracy of the signal. Sites located on sex chromosomes are excluded to avoid systematic bias caused by sex differences. Based on this, the k-nearest neighbor algorithm is used to complete missing values, reducing data loss and improving sample utilization while preserving the sample information structure. Furthermore, methylation sites detected jointly in both the training and validation sets are extracted to ensure feature consistency and comparability, enhancing the robustness and generalization ability of subsequent model construction and external validation. Through preprocessing, the quality and analytical reliability of DNA methylation data are improved.
[0028] Specifically, in one embodiment, the number of methylation sites detected jointly by the training set and the validation set is 377,714.
[0029] In one embodiment, the lactation-related gene set is derived from the MSigDB database, which consists of the union of genes from five entries retrieved using the keyword "lactic". The five entries are GOBP_LACTATE_METABOLIC_PROCESS, HP_INCREASED_CIRCULATING_LACTATE_CONCENTRATION, HP_LACTIC_ACIDOSIS, HP_LACTICACIDURIA, and HP_SEVERE_LACTIC_ACIDOSIS.
[0030] Specifically, the MSigDB database contains 284 genes, corresponding to 2885 methylation sites.
[0031] In one embodiment, differentially methylated sites are screened based on subtypes obtained from clustering, and candidate methylation site pairs are constructed, including: Differential methylation analysis was performed on the subtypes obtained from clustering. The p-values were corrected using the Benjamini-Hochberg method. Methylation sites with a corrected p-value less than 0.05 and |Δβ|>0.1 were selected as differential methylation sites. Differentially methylated sites are combined in pairs to obtain candidate methylation site pairs.
[0032] Specifically, refer to Figure 2Based on the relative methylation order of each methylation site pair in each sample, the actual methylation level in each sample is replaced with the relative size relationship between methylation sites to obtain the relative expression order matrix X of candidate prognostic related methylation site pairs; matrix X is a 0-1 matrix, where 1 represents a methylation site pair. In the sample i The expression order relationship in the methylation site is Larger than methylation site ;0 indicates that the methylation site is positive for the methylation site. In the sample i The expression order relationship in the methylation site is Less than or equal to methylation site Methylation sites with a corrected P-value less than 0.05 and |Δβ| > 0.1 were selected as differentially methylated sites, where is the absolute value of the difference in methylation levels between the two groups of |Δβ|. Candidate methylation site pairs were constructed. It is understood that the constructed candidate methylation site pairs are methylation site pairs that are significantly associated with the prognosis of melanomas with lactation-related gene features.
[0033] In one embodiment, the methylation site pair matrix is screened using a Cox regression model to obtain methylation site pairs that are significantly associated with melanoma prognosis, and the method further includes: If the risk ratio corresponding to a relative methylation order value of 1 for a methylation site pair is less than 1, then the site positions of the current methylation site pair are swapped.
[0034] In this embodiment, by swapping the site positions of the current methylation site pairs, it is ensured that a state with a methylation order value of 1 represents a high risk.
[0035] It should be understood that although the steps in the flowchart are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order requirement for the execution of these steps, and they can be executed in other orders. Furthermore, Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.
[0036] In one embodiment, a method for validating a melanoma prognostic model based on lactation gene methylation features is provided, comprising the following steps: Step 901: Obtain DNA methylation data and clinical information of melanoma from the Cancer Genome Atlas Database and the Gene Expression Comprehensive Database, respectively. Preprocess the DNA methylation data, using the DNA methylation data from the Cancer Genome Atlas Database as the training set and the DNA methylation data from the Gene Expression Comprehensive Database as the validation set.
[0037] Step 902: Based on the methylation sites corresponding to the lactation-related gene set, hierarchical clustering is used to cluster the training set samples. Differential methylation sites are screened according to the subtypes obtained from the clustering, candidate methylation site pairs are constructed, and a methylation site pair matrix is constructed based on the relative methylation order relationship of the methylation sites.
[0038] Step 903: The methylation site pair matrix is screened using a Cox regression model to obtain methylation site pairs that are significantly associated with the prognosis of melanoma, and the C-index value, P-value, and hazard ratio of the obtained methylation site pairs are evaluated.
[0039] Step 904: Using the methylation site pair with the largest C-index value obtained from the forward selection order as the seed, a greedy algorithm is used to screen methylation site pairs that are significantly related to the prognosis of melanoma. If adding a methylation site pair increases the C-index value, the current methylation site is retained; if the C-index value does not increase, the current methylation site is not added. This process continues until the C-index value no longer increases, resulting in a combination of methylation site pairs. The combination of methylation site pairs is then validated using a validation set. The validated combination of methylation site pairs is used as a prognostic biomarker model for lactation-related gene methylation features in melanoma.
[0040] The prognostic biomarker model for lactation-related gene methylation characteristics of melanoma was validated on a validation set, including: survival analysis, multivariate Cox regression analysis, functional enrichment analysis, immune infiltration analysis, immunotherapy efficacy analysis, and chemotherapy drug sensitivity analysis.
[0041] In one embodiment, survival analysis and multivariate Cox regression analysis were performed on a prognostic biomarker model of lactation-related gene methylation characteristics in melanoma, including: Based on candidate prognostic risk marker combinations, the relative methylation order sum (RMOComb) value of each combination in the training set is calculated. A prognostic risk model is constructed based on the RMOComb value of each sample, and a risk threshold is determined based on the point with the largest Youden index on the 5-year ROC curve in the training set. If the RMOComb value is greater than the threshold, it is considered high risk; otherwise, it is considered low risk. Kaplan-Meier survival analysis is performed using "survival" and "survminer" in the R package to verify whether there is a survival difference between the high and low risk groups predicted by the prognostic marker model of lactation-related genes in melanoma. Finally, multivariate Cox regression analysis is performed on the training and validation sets in combination with age and stage information.
[0042] Specifically, a prognostic biomarker model based on the methylation characteristics of lactation-related genes in melanoma was used to score the risk of samples in each dataset, and the high and low risk of samples were determined based on risk thresholds. In the training set, based on the order of the prognostic biomarker model based on the methylation characteristics of lactation-related genes in melanoma, 237 and 221 samples were assigned to the high and low risk groups, respectively, and there was a significant difference in survival analysis between the two groups (P<0.0001, HR=0.20, 95%CI: 0.15–0.27). In the two validation sets, 29 and 18, and 82 and 112 samples, respectively, were assigned to the high and low risk groups, and there were significant differences in survival analysis between the two groups (P=0.016, HR=0.45, 95%CI: 0.24–0.87) and (P=0.00019, HR=0.51, 95%CI: 0.36–0.73). Kaplan-Meier survival curves showed that, in the training set, high-risk melanoma patients had lower overall survival, while low-risk patients generally exhibited longer survival times, with a significant difference between the two groups. The mean AUC in the training set was 0.757; the mean AUC in the validation set GSE51547 was 0.64; and the mean AUC in the validation set GSE144487 was 0.63. This indicates that methylation markers have significant prognostic value.
[0043] In one embodiment, functional enrichment analysis was performed on a prognostic biomarker model of lactation-related gene methylation characteristics in melanoma, including: A Bayesian modified t-test using the limma linear model was employed to identify differentially methylated sites between different groups. A site was defined as a differentially methylated site if the p-value after Benjamini-Hochberg correction was less than 0.05 and |Δβ|>0.1. Functional enrichment analysis of methylated sites of interest was performed using the R package “missMethyl” based on gene ontology and the Kyoto Encyclopedia of Genes and Genomes database.
[0044] Specifically, based on the Kyoto Encyclopedia of Genes and Genomes and Gene Ontology, functional enrichment analysis was performed on differentially methylated sites between high- and low-risk patients in a prognostic biomarker model of lactation-related gene methylation characteristics in melanoma. The enrichment results showed that differentially methylated sites in the Gene Ontology database were mainly enriched in biological processes such as immune system processes, immune cell activation, and signal transduction regulation. In the Kyoto Encyclopedia of Genes and Genomes database, these enriched gene pathways mainly included chemokine signaling pathways, hematopoietic cell lineages, and Th1 and Th2 cell differentiation pathways. These pathways are closely related to the formation of the tumor immune microenvironment, immune escape mechanisms, and the regulation of anti-tumor immune responses.
[0045] In one embodiment, an immune infiltration analysis was performed on a prognostic biomarker model of lactation-related gene methylation characteristics in melanoma, including: The infiltration levels of these seven immune cell types in the sample were scored using the R package “EpiDISH” with reference to methylation sites specific to seven immune cell types (B cells, NK cells, CD4+ T cells, CD8+ T cells, monocytes, neutrophils, and eosinophils); the Wilcoxon test was used for comparison.
[0046] The results showed that patients in the low-risk group had higher levels of CD8+ T cells and B cells, while patients in the high-risk group had higher levels of monocytes (P<0.05, Wilcoxon test).
[0047] In one embodiment, an analysis of the immunotherapy efficacy was performed on a prognostic biomarker model of lactation-related gene methylation characteristics in melanoma, including: The differences in immunotherapy response between high- and low-risk groups were assessed based on immunophenotypic scores, cytolytic activity scores, and tumor mutational burden. Immunophenotypic score data for melanoma patients were obtained from the Cancer Immunoglossary Database. All patients were divided into four groups based on the expression status (positive or negative) of PD-1 and CTLA-4, and the differences in immunophenotypic scores between the high-risk and low-risk groups within each group were compared. Cytolytic activity scores were represented by the geometric mean of granzyme A and perforin gene expression. Somatic mutations were extracted using the R package "maftools" to calculate the tumor mutational burden for each patient. The Wilcoxon test was used for comparison.
[0048] The results showed that patients with low immunophenotypic scores had significantly higher scores than those with high scores across the four groups (P<0.05, Wilcoxon test). Analysis of cytolytic activity scores revealed higher scores in the low-risk group (P<0.05, Wilcoxon test). Wilcoxon test results also showed a higher mutation frequency in low-risk patients (P<0.05, Wilcoxon test).
[0049] In one embodiment, chemotherapy drug sensitivity analysis was performed on a prognostic biomarker model of lactation-related gene methylation characteristics in melanoma, including: Based on molecular markers of cancer cell drug response and drug sensitivity data provided in the anticancer drug sensitivity genomics database, the R package "oncoPredict" was used to analyze several commonly used chemotherapeutic drugs (including dacarbazine, cisplatin, and temozolomide) in melanoma. Chemotherapy drug sensitivity was represented by the half-maximal inhibitory concentration (WMC) of each tumor sample, and comparisons were made using the Wilcoxon test. Results showed that the WMC values of patients in the high-risk group were significantly higher than those in the low-risk group, indicating less sensitivity to chemotherapy (P<0.05, Wilcoxon test).
[0050] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0051] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A method for constructing a melanoma prognostic model based on lactation gene methylation characteristics, characterized in that, Includes the following steps: DNA methylation data and clinical information of melanoma were obtained from the Cancer Genome Atlas Database and the Gene Expression Comprehensive Database, respectively. The DNA methylation data were preprocessed, and the DNA methylation data from the Cancer Genome Atlas Database was used as the training set and the DNA methylation data from the Gene Expression Comprehensive Database was used as the validation set. Based on the methylation sites corresponding to the lactation-related gene set, a hierarchical clustering method was used to cluster the training set samples. Differential methylation sites were screened according to the subtypes obtained from the clustering, candidate methylation site pairs were constructed, and a methylation site pair matrix was constructed based on the relative methylation order relationship of the methylation sites. The methylation site pair matrix was screened using a Cox regression model to obtain methylation site pairs that were significantly associated with the prognosis of melanoma, and the C-index value, P-value, and hazard ratio of the obtained methylation site pairs were evaluated. Using the methylation site pair with the highest C-index value obtained from the forward selection order as the seed, a greedy algorithm is used to screen methylation site pairs that are significantly associated with melanoma prognosis. If adding a methylation site pair increases the C-index value, the current methylation site is retained; otherwise, the current methylation site is not added. This process continues until the C-index value no longer increases, resulting in a methylation site pair combination. The methylation site pair combination is then validated using a validation set, and the validated methylation site pair combination is used as a prognostic biomarker model for lactation-related gene methylation features in melanoma.
2. The method for constructing a melanoma prognostic model based on lactation gene methylation characteristics according to claim 1, characterized in that, The preprocessing includes: Remove methylation sites with a missing value of more than 70% in DNA methylation data, non-CpG methylation sites, methylation sites that interact with SNPs, methylation sites with cross-reactivity, and methylation sites located on sex chromosomes. The k-nearest neighbor method was used to complete the removed methylation sites in the DNA methylation data, and the methylation sites detected by both the training and validation sets were extracted after completion.
3. The method for constructing a melanoma prognostic model based on lactation gene methylation characteristics according to claim 1, characterized in that, The training set consists of melanoma DNA methylation data from the Cancer Genome Atlas Database; the validation set consists of melanoma DNA methylation data from the Gene Expression Comprehensive Database for GSE51547 and GSE144487.
4. The method for constructing a melanoma prognostic model based on lactation gene methylation characteristics according to claim 1, characterized in that, The set of lactation-related genes comes from the MSigDB database, which consists of the union of genes from five entries retrieved using the keyword "lactic". The five entries are GOBP_LACTATE_METABOLIC_PROCESS, HP_INCREASED_CIRCULATING_LACTATE_CONCENTRATION, HP_LACTIC_ACIDOSIS, HP_LACTICACIDURIA, and HP_SEVERE_LACTIC_ACIDOSIS.
5. The method for constructing a melanoma prognostic model based on lactation gene methylation characteristics according to claim 1, characterized in that, Differential methylation sites were screened based on the subtypes obtained from clustering, and candidate methylation site pairs were constructed, including: Differential methylation analysis was performed on the subtypes obtained from clustering. The p-values were corrected using the Benjamini-Hochberg method. Methylation sites with a corrected p-value less than 0.05 and |Δβ|>0.1 were selected as differential methylation sites. Differentially methylated sites are combined in pairs to obtain candidate methylation site pairs.
6. The method for constructing a melanoma prognostic model based on lactation gene methylation characteristics according to claim 5, characterized in that, The methylation site pair matrix was screened using a Cox regression model to identify methylation site pairs that were significantly associated with melanoma prognosis, including: Methylation sites with a p-value less than 0.05 after Benjamini-Hochberg correction were identified as methylation sites significantly associated with melanoma prognosis. The methylation site pair matrix was screened using a Cox regression model to obtain methylation site pairs significantly associated with melanoma prognosis.
7. The method for constructing a melanoma prognostic model based on lactation gene methylation characteristics according to claim 6, characterized in that, The methylation site pair matrix was screened using a Cox regression model to identify methylation site pairs that were significantly associated with melanoma prognosis, including: If the risk ratio corresponding to a relative methylation order value of 1 for a methylation site pair is less than 1, then the site positions of the current methylation site pair are swapped.
8. A method for validating a melanoma prognostic model based on lactation gene methylation characteristics, characterized in that, Includes the following steps: DNA methylation data and clinical information of melanoma were obtained from the Cancer Genome Atlas Database and the Gene Expression Comprehensive Database, respectively. The DNA methylation data were preprocessed, and the DNA methylation data from the Cancer Genome Atlas Database was used as the training set and the DNA methylation data from the Gene Expression Comprehensive Database was used as the validation set. Based on the methylation sites corresponding to the lactation-related gene set, a hierarchical clustering method was used to cluster the training set samples. Differential methylation sites were screened according to the subtypes obtained from the clustering, candidate methylation site pairs were constructed, and a methylation site pair matrix was constructed based on the relative methylation order relationship of the methylation sites. The methylation site pair matrix was screened using a Cox regression model to obtain methylation site pairs that were significantly associated with the prognosis of melanoma, and the C-index value, P-value, and hazard ratio of the obtained methylation site pairs were evaluated. Using the methylation site pair with the highest C-index value obtained through forward selection as the seed, a greedy algorithm is used to screen methylation site pairs that are significantly associated with melanoma prognosis. If adding a methylation site pair increases the C-index value, the current methylation site is retained; otherwise, it is not added. This process continues until the C-index value no longer increases, resulting in a methylation site pair combination. The methylation site pair combination is then validated using a validation set, and the validated methylation site pair combination is used as a prognostic biomarker model for lactation-related gene methylation features in melanoma. The prognostic biomarker model for melanoma based on lactation-related gene methylation was validated on a validation set, including: survival analysis, multivariate Cox regression analysis, functional enrichment analysis, immune infiltration analysis, immunotherapy efficacy analysis, and chemotherapy sensitivity analysis.