A biomarker for osteosarcoma, its application, and a method for constructing a prognostic prediction model.

Abstract

This invention discloses an osteosarcoma biomarker, relating to the field of tumor prognostic prediction technology. The osteosarcoma biomarker includes the genes CGREF1, COL13A1, KIF25, and RPL35AP7. This invention also discloses a method for constructing an osteosarcoma prognostic prediction model, using the expression levels of the biomarkers CGREF1, COL13A1, KIF25, and RPL35AP7 and their corresponding multivariate analysis coefficients to construct the osteosarcoma prognostic prediction model. The osteosarcoma biomarker and its screening method provided by this invention, with osteoblasts as the core research object, avoid the problem of mutual masking of signals from different cells in overall transcriptome analysis. By mapping osteoblast-specific molecular characteristics to the patient's overall transcriptome data, it achieves clinical translational application.

Description

Technical Field

[0001] This invention relates to the field of tumor prognostic prediction technology. Specifically, it relates to a biomarker for osteosarcoma, its application, and a method for constructing a prognostic prediction model. Background Technology

[0002] Osteosarcoma is a highly malignant primary bone tumor, commonly affecting adolescents and children. It is characterized by rapid growth, strong invasiveness, and a high rate of recurrence and distant metastasis. Currently, the preferred treatment for osteosarcoma patients is neoadjuvant chemotherapy combined with surgery. This treatment has significantly improved the 5-year survival rate; however, once lung metastasis occurs, the 5-year survival rate remains below 20%. Therefore, exploring the mechanisms of osteosarcoma occurrence and metastasis at the molecular level is of significant scientific and social importance, laying the foundation for early diagnosis and prognostic assessment. Currently, molecular subtyping and prognostic prediction of osteosarcoma are mainly based on whole-tissue transcriptome sequencing data to analyze the molecular characteristics of osteosarcoma and attempt to construct prognostic gene models. However, these methods are typically based on the "average expression" signal of tumor tissue, ignoring the significant heterogeneity between different cell types in the tumor microenvironment. This makes it difficult to reveal the true biological role of specific cell populations in tumor occurrence, recurrence, and progression. Therefore, the predictive results have limitations in terms of stability and biological interpretability.

[0003] With the development of single-cell transcriptome sequencing technology, the cellular composition and functional status of the osteosarcoma tumor microenvironment at the single-cell level have been analyzed. This has revealed that osteosarcoma tissue contains various cell types in addition to tumor cells, including immune cells, vascular-associated cells, and stromal cells. Different cell populations may play different roles in tumor progression. However, current single-cell studies mostly focus on cell type identification or immune microenvironment description. There is still a lack of mature and reproducible technical solutions for systematically translating cell-specific molecular characteristics obtained at the single-cell level into patient-level molecular subtyping and prognostic assessment tools.

[0004] Osteoblasts, as the most numerous cell type in osteosarcoma tissue and closely related to tumor origin, have not yet had their molecular characteristics in tumor recurrence and malignant progression fully explored. Current technologies have not yet established a system that uses osteoblasts as the core to screen key gene features from single-cell transcriptome data and further constructs a system that can be used for patient prognosis assessment and risk stratification. In actual research, there are many difficulties, such as large sample heterogeneity, unstable feature gene screening, and difficulty in effectively integrating single-cell results with clinical data.

[0005] Therefore, how to overcome the shortcomings of existing whole transcriptome analysis that ignores cellular heterogeneity, make full use of single-cell transcriptome data, and clarify the molecular characteristics of key cell populations in osteosarcoma remains a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] Therefore, the technical problem to be solved by the present invention is to provide a biomarker for osteosarcoma and a method for constructing its application and prognostic prediction model. With osteoblasts as the core research object, it avoids the problem of different cell signals masking each other in the overall transcriptome analysis. By mapping osteoblast-specific molecular characteristics to the patient's overall transcriptome data, it realizes clinical translational application.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] An osteosarcoma biomarker, said osteosarcoma biomarker comprising the genes CGREF1, COL13A1, KIF25 and RPL35AP7.

[0009] This application also provides the use of reagents for detecting the expression levels of CGREF1, COL13A1, KIF25, and RPL35AP7 genes in the preparation of products for predicting the prognosis of osteosarcoma, wherein the reagents in the products include one or more of primers that specifically amplify the prognostic markers, probes that specifically recognize the prognostic markers, and binding agents that specifically bind to the proteins encoded by the prognostic markers.

[0010] Furthermore, the products include one or more of the following: reagent kits, chips, and systems.

[0011] This application also provides a method for screening biomarkers to predict the prognosis of osteosarcoma, comprising the following steps: S1, acquisition and preprocessing of osteosarcoma single-cell transcriptome data: acquiring single-cell transcriptome sequencing data of primary and recurrent osteosarcoma samples, performing quality control on the single-cell transcriptome sequencing data, removing low-quality cells and double cells, and standardizing and integrating multiple samples of the retained single-cell data to obtain a high-quality single-cell expression matrix; S2, tumor microenvironment cell clustering and cell type annotation: performing dimensionality reduction and unsupervised clustering based on the single-cell expression matrix obtained in S1, identifying each transcriptome cluster according to preset cell type marker genes, annotating each transcriptome cluster with cell type, and determining the main cell population in the tumor microenvironment; S3, screening and differential molecular feature identification of osteoblast subpopulations: screening osteoblast subpopulations from the cell types obtained in S2, comparing the gene expression differences of osteoblasts in recurrent samples and primary samples, and obtaining the differences in gene expression between recurrent and primary samples. S4. Functional analysis of osteoblast-related molecular features: Pathway enrichment and gene set enrichment analyses were performed on the osteoblast-related differentially expressed gene set in S3 to screen osteoblast molecular features closely related to tumor invasiveness and biological progression, providing a biological basis for subtyping and modeling; S5. Patient molecular subtyping based on osteoblast-related genes: The osteoblast-related genes obtained in S4 were mapped to the overall transcriptome data of osteosarcoma patients to perform unsupervised molecular subtyping, obtaining at least two molecular subtypes with significant biological and clinical outcome differences; S6. Screening of key genes: Based on the molecular subtyping obtained in S5, key genes CGREF1, COL13A1, KIF25, and RPL35AP7, which are significantly related to survival, were screened, and a prognostic risk assessment model based on the expression levels of key genes CGREF1, COL13A1, KIF25, and RPL35AP7 was constructed.

[0012] S7. Using the expression levels of genes screened in S6 and their corresponding multivariate analysis coefficients, a prognostic prediction model for osteosarcoma was constructed.

[0013] The process of constructing the prognostic prediction model for osteosarcoma is as follows:

[0014] Based on the four key genes CGREF1, COL13A1, KIF25, and RPL35AP7 previously screened through univariate Cox regression, LASSO regression, and stepwise regression, a multivariate Cox proportional hazards regression model was constructed with the overall survival time and survival status of osteosarcoma samples as dependent variables and the expression values ​​of the four genes as independent variables. Through model fitting, the regression coefficient of each gene was obtained, which reflects the relative contribution of each gene to the patient's survival risk.

[0015] Risk score calculation:

[0016] A linear risk score formula was established based on the regression coefficients output by the multivariate Cox regression model. The regression coefficients for each gene are: CGREF1 (0.03575), COL13A1 (0.07057), KIF25 (0.46774), and RPL35AP7 (0.66914). The risk score is calculated as follows: Risk Score = 0.03575 × CGREF1 expression value + 0.07057 × COL13A1 expression value + 0.46774 × KIF25 expression value + 0.66914 × RPL35AP7 expression value. Based on this formula, a risk score is calculated for each osteosarcoma sample. A higher score indicates a higher risk of death for the patient. The expression values ​​for each gene are the expression levels of FPKM derived from TCGA.

[0017] Risk level classification criteria:

[0018] To transform continuous risk scores into binary variables for clinical decision-making, the `surv_cutpoint` function was used to calculate the optimal cutoff value for the risk score. This function, based on the principle of maximum selection rank statistic, determines the critical point that maximizes the survival difference between high-risk and low-risk groups. The calculated optimal cutoff value was 2.026652. Patients were divided into high-risk and low-risk groups based on this value: patients with risk scores greater than 2.026652 were assigned to the high-risk group, and those with scores less than or equal to this value were assigned to the low-risk group. This grouping method was validated using Kaplan-Meier survival curves, showing a significant survival difference between the two groups, demonstrating that this classification method has good prognostic discrimination ability.

[0019] Furthermore, the primary and recurrent samples are selected from at least one of the following: serum, plasma, cerebrospinal fluid, tissue or tissue lysate, cell culture supernatant, semen, and saliva samples from osteosarcoma patients.

[0020] The technical solution of the present invention achieves the following beneficial technical effects:

[0021] 1. The osteosarcoma biomarkers and prognostic prediction model provided by this invention reveal the key molecular characteristics of osteosarcoma at the cellular level. With osteoblasts as the core research object, it avoids the problem of different cell signals masking each other in the overall transcriptome analysis, so that the obtained molecular characteristics have clear cell origin and biological significance.

[0022] 2. The osteosarcoma biomarkers and prognostic prediction model provided by this invention achieve an effective connection between single-cell data and clinical prognosis. By mapping osteoblast-specific molecular characteristics to the patient's overall transcriptome data and performing systematic typing and modeling, it solves the problem that existing single-cell research results are difficult to translate into clinical applications.

[0023] 3. The osteosarcoma biomarkers and prognostic prediction models provided by this invention improve the accuracy and stability of osteosarcoma prognostic assessment. The molecular subtyping and risk model constructed based on the method of this invention shows significant discriminative ability in overall survival and event-free survival prediction, and can achieve effective risk stratification of patients.

[0024] 4. The osteosarcoma biomarkers and prognostic prediction models provided by this invention have good reproducibility and application value. Based on standardized data processing procedures and clear gene feature screening steps, they are applicable to osteosarcoma sample data from different sources and have the potential for application in scientific research and clinical decision support. Attached Figure Description

[0025] Figure 1 UMAP clustering diagram of this invention;

[0026] Figure 2 UMAP cell type annotation diagram of this invention;

[0027] Figure 3 Stacked violin diagram of cell type marker gene expression in this invention;

[0028] Figure 4 Volcano diagram of differentially expressed genes in osteoblasts in this invention;

[0029] Figure 5 GSEA enrichment point map of the C7 gene set in this invention;

[0030] Figure 6 The rank evaluation diagram of the NMF classification in this invention;

[0031] Figure 7 This invention provides a heatmap of the NMF consistency matrix.

[0032] Figure 8 The overall survival Kaplan-Meier curve of NMF typing in this invention;

[0033] Figure 9 Kaplan-Meier curves for event-free survival in NMF typing according to this invention;

[0034] Figure 10 The overall survival Kaplan-Meier curve of the transition state grouping in this invention;

[0035] Figure 11 The event-free survival Kaplan-Meier curve for transition state grouping in this invention;

[0036] Figure 12 Stacked bar chart showing the ratio of NMF typing to overall survival outcome in this invention;

[0037] Figure 13A stacked bar chart showing the ratio of NMF typing to event-free survival outcomes in this invention;

[0038] Figure 14 A stacked bar chart showing the ratio of NMF classification to transition states in this invention;

[0039] Figure 15 A stacked bar chart showing the ratio of transition states to overall survival outcomes in this invention;

[0040] Figure 16 A stacked bar chart showing the ratio of transition states to event-free survival outcomes in this invention;

[0041] Figure 17 This invention relates to a volcano plot of differentially expressed genes related to genotyping.

[0042] Figure 18 This invention provides a LASSO-Cox regression coefficient path diagram;

[0043] Figure 19 LASSO-Cox cross-validation curve of this invention;

[0044] Figure 20 Time-dependent ROC curve of the risk score of this invention;

[0045] Figure 21 The overall survival Kaplan-Meier curves for the high / low risk groups of this invention;

[0046] Figure 22 Risk diagram of the prognostic model of this invention;

[0047] Figure 23 Kaplan–Meier survival curve analysis of CGREF1, the candidate gene of this invention;

[0048] Figure 24 Kaplan–Meier survival curve analysis of the candidate gene COL13A1 in this invention;

[0049] Figure 25 Kaplan–Meier survival curve analysis of the candidate gene KIF25 of this invention;

[0050] Figure 26 Kaplan–Meier survival curve analysis of the candidate gene RPL35AP7 of this invention;

[0051] Figure 27 A heatmap showing the correlation between the risk score of this invention and the KEGG pathway ssGSEA score. Detailed Implementation

[0052] Example 1: Screening of prognostic markers for osteosarcoma

[0053] 1. Acquisition and preprocessing of single-cell transcriptome data of osteosarcoma: Single-cell transcriptome sequencing data of primary and recurrent osteosarcoma samples were acquired. The data were subjected to quality control, low-quality cells and double cells were removed, and the retained single-cell data were standardized and multi-sample integrated to obtain a high-quality single-cell expression matrix.

[0054] Figure 1 The UMAP cluster map (0–15 clusters) shows the distribution of integrated osteosarcoma single-cell transcriptome data in UMAP space, with cell clusters of 0–15 displayed by color according to the Seurat unsupervised clustering results. The processing results are as follows... Figure 1 As shown, after uniform quality control, normalization and cross-sample integration of single-cell transcriptome data from primary and recurrent osteosarcoma samples, a total of 61,879 high-quality single cells were obtained for subsequent analysis. Based on principal component analysis (PCA) and nonlinear dimensionality reduction (UMAP), and using unsupervised graph clustering, 16 transcriptome clusters (0–15 clusters) were identified in the nonlinear dimensionality reduction space.

[0055] 2. Tumor microenvironment cell clustering and cell type annotation: Dimensionality reduction and unsupervised clustering are performed based on single-cell expression matrices, and cell type annotation is performed on each cell cluster according to preset cell type marker genes to identify the main cell populations in the tumor microenvironment, including osteoblasts, immune cells, vascular-associated cells and stromal cells, etc.

[0056] Figure 2 Annotation map of UMAP cell types, in Figure 1 Based on this, cell type annotation of cell clusters is performed by combining marker gene expression, and the distribution and aggregation characteristics of various cell types are displayed in UMAP space; Figure 3 A stacked violin plot of cell type marker gene expression shows the expression distribution of typical marker genes in different cell types. The width of the violin represents the expression density, used to verify the rationality of cell type annotation. The annotation results are as follows: Figure 2 and Figure 3 As shown, after systematically annotating each cell cluster using classic cell type marker genes, it was found that the osteosarcoma microenvironment is mainly composed of seven cell populations, including myeloid cells (16,155), lymphocytes (3,499), endothelial cells (2,685), parietal cells (pericytes / smooth muscle cells) (4,477), fibroblasts (2,627), osteoclasts (2,425), and osteoblasts (30,011). Among them, osteoblasts are the most prevalent cell type, accounting for approximately 48.5% of all cells, suggesting that they may play a core role in the occurrence and development of osteosarcoma.

[0057] 3. Screening and differential molecular feature identification of osteoblast subpopulations: Screen osteoblast subpopulations from cell types, compare the gene expression differences of osteoblasts in relapsed samples and primary samples, and obtain the set of differentially expressed genes of osteoblasts that are significantly related to relapsed samples.

[0058] Figure 4 This is a volcano plot of differentially expressed genes in osteoblasts, comparing differentially expressed genes in osteoblasts between relapsed and primary samples. The horizontal axis represents log2FC, and the vertical axis represents −log10 (corrected P-value). Significantly differentially expressed genes are highlighted in the plot. Given the dominant role of osteoblasts in osteosarcoma tissue, we further focused on osteoblast subpopulations, comparing transcriptional differences between relapsed and primary samples. The results of differential expression analysis are as follows: Figure 4 As shown, a total of 1,011 differentially expressed genes were identified in osteoblasts (|log2FC|>0.585, corrected P<0.05). Among them, 360 genes were upregulated and 651 genes were downregulated in relapsed samples. There were significant transcriptomic differences between osteosarcoma osteoblasts in relapsed and primary samples. These differentially expressed genes as a whole reflect that osteoblasts underwent significant transcriptional reprogramming during the relapse stage, suggesting that their functional status may have shifted from a relatively differentiation / stromal generation-related phenotype to a state with more tumor-related characteristics.

[0059] 4. Functional analysis of osteoblast-related molecular features: Pathway enrichment analysis and gene set enrichment analysis were performed on differentially expressed genes to screen osteoblast molecular features closely related to tumor invasiveness and biological progress, providing a biological basis for subsequent subtyping and modeling;

[0060] Figure 5 This is a GSEA enrichment map of the C7 gene set. GSEA analysis (MSigDB C7 immunophenotypic gene set) was performed based on differential expression results. The position of the dots reflects the enrichment intensity (e.g., NES), the dot size represents the gene set size / number of enriched genes, and the color indicates significance (FDR / q value). To systematically elucidate the biological significance of osteoblast differential expression characteristics at the pathway level, genes were sorted based on the differential analysis results, and gene set enrichment analysis (GSEA) was performed. The enrichment analysis results are shown below. Figure 5 As shown, in the immune signature gene set (MSigDB C7), multiple gene sets associated with tumor invasiveness and progression were significantly enriched in the recurrence-associated osteoblast expression profile. Among them, the Anastasiauddo cancer invasive signature gene set (enrichment score (NES) = −2.53, corrected p = 3.29 × 10⁻) was particularly rich. 4 ), Differentially expressed gene set of Liu's ovarian cancer tumor and xenograft (downregulated) (enrichment score (NES) = -3.12, corrected p = 1.60 × 10⁻¹) 4Gene sets such as the Schutz breast cancer ductal invasion upregulated gene set (enrichment score (NES) = -3.01, corrected p = 4.26 × 10⁻¹¹) showed highly significant negative enrichment, suggesting that the transcriptional characteristics of recurrence-related osteoblast genes are closely related to increased tumor invasiveness.

[0061] 5. Patient molecular subtyping based on osteoblast-related genes: Map osteoblast-related genes to the overall transcriptome data of osteosarcoma patients to perform unsupervised molecular subtyping of patients and obtain at least two molecular subtypes with significant biological and clinical outcome differences;

[0062] To assess the clinical significance of osteoblast-related genes, candidate genes screened in single-cell analysis were mapped to transcriptomic data of an osteosarcoma patient cohort, and analyzed in conjunction with overall survival (OS) and event-free survival (EFS) information. After initial screening by univariate Cox regression, patients were subjected to unsupervised molecular subtyping using nonnegative matrix factorization.

[0063] Figure 6 This is a rank evaluation plot for NMF typing, where nonnegative matrix factorization (NMF) is used to perform molecular typing of the samples; it shows the changes in model indices (such as consistency / stability) under different ranks to determine the optimal number of typings. Figure 7 A heatmap of the NMF consistency matrix is ​​generated, plotting the consistency matrix at the optimal rank to show the clustering consistency and clarity of the fractal boundaries among samples. By systematically evaluating the consistency and stability of the model under different decomposition ranks, the optimal number of fractals is ultimately determined as follows: Figure 6 The consistency matrices constructed under the optimal decomposition rank for the two classes shown are as follows: Figure 7 As shown, the genotyping boundary between the two types of samples is clear, and the consistency within the groups is good.

[0064] Figure 8 The Kaplan-Meier curves for overall survival of NMF subtypes are shown to compare the differences in overall survival between different NMF subtypes (Cluster 1 / 2). The curves are estimated by KM, and the Cox proportional hazards model hazard ratios and 95% confidence intervals are marked on the graph. Figure 9 The event-free survival Kaplan-Meier curves for NMF classifications are shown to compare the differences in event-free survival among different NMF classifications. The curves are estimated by KM and the hazard ratios and 95% confidence intervals are marked. Figure 10 Kaplan-Meier curves of overall survival grouped by migration status, comparing differences in overall survival based on whether or not migration occurred (migration vs. non-migration). Figure 10 The risk ratio, 95% confidence interval, and p-value are given in the text; Figure 11Kaplan-Meier (KM) survival curves for event-free survival grouped by transfer status are shown, comparing differences in event-free survival between groups with and without transfer. The curves are estimated using the KM method, and the hazard ratio, 95% confidence interval, and p-value are labeled. The Kaplan-Meier survival analysis results show... Figure 8 and Figure 9 As shown, there were significant differences in overall survival (OS) and event-free survival (EFS) among patients with different non-negative matrix factorization subtypes; furthermore, after grouping patients according to whether metastasis occurred, there were also significant differences in overall survival and event-free survival between metastatic and non-metastatic patients. Figure 10 and Figure 11 The significantly different survival outcomes are shown.

[0065] Figure 12 Stacked bar charts showing the proportion of survival and death among different NMF subtypes were used to evaluate differences between groups and p-values ​​were labeled. Figure 13 Stacked bar charts showing the proportion of no-event and event-occurring outcomes in different NMF subtypes; Fisher's exact test provides the p-value for differences between groups. Figure 14 A stacked bar chart showing the proportion of metastatic and non-metastatic states in different NMF types is used to evaluate differences using Fisher's precise test and label the p-values. Figure 15 Stacked bar charts showing the proportion of metastasis status to overall survival outcome (survival / death) in the metastasis / non-metastasis groups; Fisher's exact test was used to assess the differences and p-values ​​were labeled. Figure 16 A stacked bar chart showing the proportion of event-free survival outcomes in the metastasis and non-metastasis groups was used; Fisher's exact test was used to assess the differences and p-values ​​were labeled. Further proportion analysis results are as follows: Figure 12 – Figure 16 As shown, different nonnegative matrix factorization types exhibit significant differences in the proportions of death outcomes, event occurrences, and transition states.

[0066] Figure 17 This is a volcano plot of differentially expressed genes related to genotypes, comparing differentially expressed genes among genotype groups; the horizontal axis is logFC, the vertical axis is −log10 (corrected P-value), and significant genes are distinguished by |logFC| and the corrected P-value threshold.

[0067] 6. Key gene screening and prognostic risk model construction: Based on molecular subtyping, key genes that are significantly related to patient survival are screened, a prognostic risk assessment model based on the expression level of key genes is constructed, and patients are divided into different risk levels according to the risk score to predict the overall survival and disease progression risk of patients.

[0068] The prognostic prediction model was constructed based on multivariate Cox proportional hazards regression analysis. First, the expression values ​​of four selected genes—CGREF1, COL13A1, KIF25, and RPL35AP7—were used as independent variables, with overall survival time and survival status as dependent variables. A multivariate Cox regression model was established, yielding regression coefficients for each gene: CGREF1 (0.03575), COL13A1 (0.07057), KIF25 (0.46774), and RPL35AP7 (0.66914). Based on these coefficients, a risk score formula was constructed: Risk Score = 0.03575 × CGREF1 expression value + 0.07057 × COL13A1 expression value + 0.46774 × KIF25. The risk score for each sample was calculated using the formula: IF25 expression value + 0.66914 × RPL35AP7 expression value. The optimal cutoff value of 2.026652 for the risk score was determined using the surv_cutpoint function. Patients were divided into high-risk and low-risk groups. To evaluate the predictive efficacy of the model, time-dependent receiver operating characteristic (ROC) curves were plotted. The results showed that the areas under the curves for 1-year, 3-year, and 5-year survival predictions were 0.77, 0.71, and 0.66, respectively. Furthermore, the survival differences between the two groups were compared using Kaplan-Meier survival curves, and the hazard ratio and significance p-value were calculated, further validating the good discriminative ability of this risk scoring model for osteosarcoma patients.

[0069] Figure 18 The LASSO-Cox regression coefficient path diagram (λ change) shows the shrinkage trajectory of the regression coefficients of each candidate gene during the change of the LASSO penalty parameter λ, which is used to observe the process of features entering / exiting the model; Figure 19 The LASSO-Cox cross-validation curve (selecting the optimal λ) shows the change of model bias / error with log(λ) under cross-validation, and determines the optimal λ (e.g., λ_min / λ_1se) to construct a robust prognostic model; LASSO-Cox regression is used to perform feature compression and screening of candidate genes, and the results are as follows. Figure 18 As shown, with the increase of the penalty parameter λ, the regression coefficients of most genes gradually shrink to zero. This was confirmed through cross-validation. Figure 19 The optimal penalty parameter λ_min = 0.1716214 was determined, and four key genes, CGREF1, COL13A1, KIF25 and RPL35AP7, were finally selected to construct the prognostic risk model.

[0070] Figure 20 This is a time-dependent ROC curve for risk scores, used to assess the overall survival prediction performance at 1, 3, and 5 years based on risk scores. The curves represent the time-dependent ROC, and the corresponding AUC values ​​are labeled on the graph. Figure 21 The Kaplan-Meier curves for overall survival of the high-risk and low-risk groups are shown. The groups are divided into high-risk and low-risk groups according to their risk scores. The differences in overall survival between the two groups are compared. The hazard ratio, 95% confidence interval and p-value are shown in the figure. Figure 22 The prognostic model risk map (ggrisk) displays a comprehensive visualization of the Cox prognostic model (risk score distribution, outcome status distribution, and key gene expression patterns, etc.), used to assess model stratification and sample prognostic differences. Time-dependent ROC curve analysis is shown below. Figure 20 As shown, the model has good discriminative ability in predicting overall survival, with area under the curve (AUC) of 0.77, 0.71, and 0.66 for 1-year, 3-year, and 5-year OS, respectively. After dividing patients into high-risk and low-risk groups based on risk scores, the Kaplan-Meier survival analysis results are as follows: Figure 21 As shown, the overall survival of patients in the high-risk group was significantly shortened; the comprehensive visualization results of risk score distribution, outcome status, and key gene expression patterns are as follows: Figure 22 As shown, the constructed prognostic prediction model demonstrates its ability to predict overall survival in osteosarcoma and further validates the model's stability in prognostic stratification.

[0071] Example 2: Validation of the osteosarcoma prognostic prediction model

[0072] To validate the independent survival predictive value of each gene in the prognostic model, Kaplan-Meier survival analysis was performed on CGREF1, COL13A1, KIF25, and RPL35AP7.

[0073] The results are as follows Figures 23-26 As shown, Figure 23 The Kaplan-Meier survival curve analysis of candidate gene CGREF1 is shown. CGREF1 is divided into high expression and low expression groups according to the median expression level of CGREF1, and the overall survival difference is compared. The Cox model hazard ratio and 95% confidence interval are marked in the figure. Figure 24 Kaplan–Meier survival curve analysis for candidate gene COL13A1, comparing overall survival differences by grouping according to median COL13A1 expression, and marking hazard ratios and 95% confidence intervals; Figure 25 The Kaplan-Meier survival curve analysis of candidate gene KIF25 is shown in the figure. The overall survival difference is compared between high and low expression groups based on the median KIF25 expression. The hazard ratio and 95% confidence interval are marked in the figure. Figure 26 Kaplan-Meier survival curve analysis of candidate gene RPL35AP7 was performed. Overall survival differences were compared by grouping according to the median expression of RPL35AP7, and hazard ratios and 95% confidence intervals were marked. The expression levels of these four genes were significantly associated with overall patient survival (P<0.05).

[0074] Will Figure 23-26 The single-gene model shown is similar to Figure 20-21 A systematic comparison of the four-gene combined model shown indicates that:

[0075] In the time-dependent ROC analysis of single genes, different genes showed significant differences in predictive performance at different time scales. RPL35AP7 performed best in 5-year survival prediction (AUC=0.79), but its performance in short-term prediction was average (AUC=0.72). KIF25 showed relatively stable performance in 1-year and 3-year prediction (AUC=0.73 and 0.72, respectively), but its overall predictive ability was at a moderate level. COL13A1 had a low AUC in 1-year prediction (0.63), but improved in 3-year and 5-year prediction (0.75 and 0.76, respectively), indicating that its predictive ability is time-dependent. CGREF1 performed best in 1-year prediction (AUC=0.79), but its performance in 5-year prediction decreased significantly. These results indicate that the predictive performance of single-gene models fluctuates greatly at different follow-up time points, lacks stability, and is difficult to maintain consistent predictive performance across all time scales.

[0076] In contrast, the four-gene combined model based on RPL35AP7, KIF25, COL13A1, and CGREF1 had AUCs of 0.77, 0.71, and 0.66 for 1-year, 3-year, and 5-year survival predictions, respectively. Although it did not surpass the best single-gene model at individual time points, it maintained relatively stable predictive performance across different time scales, especially outperforming most single-gene models (such as COL13A1) in 1-year predictions. This result suggests that the multi-gene combined model effectively compensates for the limitations of single-gene models by integrating the predictive information of different genes at different time dimensions.

[0077] In summary, each single gene has different temporal advantages in prognostic prediction, but no single gene can maintain the best predictive ability at all time points. The four-gene joint model achieves balanced prediction of short-term and medium-to-long-term prognosis through multi-dimensional information fusion, thereby significantly improving the overall stability and generalization ability of the model. Therefore, the four-gene model provided by this invention has higher application potential in osteosarcoma prognostic assessment and can be used as a robust multi-indicator joint prediction tool.

[0078] The above results indicate that the four-gene combined model can significantly improve predictive performance, especially in the short term, within a specific time window, demonstrating the technological advancement of the multi-gene combined model compared to the single-gene model.

[0079] Figure 27 To create a heatmap of the correlation between risk scores and KEGG pathway ssGSEA scores, Pearson correlation analysis was performed on the KEGG gene set ssGSEA scores and risk scores. Pathways with p-values ​​< 0.01 and absolute correlation coefficients > 0.35 were selected, and a correlation matrix heatmap was plotted. Hierarchical clustering was then used to illustrate the correlation structure between pathways. Correlation analysis was also performed between risk scores and single-sample gene set enrichment analysis (ssGSEA) scores from the Kyoto Encyclopedia of Genes and Genomes database. Figure 27 The study showed that the risk score was significantly correlated with multiple tumor-related pathways, which mainly involve key biological processes such as extracellular matrix interaction, signal transduction, and tumor metabolic reprogramming.

[0080] Example 3

[0081] A product for osteosarcoma prognosis, the product containing a reagent for detecting the expression level of at least one of CGREF1, COL13A1, KIF25 and RPL35AP7, specifically a gene chip for detecting gene expression levels. The higher the expression level of the gene being measured, the shorter the survival time, the lower the survival rate within a certain time period, the higher the risk of metastasis, and the lower the immune score of the tumor microenvironment.

[0082] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of the claims of this patent application.

Claims

1. A biomarker for osteosarcoma, characterized in that, The osteosarcoma biomarkers include the genes CGREF1, COL13A1, KIF25, and RPL35AP7.

2. Application of reagents for detecting the expression levels of CGREF1, COL13A1, KIF25 and RPL35AP7 genes in the preparation of products for predicting the prognosis of osteosarcoma.

3. The application according to claim 2, characterized in that, The products include one or more of the following: reagent kits, chips, and systems.

4. A method for constructing a prognostic prediction model for osteosarcoma, characterized in that, Includes the following steps: S1. Acquisition and preprocessing of single-cell transcriptome data from osteosarcoma: Single-cell transcriptome sequencing data of primary and recurrent osteosarcoma samples were obtained. The single-cell transcriptome sequencing data were subjected to quality control, low-quality cells and double cells were removed, and the retained single-cell data were standardized and integrated from multiple samples to obtain a high-quality single-cell expression matrix. S2. Tumor microenvironment cell clustering and cell type annotation: Dimensionality reduction and unsupervised clustering were performed based on the single-cell expression matrix obtained by S1, and each transcriptomics cluster was identified according to the preset cell type marker genes. Cell type annotation was performed on each transcriptomics cluster to determine the main cell population in the tumor microenvironment. S3. Screening and differential molecular signature identification of osteoblast subsets: Osteoblast subpopulations were screened from the cell types obtained from S2, and the gene expression differences of osteoblasts in relapsed samples and primary samples were compared to obtain a set of differentially expressed genes in osteoblasts that were significantly associated with relapsed samples. S4. Functional analysis of osteoblast-related molecular characteristics: Pathway enrichment analysis and gene set enrichment analysis were performed on the differentially expressed gene set of osteoblasts in S3 to screen osteoblast molecular features closely related to tumor invasiveness and biological progress, providing a biological basis for subtyping and modeling. S5. Molecular subtyping of primary and recurrent samples based on osteoblast-related genes: The osteoblast-related genes obtained in S4 were mapped to the overall transcriptome data of primary and recurrent samples, and unsupervised molecular typing was performed on the primary and recurrent samples to obtain at least two molecular subtypes with significant biological and clinical outcome differences. S6. Key gene screening; Based on the molecular typing obtained in S5, key genes COL13A1, KIF25, CGREF1 and RPL35AP7 that are significantly related to survival were screened, and a prognostic risk assessment model based on the expression levels of key genes COL13A1, KIF25, CGREF1 and RPL35AP7 was constructed. S7. Using the expression levels of genes screened in S6 and their corresponding multivariate analysis coefficients, a prognostic prediction model for osteosarcoma was constructed.

5. The method according to claim 4, characterized in that, The osteosarcoma prognostic prediction model includes the following calculation formula: Risk score = 0.03575 × CGREF1 expression value + 0.07057 × COL13A1 expression value + 0.46774 × KIF25 expression value + 0.66914 × RPL35AP7 expression value; A risk score > 2.026652 indicates a high risk of osteosarcoma, while a risk score ≤ 2.026652 indicates a low risk of osteosarcoma.

6. The method according to claim 5, characterized in that, The CGREF1, COL13A1, KIF25, and RPL35AP7 expression values ​​were all FPKM expression levels derived from TCGA.

Images