Prediction model for judging prognosis and immunotherapy response of ovarian cancer based on machine learning and application

Abstract

The invention relates to a prediction model for judging prognosis and immunotherapy response of ovarian cancer based on machine learning. The prediction model comprises the following steps: determining gene expression quantities of eight markers, namely ANGPTL4, AREG, CXCL10, FANCB, GMPR, IGF1, LYG1 and WNK 1. The invention also provides a gene combination kit for predicting the prognosis and immunotherapy response of the ovarian cancer and application of the gene combination kit. The invention provides a prediction model aiming at ovarian cancer, which is used for judging the prognosis of a patient, can assist in guiding immunotherapy and prognosis judgment, and helps to realize personalized treatment aiming at a specific patient. Experiments and tumor big data support the prognosis model, the prognosis and immunotherapy response of a patient can be effectively predicted by integrating various biological and clinical data, powerful support can be provided for individualized treatment, a treatment scheme can be optimized, and the survival rate and life quality of the patient can be improved.

Description

Technical Field

[0001] The present invention relates to the field of medicine and biology, and in particular to a prediction model based on machine learning for judging the prognosis of ovarian cancer and immunotherapy response and its application. Background Art

[0002] Ovarian cancer is highly heterogeneous, including differences in the molecular characteristics of tumors, clinical manifestations, and responses to treatment. Different patients have very different responses to treatment and their prognosis and survival time are completely different. Personalized treatment for specific patients can effectively improve the survival time and quality of patients. However, there is currently no way to reliably distinguish the disease status of different patients and assist in the realization of precision treatment. With the development of high-throughput cancer genomics, clinical practices of judging the prognosis of patients and the efficacy of immunotherapy targeted drugs by sending tumor tissues for testing and sequencing are gradually increasing. Machine learning is a data analysis method. Machine learning is used in tumor research for early diagnosis, prognosis prediction, formulation of personalized treatment plans, and drug discovery. By analyzing gene expression, imaging data, and clinical information, machine learning models can identify tumor characteristics, predict patients' treatment response and survival rate, optimize treatment effects, and promote the development of precision medicine. Some prognostic models have been developed to predict the prognosis of ovarian cancer, such as models related to apoptosis, pyroptosis, and copper death [1-3]. At present, the accuracy of ovarian cancer prognostic models is not high, and most of them only use genomic sequencing data, and no models that can combine clinical characteristics and omics data have been found.

[0003] Chinese patent document CN113096739B discloses an analysis method for an immune prognostic diagnostic marker combination for ovarian cancer, wherein the marker combination includes the following genes: CSPG5, CXCL10, CXCL11, DKK1, PI3, TNFRSF17, DUOX1, TNFRSF13B and PAEP; the analysis method includes: S1, data collection and preprocessing; S2, calculation of TMB scores and prognostic analysis; S3, differential analysis; S4, construction and evaluation of TMB-immune prognostic related risk score model; S5, establishment of the relationship between clinical pathological factors and TMB-IRS model; S6, CIBERSORT database analysis; S7, statistical analysis. The present invention constructs an immune risk scoring model related to TMB through 9 key tumor mutation load-related immune prognostic genes in ovarian cancer, divides patients into high-risk and low-risk groups, and proves that the high-risk group is associated with a poor prognosis, and that the immune cell infiltration ratio of patients in the low-risk group is higher; it provides prognostic hints for the clinical treatment of ovarian cancer and can serve as a new biomarker and potential therapeutic target for predicting the prognosis of ovarian cancer.

[0004] Chinese patent document CN108977545B discloses an ovarian cancer diagnosis and prognosis composition and its application. An ovarian cancer diagnosis or prognosis composition, composed of PDK1, PRKAA2 and PRKAB1. The use of the ovarian cancer prognosis composition of the present invention in the preparation of an ovarian cancer diagnosis kit or a kit for predicting the prognosis of ovarian cancer. A kit for predicting the prognosis of ovarian cancer, comprising reagents for detecting the expression levels of PDK1, PRKAA2 and PRKAB1. The combined detection of PDK1, PRKAA2 and PRKAB1 can effectively determine the prognosis of ovarian cancer patients, while the single detection of any one of the indicators cannot effectively determine the prognosis of ovarian cancer patients. The present invention provides an effective prognostic indicator for clinical use.

[0005] However, there are currently no reports on predictive models and applications based on machine learning for determining ovarian cancer prognosis and immunotherapy response. Summary of the invention

[0006] The purpose of the present invention is to address the deficiencies in the prior art and provide a prediction model and application based on machine learning for judging the prognosis of ovarian cancer and immunotherapy response.

[0007] In a first aspect, the present invention provides a prediction model based on machine learning for judging the prognosis and immunotherapy response of ovarian cancer, including determining: the expression levels of the eight-linked marker genes ANGPTL4, AREG, CXCL10, FANCB, GMPR, IGF1, LYG1 and WNK 1.

[0008] As a preferred example, the invention comprises an operating instruction manual, which records the following content: Riskscore=LYG1*-0.23654+AREG*0.087816+CXCL10*-0.06992+GMPR*-0.15972+WNK1*0.120974+IGF1*0.052135+ANGPTL4*0.099811+FANCB*-0.20232. When the score Riskscore<-0.028, it is considered that the ovarian cancer patient is sensitive to chemotherapy and immunotherapy; when the score Riskscore≧-0.028, it is considered that the ovarian cancer patient is insensitive to chemotherapy and immunotherapy.

[0009] In a second aspect, the present invention provides an application of a gene combination kit in the preparation of a reagent for predicting the prognosis of ovarian cancer and immunotherapy response, wherein the gene combination is the expression levels of eight marker genes: ANGPTL4, AREG, CXCL10, FANCB, GMPR, IGF1, LYG1 and WNK 1.

[0010] As a preferred example, the invention further comprises an operating instruction manual, which records the following content: Riskscore=LYG1*-0.23654+AREG*0.087816+CXCL10*-0.06992+GMPR*-0.15972+WNK1*0.120974+IGF1*0.052135+ANGPTL4*0.099811+FANCB*-0.20232. When the score Riskscore<-0.028, it is considered that the ovarian cancer patient is sensitive to chemotherapy and immunotherapy; when the score Riskscore≧-0.028, it is considered that the ovarian cancer patient is insensitive to chemotherapy and immunotherapy.

[0011] In a third aspect, the present invention provides a gene combination kit for predicting the prognosis and immunotherapy response of ovarian cancer, wherein the gene combination is the expression of eight marker genes including ANGPTL4, AREG, CXCL10, FANCB, GMPR, IGF1, LYG1 and WNK1.

[0012] As a preferred example, the invention further comprises an operating manual, wherein the operating manual contains the following contents: Riskscore=LYG1*-0.23654+AREG*0.087816+CXCL10*-0.06992+GMPR*-0.15972+WNK1*0.120974+IGF1*0.052135+ANGPTL4*0.099811+FANCB*-0.20232

[0013] If the score Riskscore < -0.028, the ovarian cancer patient is considered to be sensitive to chemotherapy and immunotherapy, and when the score Riskscore ≧ -0.028, the ovarian cancer patient is considered to be insensitive to chemotherapy and immunotherapy.

[0014] In this article, the relevant gene families refer to Genebank,

[0015] The "ANGPTL4" gene can be referenced to Gene ID: 51129

[0016] The "AREG" gene can be found in Gene ID: 374

[0017] The "CXCL10" gene can be found in Gene ID: 3627

[0018] The "FANCB" gene can be referenced to Gene ID: 2187

[0019] The "GMPR" gene can be referenced to Gene ID: 2766

[0020] The "IGF1" gene can be found in Gene ID: 3479

[0021] The "LYG1" gene can be found in Gene ID: 129530

[0022] The “WNK1” gene may refer to Gene ID: 65125.

[0023] The advantages of the present invention are: providing a prediction model for ovarian cancer, which is used to judge the prognosis of patients, can assist in guiding immunotherapy and prognosis judgment, and help achieve personalized treatment for specific patients. Experimental and tumor big data support the prognosis model of the present invention. This model can effectively predict the prognosis of patients and the response to immunotherapy by integrating a variety of biological and clinical data, and can provide strong support for individualized treatment, optimize treatment plans, and improve patients' survival rate and quality of life. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Flowchart of the present invention.

[0025] Figure 2 The construction process of the prognostic model. (A) shows the screening of genes related to the prognosis of cellular senescence and ovarian aging. (B) shows the best candidate genes selected in the LASSO model. (C) shows the LASSO coefficients of prognostic-related genes, where each curve represents a gene. (D) shows the time-dependent ROC curve of the 5-year overall survival in the TCGA-OV cohort to evaluate the predictive performance of the model.

[0026] Figure 3Clinical functional analysis of the prognostic model. (A) The distribution of risk scores in TCGA and the distribution of patients in high-risk and low-risk groups according to overall survival (OS) status are shown, and the heat map shows the expression profiles of the eight CSOARGs. (B) A forest plot showing the results of univariate Cox regression analysis between the expression of candidate CSOARGs and overall survival, and the results of univariate Cox regression analysis of the eight model genes are shown. (C) A forest plot showing the results of univariate Cox regression analysis between clinical characteristics (age, tumor stage, grade), risk score, and overall survival (D) The Kaplan-Meier survival curves of patients in high-risk and low-risk groups are shown. (E) The Kaplan-Meier survival curves of patients in sequential and younger groups are shown. (F) The nomogram of the TCGA-OV training cohort integrating CSOARG risk score, age, stage, and grade is shown. (G) The calibration plot of the 5-year OS probability in the TCGA-OV cohort is shown. (H) A bar chart showing the gene ontology (GO) enrichment analysis of differentially expressed genes in high-risk and low-risk groups is shown. (I) Bar graph showing the KEGG enrichment analysis of differentially expressed genes in the high-risk and low-risk groups.

[0027] Figure 4 Analysis of immunotherapy response in high-risk and low-risk groups. (AD) Comparison of immunophenotypic scores (IPS) between low-risk and high-risk groups. (A) Both PD-1 and CTLA-4 were negatively expressed. (B) Only PD-1 was positively expressed. (C) Only CTLA-4 was positively expressed. (D) Both PD-1 and CTLA-4 were positively expressed. (EH) Comparison of tumor microenvironment scores calculated by TIDE. (E) Total TIDE score. (F) T cell dysfunction score. (G) T cell exclusion score. (H) Microsatellite instability score.

[0028] Figure 5 Association of risk models with chemotherapeutic drugs, endocrine therapy, and targeted therapy. (AC) Models predicted sensitivity to chemotherapeutic drugs. Results estimated that low-risk patients had lower IC50 values ​​for 5-fluorouracil, capecitabine, and cisplatin. (DH) Models predicted sensitivity to targeted therapy. Results estimated that low-risk patients had lower IC50 values ​​for gefitinib, nilotinib, bosutinib, selumetinib, and dabrafenib. (I) Models predicted sensitivity to endocrine therapy. Results estimated that low-risk patients had lower IC50 values ​​for temozolomide.

[0029] Figure 6Application of machine learning in optimizing prognostic models. (A) Heatmap compares the performance of various machine learning models on the TCGA-OV dataset. The heatmap shows the accuracy, AUC, recall, precision, F1 score, Kappa value, and Mathews correlation coefficient (MCC) of 14 classifiers. (B) Feature importance graph showing the best model (Extra Trees classifier) ​​in predicting 5-year overall survival of TCGA-OV. (C) ROC curve showing the best model (Extra Trees classifier) ​​in predicting 5-year overall survival of TCGA-OV. (D) AUC values ​​of the prognostic models in this article are compared with those in other literature.

[0030] Figure 7 Expression level and function of WNK1 in ovarian cancer. (A) The expression of model genes in patients with premenopause (<45 years old) and menopause and postmenopause (>=45 years old) in the TCGA database was compared. (B) The differential expression of prognostic model genes in clinical patient tissues was verified by qRT-PCR. (C) The differential expression of WNK1 RNA level in ovarian cancer cells was verified by qRT-PCR. (D) The differential expression of WNK1 protein level in cells was verified by Western blot analysis. (E) The knockdown efficiency of WNK1 in ovarian cancer was verified. (F) The effect of WNK1 knockdown on the proliferation function of A2780 ovarian cancer cells was verified by CCK-8 assay. (G) The effect of WNK1 knockdown on the proliferation function of OVCAR-3 ovarian cancer cells was verified by CCK-8 assay. (H,I) The effect of WNK1 knockdown on the migration function of A2780 ovarian cancer cells was verified by scratch assay (scale bar: 50μm). (J, K) The effect of WNK1 knockdown on the migration function of OVCAR-3 ovarian cancer cells was verified by scratch assay (scale bar: 200 μm). (L, M) The effect of WNK1 knockdown on the migration function of ovarian cancer cells was verified by Transwell migration assay (scale bar: 200 μm). DETAILED DESCRIPTION

[0031] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content of the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0032] It should be noted that the XML file submitted for the sequence listing involved in the examples needs to replace all "U" in SEQ ID NO: 1 and SEQ ID NO: 2 with "T" due to format issues. The specific details are subject to those shown in the examples.

[0033] Example 1

[0034] The present invention is a prediction method based on machine learning for judging the prognosis of ovarian cancer and the response to immunotherapy. Figure 1 :From the perspective of aging, we collected gene sets related to cell aging and ovarian aging, selected genes related to the prognosis of ovarian cancer through the TCGA database, and obtained prognostic gene sets related to cell aging and ovarian aging. Through unicox regression and lasso regression analysis, a prognostic model based on the eight markers of ANGPTL4, AREG, CXCL10, FANCB, GMPR, IGF1, LYG1, and WNK 1 was obtained. Then, based on these eight genes and the data of age and International Federation of Gynecology and Obstetrics stage, the best prognostic model was screened out from 14 algorithms by using machine learning technology, and then the final prediction model was further fine-tuned. At the same time, on the one hand, further bioinformatics analysis was carried out, and patients were divided into high-risk group and low-risk group according to the risk score, and the differences in immune function, immunotherapy and drug sensitivity, genomic stability, and single-cell sequencing between the two groups were compared. On the other hand, molecular biological experiments were carried out to screen out the most significantly overexpressed gene WNK1, and the oncogene function of WNK1 was verified by RT-qPCR, protein blotting, proliferation assay, scratch assay, and migration assay.

[0035] (1) Experimental methods

[0036] 1. Data Source

[0037] The Cancer Genome Atlas Ovarian Cancer (TCGA-OV) dataset was downloaded from the UCSC Xena browser (https: / / xenabrowser.net / ). After removing samples without clinical information such as age, stage, or overall survival less than 30 days, a total of 407 tumor samples were included for further analysis. The single-cell RNA sequencing (scRNA-seq) dataset GSE165897[4] was collected from the Gene Expression Omnibus (GEO) (https: / / www.ncbi.nlm.nih.gov / geo / ). For external validation, the bulk RNA sequencing (Bulk RNA-seq) datasets GSE140082[5] and GSE30161[6] were downloaded from GEO using the “GEOquery” R package.

[0038] 2. Cellular senescence and ovarian aging-related genes (CSOARG)

[0039] First, we screened out cell senescence-related genes from the literature. A previous study [7] found a total of 314 ovarian senescence-related genes. The SenMayo genome [8] identified a total of 125 genes associated with aging phenotypes. Ovarian senescence-related genes and SenMayo genome genes were combined to generate a cell senescence and ovarian senescence-related gene (CSOARG) prediction model. Prognosis-related genes were screened to construct a prognostic model. Batch RNA-seq and scRNA-seq analyses were performed on the predictive function of the model. Clinical samples collected at Jinshan Hospital were used to analyze the relationship between model gene expression and clinical characteristics. Then, the hub gene WNK1 was screened and verified at the cellular level in vitro.

[0040] 3. Establishment of CSOASRG Signatures in Ovarian Cancer

[0041] Univariate Cox regression analysis was applied to screen prognostic genes from CSOARGs in the TCGA-OV dataset. Then, LASSOCox regression analysis was performed by the “glmnet” R package to construct a CSOARG prognostic model. The risk score was calculated as follows: Riskscore = ∑coef(gene i)*expression(gene i). Time-dependent receiver operating characteristic (ROC) analysis was performed using the “time ROC” R package to evaluate the efficiency of the model. According to Riskscore = ∑coef(gene i)*expression(gene i), patients in the TCGA-OV dataset were divided into two groups, high-risk group and low-risk group, according to the average risk score. The overall survival (OS) time of the two subgroups was compared by Kaplan-Meier analysis using the “survival” and “survminer” R packages and displayed with Kaplan-Meier plots. The “forestplot” package was used to draw forestplot plots. The calibration curve was used to test the accuracy of the nomogram. A nomogram predicting the probability of 5-year OS was constructed. The calibration curve was used to test the accuracy of the nomogram. The nomogram and calibration curve were generated by the “rms” R package. The Kyoto Encyclopedia of Genomes (KEGG) and Gene Ontology (GO) were analyzed using the “org.Hs.eg.db” and “clusterProfiler” R packages [9] to display the functional classification and related pathways of the prognostic CSOARGs.

[0042] 4. Immune function analysis

[0043] The single-sample gene set enrichment analysis (ssGSEA) algorithm was used to estimate the expression of model genes in 29 immune cell subsets using the “GSVA” R package

[10] . CIBERSORT (https: / / cibersort.stanford.edu / ) was used to analyze the relative proportions of 22 immune cell subsets between the two risk subgroups

[18] . The “limma” R package was used to determine the differential expression of 79 immune checkpoint genes from published papers [13-15] between the two risk subgroups.

[0044] 5. Prediction of immunotherapy response

[0045] The immune epigenetic score (IPS) for ovarian cancer was downloaded from the Cancer Immunobiome Atlas (TCIA) database (https: / / tcia.at / ). The IPS for the two risk subgroups were compared. The tumor immune dysfunction and rejection (TIDE) score was calculated using the TIDE database website (http: / / tide.dfci.harvard.edu)

[16] .

[0046] 6. Drug sensitivity prediction The pRRophetic software package

[17] predicts the half-maximal inhibitory concentration (IC50) representing drug sensitivity. The predicted differences in sensitivity to commonly used clinical drugs between the low-risk group and the high-risk group in the TCGA-OV dataset were compared.

[0047] 7. Analysis of Genomic Instability

[0048] Tumor mutation burden (TMB), homologous recombination deficiency (HRD) score, somatic alterations, and single nucleotide variations (SNV) data of ovarian cancer patients were downloaded from the TCGA database. TMB and HRD scores were compared between the two risk subgroups. Somatic alterations and single nucleotide variations in the two risk subgroups were analyzed and visualized using the “maftools” R package.

[0049] 8. Single-cell RNA-seq Analysis

[0050] TME analysis was based on the GSE165897 dataset [4]. Cell quality control was performed using the “Seurat” R package, and the gene expression of each cell was normalized with default parameters

[18] . In this dataset, each cell was defined by cell, and cell clusters were displayed by the T-distributed stochastic neighbor embedding (t-SNE) algorithm. The communication signal network between tumor cells and plasma cells was detected by the “NicheNETr” R package

[19] .

[0051] 9. Machine learning.

[0052] A prognostic model was constructed in the TCGA-OV database using Python software and the "PyCaret" package using machine learning methods. The model was established by features such as age, FIGO stage, and expression of eight prognostic genes. Fifteen classification algorithms were stratified and ten-fold cross-validated and compared using the "PyCaret" Python package (https: / / pycaret.org / ) to find the best model for predicting the prognosis of ovarian cancer patients.

[0053] 10. Clinical Tissue Sample Collection

[0054] Ovarian cancer tissues, paracancerous tissues, or non-malignant tumor tissues were collected from Jinshan Hospital Affiliated to Fudan University and stored in liquid nitrogen after surgery (non-cancer group, n = 24; cancer group, n = 16). The Ethics Committee of Jinshan Hospital approved the human ethics approval of this study (approval number: JIEC-2023-S68). 11. Cell culture

[0055] Human immortalized ovarian surface epithelial cell lines IOSE-80 (derived from normal ovarian surface epithelium; OriGene Technologies, Inc. ATCC) and SK-OV-3 (ascites-type ovarian endometrioid adenocarcinoma; ATCC) were cultured. Human ovarian cancer cell line OVCAR-3 (derived from ovarian adenocarcinoma in ascites; ATCC, American Type Culture Collection, Virginia, USA) was cultured with RPMI-1640 medium (Thermo Fisher) containing 20% ​​fetal bovine serum (FBS; Thermo Fisher). All cells were cultured in a humidified incubator at 37°C and 5% CO2, and their identity was confirmed by short tandem repeat (STR) analysis. In addition, the cells were tested for pathogens and mycoplasma.

[0056] 12. RNA Extraction and Quantitative Real-time Reverse Transcription-PCR (qRT-PCR)

[0057] Total RNA from tissues and cells was extracted using RNA-Quick Purification Kit (Shanghai Yishan Biotechnology Co., Ltd., Shanghai, China) and reverse transcribed into cDNA using qPCR RT kit (Roche Diagnostics, Indiana, USA).

[0058] 13. Protein Extraction and Western Blotting

[0059] Using Minute TMTotal protein was extracted using the SD-001 kit (Beijing Invent Biotechnology Co., Ltd.), and 1% phenylmethylsulfonyl fluoride (Biyuntian) and 1% phosphatase inhibitor (Nanjing Keygen Biotechnology Co., Ltd.) were added. Proteins were separated using 7.5% SDS-PAGE and then transferred to a 0.45 mm PVDF membrane. The transfer buffer containing 10% methanol and 0.1% SDS was diluted with distilled water. The primary antibodies were rabbit anti-WNK1 (1:500 dilution, product number: 28357-1-AP) and rabbit anti-Vaculin (1:5000 dilution, product number: 28357-1-AP), purchased from Wuhan Mitaka Proteintech Co., Ltd. (Wuhan, Hubei), and incubated overnight at 40°C. The secondary antibody was horseradish peroxidase-conjugated goat anti-rabbit IgG (1:10000 dilution; Proteintech). The signal was detected using the BeyoECL Moon kit (Biyuntian), and then quantified using ImageJ software.

[0060] 14. Detection of Cell Viability and Migration

[0061] To detect cell viability, cells were first transfected with WNK1-siRA (si-WNK1) and then re-cultured in 96-well plates (6000 cells per well for A2780 and 8000 cells per well for OVCAR-3). Cell viability was assessed using a cell counting kit-8 (CCK-8; Shanghai Yishan Biotechnology Co., Ltd.) at 0, 24, 48, and 72 h. A2780 and OVCAR-3 cells were transfected with si-WNK1 (sequence: 5'-GCGACGACUACGAGAUAAATT-3' (SEQ ID NO: 1)) or negative control siRNA (sequence: 5'-UUCUCCGAACGUGUCACGUTT-3' (SEQ ID NO: 2)) for 2 days. After harvesting the cells by centrifugation and resuspending them in serum-free medium, the cells were cultured in the upper chamber of a Transwell plate (Corning Incorporated, Corning, NY, USA). At the same time, complete medium supplemented with 20% FBS was added to the lower chamber of the Transwell plate. After 48 hours of culture, the cells on the bottom of the Transwell were fixed with 4% paraformaldehyde for 15 minutes and stained with crystal violet for 30 minutes. Migrating cells were photographed in three random fields of view using an inverted microscope (Olympus, Tokyo, Japan) and counted using ImageJ (National Institutes of Health, Bethesda, USA). In the wound healing assay, cells transfected with siRNA were seeded in 6-well plates, and when the cells reached 90% confluence, the cells were scraped with a 200 μL pipette tip. The original scratch width at 0 hours and the scratch width after 48 hours of culture were measured. The wound healing percentage was calculated using the following formula: (original scratch width after 48 hours of culture) / (original scratch width)×100%.

[0062] 15. Statistical Analysis

[0063] All statistical analyses were performed using R software (v4.2.1, R Foundation for Statistical Computing, Vienna, Austria). The t-test and Wilcoxon rank-sum test were used to analyze the differences between the two groups. Multiple groups were compared using one-way analysis of variance (ANOVA) followed by Tukey's multiple comparison test. Data are presented as mean ± standard deviation (SD). A P value less than 0.05 was defined as statistically significant.

[0064] (2) Results and analysis

[0065] 1. Construction of a prognostic model for ovarian cancer

[0066] We selected cellular senescence and ovarian aging-related genes (CSOARG) and prognosis-related genes from TCGA, and finally used LASSO regression analysis to develop an 8-gene model with an AUC of 0.68 for the prediction of 5-year overall survival. Figure 2 : The construction process of the prognostic model. (A) shows the screening of genes related to the prognosis of cellular senescence and ovarian aging. (B) shows the best candidate genes selected in the LASSO model. (C) shows the LASSO coefficients of prognosis-related genes, where each curve represents a gene. (D) shows the time-dependent ROC curve of the 5-year overall survival in the TCGA-OV cohort to evaluate the predictive performance of the model.

[0067] 2. Analysis of the clinical significance of the prognostic model

[0068] The risk score of the samples was calculated by the model and the patients were divided into high-risk and low-risk groups. Univariate Cox regression analysis showed that the risk score was associated with patient prognosis, age and FIGO stage. Survival analysis showed that the high-risk group had a poor prognosis. The constructed nomogram was used to predict the 5-year survival rate, and its reliability was verified by the calibration curve. Figure 3 Clinical functional analysis of the prognostic model. (A) The distribution of risk scores in TCGA and the distribution of patients in high-risk and low-risk groups according to overall survival (OS) status are shown, and the heat map shows the expression profiles of eight CSOARGs (cellular senescence and ovarian aging-related genes). (B) The forest plot showing the results of univariate Cox regression analysis between candidate CSOARGs expression and overall survival, and the results of univariate Cox regression analysis of eight model genes are shown. (D) The Kaplan-Meier survival curves of patients in high-risk and low-risk groups are shown. (E) The Kaplan-Meier survival curves of patients in sequential and young groups are shown. (F) The nomogram of the TCGA-OV training cohort integrating CSOARG risk score, age, stage, and grade is shown. (G) The calibration plot of the 5-year OS probability in the TCGA-OV cohort is shown. (H) The bar chart showing the gene ontology (GO) enrichment analysis of differentially expressed genes in high-risk and low-risk groups is shown. (I) Bar graph showing the KEGG enrichment analysis of differentially expressed genes in the high-risk and low-risk groups.

[0069] Prediction of immunotherapy response

[0070] By comparing the IPS scores of the high-risk group and the low-risk group, it was found that the low-risk group had higher expression of PD-1 and CTL-4 molecules and better response to immunotherapy. The Tide score showed that the high-risk group had more severe T cell rejection and dysfunction, indicating that its T cell-related cellular immunity was weaker. This also means that the high-risk group has a weaker response to immunotherapy. Figure 4 Analysis of immunotherapy response in high-risk and low-risk groups. (AD) Comparison of immunophenotypic scores (IPS) between low-risk and high-risk groups. (A) Both PD-1 and CTLA-4 were negatively expressed. (B) Only PD-1 was positively expressed. (C) Only CTLA-4 was positively expressed. (D) Both PD-1 and CTLA-4 were positively expressed. (EH) Comparison of tumor microenvironment scores calculated by TIDE. (E) Total TIDE score. (F) T cell dysfunction score. (G) T cell exclusion score. (H) Microsatellite instability score.

[0071] IV. Prediction of drug sensitivity

[0072] The high-risk group had higher IC50 values ​​for chemotherapy drugs (such as cisplatin, 5-FU, and camptothecin). The high-risk group also had higher IC50 values ​​for targeted therapy drugs (such as gefitinib, nilotinib, bosutinib) and endocrine drugs (such as temozolomide). This suggests that the high-risk group is more responsive to inferior chemotherapy and targeted therapy. Figure 5 Association of risk models with chemotherapeutic drugs, endocrine therapy, and targeted therapy. (AC) Models predicted sensitivity to chemotherapeutic drugs. Results estimated that low-risk patients had lower IC50 values ​​for 5-fluorouracil, capecitabine, and cisplatin. (DH) Models predicted sensitivity to targeted therapy. Results estimated that low-risk patients had lower IC50 values ​​for gefitinib, nilotinib, bosutinib, selumetinib, and dabrafenib. (I) Models predicted sensitivity to endocrine therapy. Results estimated that low-risk patients had lower IC50 values ​​for temozolomide.

[0073] 5. Machine learning selects the best algorithm to improve prediction accuracy

[0074] By comparing 14 algorithms using the machine learning software package "pycaret", the optimal model was selected, and the AUC after optimization reached 0.82. This model has better predictive performance than other models reported in the literature.

[0075] See Figure 6Application of machine learning in optimizing prognostic models. (A) Heatmap compares the performance of various machine learning models on the TCGA-OV dataset. The heatmap shows the accuracy, AUC, recall, precision, F1 score, Kappa value, and Mathews correlation coefficient (MCC) of 14 classifiers. (B) Feature importance graph showing the best model (Extra Trees classifier) ​​in predicting 5-year overall survival of TCGA-OV. (C) ROC curve showing the best model (Extra Trees classifier) ​​in predicting 5-year overall survival of TCGA-OV. (D) AUC values ​​of the prognostic models in this article are compared with those in other literature.

[0076] ExtraTreesClassifier(bootstrap=False,ccp_alpha=0.0,class_weight=None,criterion='gini',max_depth=None,max_features='sqrt',max_leaf_nodes=None,max_samples=None,min_impurity _decrease=0.0,min_samples_leaf=1,min_samples_split=2,min_weight_fraction_leaf=0.0,n_estimators=100,n_jobs=-1,oob_score=False,random_state=1168,verbose=0,warm_start=False)

[0077] VI. Experimental verification of the biological function of the key gene WNK1

[0078] WNK1 is an oncogene that is highly expressed in ovarian cancer and can promote tumor proliferation and migration. Figure 7:Expression level and function of WNK1 in ovarian cancer. (A) The expression of model genes in patients with premenopause (<45 years old) and menopause and postmenopause (>=45 years old) in the TCGA database was compared. (B) The differential expression of prognostic model genes in clinical patient tissues was verified by qRT-PCR. (C) The differential expression of WNK1 RNA level in ovarian cancer cells was verified by qRT-PCR. (D) The differential expression of WNK1 protein level in cells was verified by Western blot analysis. (E) The knockdown efficiency of WNK1 in ovarian cancer was verified. (F) The effect of WNK1 knockdown on the proliferation function of A2780 ovarian cancer cells was verified by CCK-8 assay. (G) The effect of WNK1 knockdown on the proliferation function of OVCAR-3 ovarian cancer cells was verified by CCK-8 assay. (H,I) The effect of WNK1 knockdown on the migration function of A2780 ovarian cancer cells was verified by scratch assay (scale bar: 50μm). (J, K) The effect of WNK1 knockdown on the migration function of OVCAR-3 ovarian cancer cells was verified by scratch assay (scale bar: 200 μm). (L, M) The effect of WNK1 knockdown on the migration function of ovarian cancer cells was verified by Transwell migration assay (scale bar: 200 μm).

[0079] References

[0080] [1] Zhang, B., et al., Exploration of pyroptosis-associated prognosticgene signature and lncRNA regulatory network in ovarian cancer. Comput BiolMed, 2023.164:p.107343.

[0081] [2] Liu, L., et al., Developing four cuproptosis-related lncRNAssignature to predict prognosis and immune activity in ovarian cancer. JOvarian Res, 2023.16(1):p.88.

[0082] [3]S.,et al.,Development of a necroptosis-related gene signature and the immune landscape in ovarian cancer.J Ovarian Res,2023.16(1):p.82.

[0083]

[10] K.Zhang,EPErkan,S.Jamalzadeh,J.Dai,N.Andersson,K.Kaipio,T.Lamminen,N.Mansuri,K.Huhtinen,O.Jarpénen,S.Hietanen,J.Oikkonen,J.Hynninen,A.Virtanen,A. S.Hautaniemi,A. Longitudinal single-cell RNA-seqanalysis reveals stress-promoted chemoresistance in metastatic ovariancancer,Sci Adv 8(8)(2022)eabm1831.

[0084]

[11] S.Kommoss,B.Winterhoff,ALOberg,GEKonecny,C.Wang,SMRiska,JBFan,MJMaurer,C.April,V.Shridhar,F.Kommoss,Adu Bois,F.Hilpert,S.Mahner,K.Baumann,W.Schroeder,A.Burges,U.Canzler,J.Chien,ACEmbleton,M.Parmar,R.Kaplan,T.Perren,LCHartmann,ELGoode,SCDowdy,J.Pfisterer,Bevacizumab May Differentially Offer Improvements withProliferative and Mesenchymal Molecular Subtypes,Clin Cancer Res 23(14)(2017)3794-3801.

[0085] [4]J.S.Ferriss,Y.Kim,L.Duska,M.Birrer,D.A.Levine,C.Moskaluk,D.Theodorescu,J.K.Lee,Multi-gene expression predictors of single drugresponses to adjuvant chemotherapy in ovarian carcinoma:predicting platinumresistance,PLoS One 7(2)(2012)e30550.

[0086] [5]KSRuth,FRDay,J.Hussain,A.Martínez-Marchal,CEAiken,A.Azad,DJThompson,L.Knoblochova,H.Abe,JLTarry-Adkins,JMGonzalez,P.Fontanillas,A.Claringbould,OBBakker,PGakker,RGSulemers,CWTerater. o,S.Turon,M.Horikoshi,K.Lin,NCOnland-Moret,A.Sankar,EPTHertz,PNTimshel,V.Shukla,R.Borup,KWOlsen,P.Aguilera,M.Ferrer-Roda,Y.Huang,S.Stankovic,P.Timmers,TUAhelizane,BZAdeh,Naderi,A. ILAndrulis,AMarnold,KJAronson,A.Augustinsson,S.Bandinelli,CMBarbieri,RNBeaumont,H.Becher,MWBeckmann,S.Benonisdottir,S.Bergmann,M.Bochud,E.Boerwinkle,SEBojesen,MKBolla,DIBoomsma,N. Bowker,JABrody,L.Broer,JEBuring,A.Campbell,H.Campbell,JECastelao,E.Catamo,SJChanock,G.Chen evix-Trench,M.Ciullo,T.Corre,FJCouch,A.Cox,L.Crisponi,SSCross,F.Cucca,K.Jene,GDSmith,E.de Geus,R.de Mutsert,I.De Vivo,EWDemerath,J.Dennis,AMDunning,M.Dwek,M.Eriksson,T.Esko,PAFasching,JDFaul,L.Ferrucci,N.Franceschini,TMFrayling,M.Gago-Dominguez,M.Mezzavilla,M.García-Closas,C.Giger,Giles,H.G.Grallert, D.F. Gudbjartsson, V. Gudnason, P. Guénel, C.A. Haiman, N. P. Hall, C. Hayward, C. He, W. He, G. Heiss, MK JLHopper, JJHottenga, F.Hu, D.Hunter, MAIkram, R.D.Jackson, M.D.R.Joaquim, E.M.John, P.K.Joshi, D.Karasik, S.L.R.Kardia, C.Kartsonaki, R.Karlsson, C.M.Kitahara, I.Kolcic, C.Kooperberg, P.Kraft, A.W.Kurian, Z.Kutalik, M.La Bianca, G.LaChance, C.Langenberg, L.J.Launer, J.SELaven, D.A.Lawlor, L.LeMarchand, J.Li, A.Lindblom, S.Lindstrom, T.Lindstrom, M.Linet, Y.Liu, S.Liu, J.Luan, R. PKEMagnusson,M.Mangino,A.Mannermaa,B.Marco,J.Marten,NGMartin,H.Mbarek,B.McKnight,SEMedland,J.Meisinger,T.Meitinger,J.Menni,A.Metzpalu,L.Milani,RLMilne,GWMontgomery,DOMook-Ka namori,A.Mulas,AMMulligan,A.Murray,MANalls,A.Newman,R.Noordam,T.Nutile,DRNyholt,AFOlshan,H.Olsson,JNPainter,AVPatel,NLPedersen,N.Perjakova,A.Peters,U.Peters,PDPPharoah,O.Pol. asek,E.Porcu,BMPsaty,I.Rahman,G.Rennert,HSRennert,PMRidker,SMRing,A.Robino,LMRose,FRRosendaal,J.Rossouw,I.Rudan,R.Rueedi,D.Ruggiero,CFSala,E.Saloustros,DPSandler,Sanna,JS. awyer,J.Sarnowski,D.Schlessinger,MKSchmidt,MJSchoemaker,KESchrout,J.Scott,S.Shekari,A.Shrikhande,AVSmith,BHSmith,JASmith,R.Sorice,MCSouthey,TDSpector,JJSpinelli,MDStampfer,MD. JBJvan Meurs,K.Strauch,U.Styrkarsdottir,AJSwerdlow,T.Tanaka,LRTeras,A.Teumer,U. NJTimpson,D.Toniolo,M.Traglia,MATroester,T.Truong,J.Tyrrell,AGUitterlinden,S.Ulivi,CMVachon,V.Vitart,U. P.Vollenweider,H. Q.Wang,NJWareham,CRWeinberg,DRWeir,ANWilcox,KWvan Dijk,G.Willemsen,JFWilson,BHRWolffenbuttel,A.Wolk,ARWood,W.Zhao,M.Zygmunt,Z.Chen,L.Li,L.Franke,S.Burgess,P.Deelen,THPers,ML CYAndersen, A.Pujol, AJLopez-Contreras, J.A.Daniel, K.Stefansson, J.Chang-Claude, YTvan derSchouw, KLLunetta, DIChasman, D.F.Easton, J.A.Visser, SEOzanne, SHNamekawa, P.Solc, JMMurabito, KKOng, E.R.Hoffmann, A.Murray, I.Roig, JRBPerry, Genetic insights into biological mechanisms governing human ovarian ageing, Nature 596(7872)(2021)393-397.

[0087] [6] D.Saul, R.L.Kosinsky, EJAtkinson, M.L.Doolittle, senescent cells and predicts senescence-associated pathways across tissues, Nat Commun 13(1)(2022)4827.

[0088] [7]T.Wu,E.Hu,S.Xu,M.Chen,P.Guo,Z.Dai,T.Feng,L.Zhou,W.Tang,L.Zhan,X.Fu,S.Liu,X.Bo,G.Yu,clusterProfiler 4.0:A universal enrichment tool forinterpreting omics data,Innovation(Camb)2(3)(2021)100141.

[0089] [8]Y.He,Z.Jiang,C.Chen,X.Wang,Classification of triple-negativebreast cancers based on Immunogenomic profiling,J Exp Clin Cancer Res 37(1)(2018)327.

[0090] [9]S.Hanzelmann,R.Castelo,J.Guinney,GSVA:gene set variation analysisfor microarray and RNA-seq data,BMC Bioinformatics 14(2013)7.

[0091]

[10] A.M.Newman,C.L.Liu,M.R.Green,A.J.Gentles,W.Feng,Y.Xu,C.D.Hoang,M.Diehn,A.A.Alizadeh,Robust enumeration ofcell subsets from tissue expressionprofiles,Nat Methods 12(5)(2015)453-7.

[0092]

[11] D.M.Pardoll,The blockade of immune checkpoints in cancerimmunotherapy,Nat Rev Cancer 12(4)(2012)252-64.

[0093]

[12] K.S.Campbell,A.K.Purdy,Structure / function ofhuman killer cellimmunoglobulin-like receptors:lessons from polymorphisms,evolution,crystalstructures and mutations,Immunology 132(3)(2011)315-25.

[0094]

[13] H.Afrache,P.Gouret,S.Ainouche,P.Pontarotti,D.Olive,Thebutyrophilin(BTN)gene family:from milk fat to the regulation ofthe immuneresponse,Immunogenetics 64(11)(2012)781-94.

[0095]

[14] P.Jiang,S.Gu,D.Pan,J.Fu,A.Sahu,X.Hu,Z.Li,N.Traugh,X.Bu,B.Li,J.Liu,G.J.Freeman,M.A.Brown,K.W.Wucherpfennig,X.S.Liu,Signatures of T celldysfunction and exclusion predict cancer immunotherapy response,Nat Med 24(10)(2018)1550-1558.

[0096]

[15] P.Geeleher,N.Cox,R.S.Huang,pRRophetic:an R package for predictionofclinical chemotherapeutic response from tumor gene expression levels,PLoSOne 9(9)(2014)e107468.

[0097]

[16] Y.Hao,S.Hao,E.Andersen-Nissen,WMMauck,3rd,S.Zheng,A.Butler,MJLee,AJWilk,C.Darby,M.Zager,P.Hoffman,M.Stoeckius,E.Papalexi, EPMimitou,J.Jain,A.Srivastava,T.Stuart,LMFleming,B.Yeung,AJRogers,JMMcElrath,CABlish,R.Gottardo,P.Smibert,R.Satija,Integrated analysis of multimodal single-cell data, Cell 184(13)(2021)3573-3587.e29.

[0098]

[17] R. Browaeys, W. Saelens, Y. Saeys, NicheNet: modeling intercellular communication by linking ligands to target genes, Nat Methods 17(2)(2020)159-162.

[0099] The above is only a preferred embodiment of the present invention. It should be pointed out that ordinary technicians in this technical field can make several improvements and supplements without departing from the method of the present invention. These improvements and supplements should also be regarded as the scope of protection of the present invention.

Claims

1. A prediction model based on machine learning for judging the prognosis of ovarian cancer and immunotherapy response, characterized in that: Including the determination of the expression levels of eight marker genes: ANGPTL4, AREG, CXCL10, FANCB, GMPR, IGF1, LYG1 and WNK 1.

2. The prediction model according to claim 1, characterized in that Contains an operating manual, which contains the following contents: Riskscore=LYG1*-0.23654+AREG*0.087816+CXCL10*-0.06992+GMPR*-0.15972+WNK1*0.120974+IGF1*0.052135+ANGPTL4*0.099811+FANCB*-0.20232 If the score Riskscore<-0.028, it is considered that the ovarian cancer patient is sensitive to chemotherapy and immunotherapy, and when the score Riskscore≧-0.028, it is considered that the ovarian cancer patient is not sensitive to chemotherapy and immunotherapy.

3. Use of a gene combination kit in the preparation of reagents for predicting ovarian cancer prognosis and immunotherapy response, characterized in that: The gene combination is the expression level of eight marker genes including ANGPTL4, AREG, CXCL10, FANCB, GMPR, IGF1, LYG1 and WNK1.

4. The use according to claim 3, characterized in that: It also includes an operating manual, which contains the following contents: Riskscore=LYG1*-0.23654+AREG*0.087816+CXCL10*-0.06992+GMPR*-0.15972+WNK1*0.120974+IGF1*0.052135+ANGPTL4*0.099811+FANCB*-0.20232 If the score Riskscore<-0.028, it is considered that the ovarian cancer patient is sensitive to chemotherapy and immunotherapy, and when the score Riskscore≧-0.028, it is considered that the ovarian cancer patient is not sensitive to chemotherapy and immunotherapy.

5. A gene combination kit for predicting the prognosis and immunotherapy response of ovarian cancer, characterized in that the gene combination is the expression level of the eight-linked marker genes of ANGPTL4, AREG, CXCL10, FANCB, GMPR, IGF1, LYG1 and WNK1.

6. The kit according to claim 5, characterized in that It also includes an operating manual, which contains the following contents: Riskscore=LYG1*-0.23654+AREG*0.087816+CXCL10*-0.06992+GMPR*-0.15972+WNK1*0.120974+IGF1*0.052135+ANGPTL4*0.099811+FANCB*-0.20232 If the score Riskscore<-0.028, it is considered that the ovarian cancer patient is sensitive to chemotherapy and immunotherapy, and when the score Riskscore≧-0.028, it is considered that the ovarian cancer patient is not sensitive to chemotherapy and immunotherapy.

Images