Construction method of CpG island methylation phenotypic model for predicting tumor drug resistance and application of CpG island methylation phenotypic model

Abstract

The invention provides a construction method of a CpG island methylation phenotypic model for predicting tumor drug resistance and application of the CpG island methylation phenotypic model. Through the model provided by the invention, prediction of the tumor drug resistance CpG island methylation phenotype can be realized, and the model can be used for predicting the drug sensitivity and prognosis of tumor drug resistance patients and screening candidate drugs for treating the tumor drug resistance patients, provides actual support for further clinical research, and has a wide application prospect. And guidance and help are provided for future accurate personalized treatment.

Description

Technical Field

[0001] This invention belongs to the field of intelligent medicine and relates to a method for constructing a CpG island methylation phenotype model for predicting tumor drug resistance and its application. Background Technology

[0002] There are many molecular mechanisms underlying tumor drug resistance, such as increased drug efflux, alterations in drug targets, activation of cell survival pathways, and inactivation of cell death pathways. Therefore, strong molecular heterogeneity exists between different drug-resistant patients, and even between different tumor cells within the same drug-resistant patient.

[0003] However, to date, it is difficult to accurately and reliably classify drug-resistant tumors into subtypes, thus hindering effective guidance for personalized treatment. Therefore, there is a need to develop novel models or methods that can accurately predict and differentiate different drug-resistant phenotypes based on tumor molecular characteristic data, overcoming the challenge of heterogeneity and providing crucial evidence for developing targeted anti-drug-resistant treatment plans in clinical practice. Summary of the Invention

[0004] In view of this, in order to overcome the shortcomings of the prior art, the present invention is proposed.

[0005] The first aspect of this invention discloses a method for predicting the CpG island methylation phenotype (CR_CIMP phenotype) of tumor drug resistance, the method comprising:

[0006] Obtain CpG site data from cancer patients with drug-resistant tumors;

[0007] The CpG site data is input into the classification prediction model to obtain the classification results of CR_CIMP+ or CR_CIMP-. The CpG sites of CR_CIMP+ show a consistent high methylation level, while the CpG sites of CR_CIMP- show a non-consistent high methylation level.

[0008] In this invention, the consistent hypermethylation level of CpG sites means that the hypermethylation level of each CpG site tends to be consistent. For example, the hypermethylation level can be quantified by the β value. If the β value of each CpG site tends to be consistent, then the CpG sites exhibit a consistent hypermethylation level. Conversely, inconsistent hypermethylation level of CpG sites means that the hypermethylation level of each CpG site tends to be inconsistent.

[0009] In some embodiments, the training method of the classification prediction model includes:

[0010] 1) Obtain CpGs site data and classification labels from patients with drug-resistant tumors, including CR_CIMP+ and CR_CIMP-.

[0011] 2) Feature selection of CpGs site data;

[0012] 3) Use the selected CpGs site data features to train the model as a CR_CIMP phenotype classification prediction model.

[0013] In some embodiments, the CpG sites include: cg13444964, cg23570590, cg22621867, cg24935773, cg05702851, cg07209546, cg08750951, cg19786920, cg00611789, cg00157477, cg09211399, cg09510 202, cg13397359, cg06007331, cg19113375, cg17853216, cg01933836, cg21714809, cg0 5229236, cg16848054, cg04274978, cg25813864, cg10742957, cg19814309, cg23868141.

[0014] A second aspect of the present invention provides a model for predicting the CR_CIMP phenotype, which is obtained by the training method of the classification prediction model described in the first aspect of the present invention.

[0015] In this invention, the model for predicting the CR_CIMP phenotype can distinguish the CR_CIMP phenotype of tumor-resistant patients based on CpGs site data. The CR_CIMP phenotype is divided into CR_CIMP+ and CR_CIMP-.

[0016] In some embodiments, the CpGs site data of the tumor-resistant patients includes CpGs site methylation level data of the tumor-resistant patients.

[0017] The third aspect of the present invention provides any of the following applications:

[0018] (1) The application of the model described in the second aspect of the present invention in predicting drug sensitivity in patients with drug-resistant tumors;

[0019] (2) Application of the model described in the second aspect of the present invention in predicting the prognosis of patients with drug-resistant tumors;

[0020] (3) The application of the model described in the second aspect of the present invention in screening candidate drugs for treating patients with drug-resistant tumors;

[0021] (4) Application of the model described in the second aspect of the present invention in the recommended treatment of patients with tumor drug resistance.

[0022] The fourth aspect of the present invention provides the following method:

[0023] A method for predicting drug sensitivity in cancer patients with drug resistance, the method being performed by a computer device, the method comprising the following steps:

[0024] Obtain CR_CIMP phenotype data of tumor drug-resistant patients: Use a classification prediction model to process CpG site data of tumor drug-resistant patients and predict the CR_CIMP phenotype of tumor drug-resistant patients;

[0025] Predicting drug sensitivity in drug-resistant tumor patients: Predicting drug sensitivity in drug-resistant tumor patients based on the CR_CIMP phenotype.

[0026] In some embodiments, the classification prediction model is the model described in the second aspect of the present invention.

[0027] In some embodiments, the sensitivity of CR_CIMP- and CR_CIMP+ to drugs varies depending on the drug-resistant tumor and the drug.

[0028] In this invention, the tumors include, but are not limited to, adrenocortical carcinoma, bladder urothelial carcinoma, breast cancer, cervical squamous cell carcinoma, cervical endometrial adenocarcinoma, bile duct carcinoma, colonic adenocarcinoma, lymphoid tumors, esophageal cancer, glioblastoma multiforme, head and neck squamous cell carcinoma, renal chromophobe carcinoma, renal clear cell carcinoma, renal papillary cell carcinoma, leukemia, glioma, hepatocellular carcinoma, mesothelial cell carcinoma, ovarian cancer, pancreatic cancer, pheochromocytoma and paraganglioma, prostate cancer, rectal cancer, malignant sarcoma, melanoma, gastric cancer, testicular germ cell tumors, thyroid cancer, thymic carcinoma, endometrial cancer, uterine sarcoma, uveal melanoma, myeloma, lymphoma, lung cancer (lung adenocarcinoma, lung squamous cell carcinoma), sarcoma, anal cancer, melanoma, retinoblastoma, and bladder cancer.

[0029] The fourth aspect of the present invention also provides the following method:

[0030] A method for predicting the prognosis of patients with drug-resistant tumors, the method being performed by a computer device, the method comprising the following steps:

[0031] Obtain CR_CIMP phenotype data of tumor drug-resistant patients: Use a classification prediction model to process CpG site data of tumor drug-resistant patients and predict the CR_CIMP phenotype of tumor drug-resistant patients;

[0032] Predicting the prognosis of drug-resistant tumor patients: Predicting the prognosis of drug-resistant tumor patients based on the CR_CIMP phenotype.

[0033] In some embodiments, the classification prediction model is the model described in the second aspect of the present invention.

[0034] In some embodiments, within LGG_Temozolomide, the prognosis of CR_CIMP- is worse than that of CR_CIMP+. Within LGG_Temozolomide, CR_CIMP- has a higher histological grade, a higher tumor positivity rate, and is associated with older age.

[0035] In BLCA_Gemcitabine, patients with CR_CIMP+ had higher M stage, N stage, T stage, and Tumor stage.

[0036] The fourth aspect of the present invention also provides the following method:

[0037] A method for screening candidate drugs for treating drug-resistant cancer patients, the method being performed using computer equipment, the method comprising the following steps:

[0038] Obtain CR_CIMP phenotype data of tumor drug-resistant patients: Use a classification prediction model to process CpG site data of tumor drug-resistant patients and predict the CR_CIMP phenotype of tumor drug-resistant patients;

[0039] Screening for candidate drugs to treat drug-resistant tumor patients: Screening for candidate drugs to treat drug-resistant tumor patients based on the CR_CIMP phenotype of drug-resistant patients.

[0040] In some embodiments, the classification prediction model is the model described in the second aspect of the present invention.

[0041] In some embodiments, oncoPredict is used to predict the IC50 response of a drug to all patients. 50 The values ​​are then compared, and the IC values ​​between CR_CIMP+ and CR_CIMP- are determined. 50 The value has a significantly lower IC value within a certain subtype. 50 Drugs with a p.adjust value (p.adjust < 0.01) are predicted to be effective in treating this subtype.

[0042] A fourth aspect of the present invention also provides a method for recommending drugs to treat patients with drug-resistant tumors, the method being performed by a computer device, the method comprising the following steps:

[0043] Obtain CR_CIMP phenotype data of tumor drug-resistant patients: Use a classification prediction model to process CpG site data of tumor drug-resistant patients and predict the CR_CIMP phenotype of tumor drug-resistant patients;

[0044] Recommended medications for treating drug-resistant tumor patients: Medications for treating drug-resistant tumor patients are recommended based on the CR_CIMP phenotype of drug-resistant patients.

[0045] In some embodiments, the classification prediction model is the model described in the second aspect of the present invention.

[0046] In some embodiments, in LGG_Temozolomide, the candidate or recommended drugs for CR_CIMP- include one or more of AZD6482_2169, AZD8055_1059, KU-55933_1030, and ZM447439_1050.

[0047] In some embodiments, candidate or recommended drugs for CR_CIMP+ in LGG_Temozolomide include Afatinib_1032, AGI-5198_1913, Alisertib_1051, AT13148_2170, BIBR-1532_2043, Dactinomycin_1911, Daporinad_1248, Docetaxel_1007, Eg5_9814_1712, and Eleph. antin_1835, Entinostat_1593, Erlotinib_1168, Fludarabine_1813, Foretinib_2040, GDC0810_1925, GNE-317 _1926, IGF1R_3801_1738, Lapatinib_1558, Leflunomide_1578, Linsitinib_1510, Niraparib_1177, Nutlin-3a One or more of the following: (-)_1047, NVP-ADW742_1932, Olaparib_1017, Oxaliplatin_1089, Oxaliplatin_1806, P22077_1933, Paclitaxel_1080, PAK_5339_1730, Palbociclib_1054, SB505124_1194, Tamoxifen_1199, Venetoclax_1909, and Voronostat_1012.

[0048] In some embodiments, candidate or recommended drugs for CR_CIMP- in PAAD_Temozolomide include AGI-5198_1913, AGI-6780_1634, AMG-319_2045, AZD1208_1449, AZD5991_1720, AZD6482_2169, BIBR-1532_2043, BMS-754807_2171, Cyclophosphamide_1512, CZC24832_1615, Doramapimod_1042, and Entinostat_1593. , EPZ004777_1237, EPZ5676_1563, Fludarabine_1813, GSK2578215A_1927, GSK343_1627, Irinotecan_1088, IWP-2_1576, JAK1_8709_17 18. JQ1_2172, LCL161_1557, LJI308_2107, LY2109761_1852, MIRA-1_1931, Nelarabine_1814, Niraparib_1177, NU7441_1038, Nutlin-3a One or more of the following: (-)_1047, OF-1_1853, Olaparib_1017, Oxaliplatin_1806, PCI-34051_1621, Picolinici-acid_1635, PRIMA-1MET_1131, RO-3306_1052, RVX-208_1625, TAF1_5496_1732, Venetoclax_1909, WZ4003_1614, and Zoledrnate_1802.

[0049] In some embodiments, in PAAD_Temozolomide, candidate or recommended drugs for CR_CIMP+ include one or more of Acetalax_1804, ERK_6604_1714, PD0325901_1060, SCH772984_1564, Selumetinib_1736, Trametinib_1372, Ulixertinib_1908, and VX-11e_2096.

[0050] In some embodiments, the classification prediction model is the model described in the second aspect of the present invention.

[0051] The fourth aspect of the present invention also provides the following method:

[0052] A method for treating a patient with drug-resistant tumors, the method comprising the steps of administering to the patient a candidate drug for treating the drug-resistant tumor selected in this application or a drug recommended in this application for treating the drug-resistant tumor. For example, for patients with low-grade gliomas treated with temozolomide, or for patients with CR_CIMP-, one or more of AZD6482_2169, AZD8055_1059, KU-55933_1030, and ZM447439_1050.

[0053] A fifth aspect of the present invention provides a system for predicting the sensitivity of a drug to a patient with drug-resistant tumors, the system comprising:

[0054] CR_CIMP phenotype prediction module: includes a classification prediction model, which predicts the CR_CIMP phenotype of tumor-resistant patients based on CpGs site data;

[0055] Drug sensitivity prediction module: Predicts the sensitivity of drug-resistant tumor patients to drugs based on the CR_CIMP phenotype of drug-resistant patients, and obtains the classification results of whether drug-resistant tumor patients are sensitive to drugs;

[0056] Output module: Used to output classification results.

[0057] A fifth aspect of the invention also provides a system for predicting the prognosis of patients with drug-resistant tumors, the system comprising:

[0058] CR_CIMP phenotype prediction module: includes a classification prediction model, which predicts the CR_CIMP phenotype of tumor-resistant patients based on CpGs site data;

[0059] Prognostic prediction module: Predicts the prognosis of drug-resistant tumor patients based on the CR_CIMP phenotype of drug-resistant patients, and obtains the classification results of the prognosis of drug-resistant tumor patients.

[0060] Output module: Used to output classification results.

[0061] A fifth aspect of the present invention also provides a system for screening candidate drugs for treating drug-resistant tumor patients, the system comprising:

[0062] CR_CIMP phenotype prediction module: includes a classification prediction model, which predicts the CR_CIMP phenotype of tumor-resistant patients based on CpGs site data;

[0063] Candidate drug screening module: Based on the CR_CIMP phenotype of drug-resistant patients, candidate drugs for treating drug-resistant tumor patients are screened, and candidate drugs for treating drug-resistant tumor patients are obtained;

[0064] Output module: Used to output the filtering results.

[0065] A fifth aspect of the present invention also provides a drug recommendation system for treating patients with drug-resistant tumors, the system comprising:

[0066] CR_CIMP phenotype prediction module: includes a classification prediction model, which predicts the CR_CIMP phenotype of tumor-resistant patients based on CpG site data of tumor-resistant patients;

[0067] Drug recommendation module: Recommends drugs for treating drug-resistant tumor patients based on the CR_CIMP phenotype of drug-resistant patients, and screens out candidate drugs for treating drug-resistant tumor patients;

[0068] Output module: Used to output recommendation results.

[0069] The sixth aspect of the present invention provides any of the following products:

[0070] (1) A computer device, the computer device comprising a first computing device, the first computing device comprising a memory and a processor:

[0071] Memory: Used to store program instructions;

[0072] Processor: Calling program instructions, when the program instructions are executed, are used to perform the steps of the method for predicting the sensitivity of a drug to a tumor drug-resistant patient as described in the fourth aspect of the present invention, or the steps of the method for predicting the prognosis of a tumor drug-resistant patient as described in the fourth aspect of the present invention, or the steps of the method for screening candidate drugs for treating tumor drug-resistant patients as described in the fourth aspect of the present invention.

[0073] (2) A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method for predicting the sensitivity of a drug-resistant tumor patient to a drug as described in the fourth aspect of the present invention, or the steps of the method for predicting the prognosis of a drug-resistant tumor patient as described in the fourth aspect of the present invention, or the steps of the method for screening candidate drugs for treating drug-resistant tumor patients as described in the fourth aspect of the present invention.

[0074] The advantages and beneficial effects of this invention are as follows:

[0075] This invention provides a method for constructing a predictive model of CpG island methylation phenotypes for tumor drug resistance and its application. The model provided by this invention can predict the CpG island methylation phenotypes of tumor drug resistance, and can be used to predict drug sensitivity and prognosis in patients with drug-resistant tumors, as well as to screen candidate drugs for treating these patients. This provides practical support for further clinical research and guidance and assistance for future precise personalized treatment. Attached Figure Description

[0076] Figure 1 This is a schematic flowchart of a method for predicting tumor drug resistance CpG island methylation phenotype (CR_CIMP phenotype) provided in an embodiment of the present invention;

[0077] Figure 2 This is a schematic diagram of the training method of the classification prediction model provided in the embodiment of the present invention;

[0078] Figure 3 This is a schematic flowchart of a method for predicting the sensitivity of tumor drug-resistant patients to drugs, provided by an embodiment of the present invention.

[0079] Figure 4 This is a schematic flowchart of a method for predicting the prognosis of patients with drug-resistant tumors, provided by an embodiment of the present invention.

[0080] Figure 5 This is a schematic flowchart of a method for screening candidate drugs for treating drug-resistant tumor patients provided by an embodiment of the present invention;

[0081] Figure 6 This is a schematic flowchart of a method for recommending drugs to treat drug-resistant tumor patients provided by an embodiment of the present invention;

[0082] Figure 7 This is a schematic diagram of a system for predicting drug sensitivity in tumor drug-resistant patients provided by an embodiment of the present invention;

[0083] Figure 8 This is a schematic diagram of a system for predicting the prognosis of patients with drug-resistant tumors, provided by an embodiment of the present invention;

[0084] Figure 9 This is a schematic diagram of a system for screening candidate drugs for treating drug-resistant tumor patients provided by an embodiment of the present invention;

[0085] Figure 10 This is a schematic diagram of a drug recommendation system for treating drug-resistant tumor patients provided in an embodiment of the present invention;

[0086] Figure 11 This is a schematic diagram of a computer device provided in an embodiment of the present invention;

[0087] Figure 12 This is a graph showing the consensus clustering results for LGG_Temozolomide;

[0088] Figure 13 This is a diagram of the PAAD_Gemcitabine consensus clustering results;

[0089] Figure 14 This is a graph showing the BLCA_Gemcitabine consensus clustering results;

[0090] Figure 15 This is a flowchart of the CR_CIMP subtype prediction model establishment process;

[0091] Figure 16 It is the ROC curve of the subtype prediction model on the training set;

[0092] Figure 17 It is a consensus clustering identification of the independent validation set CR_CIMP subtype;

[0093] Figure 18 This is a comparative chart of clinical characteristics of CR_CIMP subtypes on an independent validation set;

[0094] Figure 19 It is the methylation spectrum of 25 features of the random forest model in the independent validation set;

[0095] Figure 20 This is the Kaplan-Meier survival curve of the subtype LGG_Temozolomide;

[0096] Figure 21 This is a diagram showing the differences in clinical characteristics among subtypes of LGG_Temozolomide;

[0097] Figure 22 This is a graph showing the differences in clinical characteristics among subtypes of BLCA_Gemcitabine;

[0098] Figure 23 This is a comparison of subtype global genomic variations in LGG_Temozolomide;

[0099] Figure 24 This is a display of the top 10 gene mutations by mutation frequency in LGG_Temozolomide;

[0100] Figure 25 This is a comparison of the number of subtype copy number variations in PAAD_Gemcitabine;

[0101] Figure 26 This is a diagram showing the top 10 gene mutations in PAAD_Gemcitabine by mutation frequency;

[0102] Figure 27 These are heatmaps and volcano maps showing differentially expressed genes among LGG_Temozolomide subtypes;

[0103] Figure 28 Visualization of CpG sites in the MEOX2 promoter region;

[0104] Figure 29 This is a display of the correlation between gene expression and methylation levels in LGG_Temozolomide;

[0105] Figure 30 This is a hypothesis regarding the mechanism by which CR_CIMP-resistance is caused by STAC promoter hypomethylation;

[0106] Figure 31 These are the top 10 KEGG pathways with the highest subtype differentially expressed genes in LGG_Temozolomide;

[0107] Figure 32 This is a Hallmark pathway scoring map showing significant activity differences among LGG_Temozolomide isotypes;

[0108] Figure 33 The difference in immunogenicity scores among subtypes in LGG_Temozolomide;

[0109] Figure 34 This is a comparison of pathway scores between subtypes in PAAD_Gemcitabine;

[0110] Figure 35 It is a prediction of candidate drugs for the CR_CIMP subtype;

[0111] Figure 36 It predicts the target pathway of drugs;

[0112] Figure 37 This is a hypothesis regarding the mechanism by which AZD6482 and AZD8055 reverse drug resistance. Detailed Implementation

[0113] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0114] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be performed in the order they appear herein, or may be performed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be performed sequentially or in parallel.

[0115] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0116] Figure 1 This invention provides a method for predicting the tumor drug resistance CpG island methylation phenotype (CR_CIMP phenotype), the method comprising:

[0117] 101: Obtain CpG site data from patients with drug-resistant tumors;

[0118] 102: Input the CpGs site data into the classification prediction model to obtain the classification results of CR_CIMP+ or CR_CIMP-. The CpGs sites of CR_CIMP+ show a consistent high methylation level, while the CpGs sites of CR_CIMP- show a non-consistent high methylation level.

[0119] Figure 2 This invention provides a training method for a classification prediction model, the method comprising:

[0120] 201: Obtain CpGs site data and classification labels from patients with drug-resistant tumors, wherein the classification labels include: CR_CIMP+, CR_CIMP-;

[0121] 202: Feature selection of CpGs site data;

[0122] 203: Use the selected CpGs site data features to train a model as a CR_CIMP phenotypic classification prediction model.

[0123] In some embodiments, the CpG sites in step 203 include: cg13444964, cg23570590, cg22621867, cg24935773, cg05702851, cg07209546, cg08750951, cg19786920, cg00611789, cg00157477, cg09211399, cg09 510202, cg13397359, cg06007331, cg19113375, cg17853216, cg01933836, cg21714809, cg 05229236, cg16848054, cg04274978, cg25813864, cg10742957, cg19814309, cg23868141.

[0124] In some embodiments, the CpGs site data of the tumor drug-resistant patients includes methylation level data of the CpGs sites of the tumor drug-resistant patients. The classification labels CR_CIMP+ and CR_CIMP- are distinguished based on the methylation level data of the CpGs sites of the tumor drug-resistant patients. The CpGs sites of CR_CIMP+ show a consistent high methylation level, while the CpGs sites of CR_CIMP- show a non-consistent high methylation level.

[0125] In some embodiments, the classification prediction model includes, but is not limited to, random forest model, linear regression model, logistic regression model, Lasso regression model, Ridge regression model, linear discriminant analysis model, nearest neighbor model, Naive Bayes model, decision tree model, perceptron model, neural network model, support vector machine model, AdaBoost model, GBDT model, XGBoost model, LightGBM model, or CatBoost model.

[0126] In some embodiments, the classification prediction model is selected from the random forest model.

[0127] In some embodiments, the step of feature selection of CpG site data during the training of a random forest model further includes: inputting the selected CpG site data features into a cross-validated random forest model.

[0128] In some embodiments, cross-validation is selected from 3x cross-validation, 5x cross-validation, and 10x cross-validation.

[0129] In some embodiments, the selected CpGs site data features are input into a 3x cross-validation random forest model. CR_CIMP+ and CR_CIMP- samples are randomly divided into 3 equal groups. One group of CR_CIMP+ and one group of CR_CIMP- samples are randomly combined as the test set, and the remaining two groups of CR_CIMP+ and CR_CIMP- are combined as the training set. The AUROC value in the test set is calculated.

[0130] In some embodiments, the selected CpGs site data features are input into a 5x cross-validation random forest model. The CR_CIMP+ and CR_CIMP- samples are randomly divided into 5 equal groups. One group of CR_CIMP+ and one group of CR_CIMP- samples are randomly combined as the test set, and the remaining 4 groups of CR_CIMP+ and CR_CIMP- are combined together as the training set.

[0131] In some embodiments, the selected CpGs site data features are input into a 10x cross-validation random forest model. CR_CIMP+ and CR_CIMP- samples are randomly divided into 10 equal groups. Three groups of CR_CIMP+ and three groups of CR_CIMP- samples are randomly combined as the test set, and the remaining seven groups of CR_CIMP+ and CR_CIMP- are combined together as the training set.

[0132] In some embodiments, cross-validation is repeated 10 times.

[0133] In some embodiments, the parameters used to train the random forest model are ntree = 500, mtry = sqrt(nfeatures).

[0134] In some embodiments, the R package Boruta is used to perform feature selection on CpGs site data.

[0135] In some embodiments, feature selection includes using Boruta to shuffle the original features to form shadow features, and then filtering features that contribute to the prediction by comparing the importance of the original features and the shadow features.

[0136] In some embodiments, the prediction method further includes a method for CR_CIMP phenotype identification, comprising the following steps:

[0137] Obtain patient DNA methylation profiles and drug response information;

[0138] Patients with drug response of "Stable disease" and "Clinical progressive disease" were classified as drug-resistant patients. A specific cancer type treated with a specific drug was defined as a condition. Conditions with multiple drug-resistant samples were retained for analysis.

[0139] For each condition, the standard deviation of the β value of each CpG site across all samples is calculated to reflect the level of variation.

[0140] CpG sites with highly variable methylation levels were selected to conduct consensus clustering to explore the overall methylation heterogeneity of drug-resistant samples and determine the optimal number of clusters K.

[0141] CpG sites with highly variable methylation levels exhibit a drug-resistant cluster with a consistent high methylation level, which is defined as CR_CIMP+, while clusters without this consistent high methylation level are defined as CR_CIMP-.

[0142] In some embodiments, the condition of more than 30 drug-resistant samples is retained for analysis.

[0143] In some embodiments, CpG sites with high methylation levels in the top 1% are selected to complete consensus clustering to explore the overall methylation heterogeneity of drug-resistant samples and determine the optimal number of clusters K.

[0144] In some embodiments, consensus clustering is performed using the R package ConsensusClusterPlus.

[0145] In some embodiments, the consensus matrix and the consensus cumulative distribution function are used to determine the optimal number of clusters K.

[0146] In some embodiments, the number of clusters K is 2.

[0147] In some embodiments, the β value is processed according to the following criteria:

[0148] 1) Filter and align to CpG sites on the X and Y chromosomes;

[0149] 2) Filter CpG sites with missing values ​​exceeding 20%;

[0150] 3) Filter samples with more than 20% missing values;

[0151] 4) The remaining missing values ​​in the data are filled in using the R package.

[0152] Figure 3 This invention provides a method for predicting the sensitivity of drug-resistant tumor patients to drugs. The method is performed by a computer device and includes the following steps:

[0153] 301 Obtaining CR_CIMP phenotype data from tumor-resistant patients: Using a classification prediction model to process CpGs site data from tumor-resistant patients and predict their CR_CIMP phenotype;

[0154] 302 Predicting drug sensitivity in drug-resistant tumor patients: Predicting drug sensitivity in drug-resistant tumor patients based on the CR_CIMP phenotype.

[0155] In some embodiments, CpGs locus data of tumor drug-resistant patients are input into a classification prediction model to predict the CR_CIMP phenotype of tumor drug-resistant patients. The CpGs loci include: cg13444964, cg23570590, cg22621867, cg24935773, cg05702851, cg07209546, cg08750951, cg19786920, cg00611789, and cg00157. 477, cg09211399, cg09510202, cg13397359, cg06007331, cg19113375, cg17853216, cg01933836, cg2 1714809, cg05229236, cg16848054, cg04274978, cg25813864, cg10742957, cg19814309, cg23868141.

[0156] In some embodiments, the CpGs site data of the tumor-resistant patients includes methylation level data of the CpGs sites of the tumor-resistant patients.

[0157] In some embodiments, the classification prediction model is the model described in the second aspect of the present invention.

[0158] Figure 4 This invention provides a method for predicting the prognosis of patients with drug-resistant tumors. The method is performed by a computer device and includes the following steps:

[0159] 401 Obtaining CR_CIMP phenotype data from cancer-resistant patients: Using a classification prediction model to process CpG site data from cancer-resistant patients and predict their CR_CIMP phenotype;

[0160] 402 Predicting the prognosis of drug-resistant tumor patients: Predicting the prognosis of drug-resistant tumor patients based on the CR_CIMP phenotype.

[0161] In some embodiments, CpGs locus data of tumor drug-resistant patients are input into a classification prediction model to predict the CR_CIMP phenotype of tumor drug-resistant patients. The CpGs loci include: cg13444964, cg23570590, cg22621867, cg24935773, cg05702851, cg07209546, cg08750951, cg19786920, cg00611789, and cg00157. 477, cg09211399, cg09510202, cg13397359, cg06007331, cg19113375, cg17853216, cg01933836, cg2 1714809, cg05229236, cg16848054, cg04274978, cg25813864, cg10742957, cg19814309, cg23868141.

[0162] In some embodiments, the CpGs site data of the tumor-resistant patients includes methylation level data of the CpGs sites of the tumor-resistant patients.

[0163] In some embodiments, the classification prediction model is the model described in the second aspect of the present invention.

[0164] Figure 5 This invention provides a method for screening candidate drugs for treating drug-resistant tumor patients, the method being performed by a computer device, and the method comprising the following steps:

[0165] 501 Obtaining CR_CIMP phenotype data from tumor-resistant patients: Using a classification prediction model to process CpGs site data from tumor-resistant patients and predict their CR_CIMP phenotype;

[0166] 502 Screening for candidate drugs to treat drug-resistant tumor patients: Screening for candidate drugs to treat drug-resistant tumor patients based on the CR_CIMP phenotype of drug-resistant patients.

[0167] In some embodiments, CpGs locus data of tumor drug-resistant patients are input into a classification prediction model to predict the CR_CIMP phenotype of tumor drug-resistant patients. The CpGs loci include: cg13444964, cg23570590, cg22621867, cg24935773, cg05702851, cg07209546, cg08750951, cg19786920, cg00611789, and cg00157. 477, cg09211399, cg09510202, cg13397359, cg06007331, cg19113375, cg17853216, cg01933836, cg2 1714809, cg05229236, cg16848054, cg04274978, cg25813864, cg10742957, cg19814309, cg23868141.

[0168] In some embodiments, the CpGs site data of the tumor-resistant patients includes methylation level data of the CpGs sites of the tumor-resistant patients.

[0169] In some embodiments, the classification prediction model is the model described in the second aspect of the present invention.

[0170] Figure 6 This invention provides a method for recommending drugs to treat drug-resistant tumor patients, which is performed by a computer device and includes the following steps:

[0171] 601: Obtaining CR_CIMP phenotype data of tumor drug-resistant patients: Using a classification prediction model to process CpGs site data of tumor drug-resistant patients and predict the CR_CIMP phenotype of tumor drug-resistant patients;

[0172] 602: Recommended drugs for treating drug-resistant tumor patients: Recommended drugs for treating drug-resistant tumor patients based on the CR_CIMP phenotype of drug-resistant patients.

[0173] In some embodiments, CpGs locus data of tumor drug-resistant patients are input into a classification prediction model to predict the CR_CIMP phenotype of tumor drug-resistant patients. The CpGs loci include: cg13444964, cg23570590, cg22621867, cg24935773, cg05702851, cg07209546, cg08750951, cg19786920, cg00611789, and cg00157. 477, cg09211399, cg09510202, cg13397359, cg06007331, cg19113375, cg17853216, cg01933836, cg2 1714809, cg05229236, cg16848054, cg04274978, cg25813864, cg10742957, cg19814309, cg23868141.

[0174] In some embodiments, the CpGs site data of the tumor-resistant patients includes methylation level data of the CpGs sites of the tumor-resistant patients.

[0175] In some embodiments, the classification prediction model is the model described in the second aspect of the present invention.

[0176] Figure 7 This invention provides a system for predicting drug sensitivity in tumor drug-resistant patients, the system comprising:

[0177] 701CR_CIMP phenotype prediction module: includes a classification prediction model, which predicts the CR_CIMP phenotype of tumor-resistant patients based on CpGs site data;

[0178] 702 Drug Sensitivity Prediction Module: Predicts the sensitivity of drug-resistant tumor patients to drugs based on the CR_CIMP phenotype of drug-resistant patients, and obtains the classification results of whether drug-resistant tumor patients are sensitive to drugs;

[0179] 703 Output Module: Used to output classification results.

[0180] In some embodiments, the CpGs site data of the tumor-resistant patients includes methylation level data of the CpGs sites of the tumor-resistant patients.

[0181] In some embodiments, the CpGs site data of the tumor drug-resistant patients does not include the methylation level data of the CpGs sites of the tumor drug-resistant patients, and the CR_CIMP phenotype prediction module includes a methylation level quantification module for quantifying the methylation level data of the CpGs site data of the tumor drug-resistant patients.

[0182] In some embodiments, the CpG sites of the tumor-resistant patients include: cg13444964, cg23570590, cg22621867, cg24935773, cg05702851, cg07209546, cg08750951, cg19786920, cg00611789, cg00157477, cg09211399, cg0 9510202, cg13397359, cg06007331, cg19113375, cg17853216, cg01933836, cg21714809, c g05229236, cg16848054, cg04274978, cg25813864, cg10742957, cg19814309, cg23868141.

[0183] Figure 8 This invention provides a system for predicting the prognosis of patients with drug-resistant tumors, the system comprising:

[0184] 801CR_CIMP phenotype prediction module: includes a classification prediction model, which predicts the CR_CIMP phenotype of tumor drug-resistant patients based on CpGs site data;

[0185] 802 Prognostic Prediction Module: Predicts the prognosis of drug-resistant tumor patients based on the CR_CIMP phenotype of drug-resistant patients, and obtains the classification results of the prognosis of drug-resistant tumor patients.

[0186] 803 Output Module: Used to output classification results.

[0187] In some embodiments, the CpGs site data of the tumor-resistant patients includes methylation level data of the CpGs sites of the tumor-resistant patients.

[0188] In some embodiments, the CpGs site data of the tumor drug-resistant patients does not include the methylation level data of the CpGs sites of the tumor drug-resistant patients, and the CR_CIMP phenotype prediction module includes a methylation level quantification module for quantifying the methylation level data of the CpGs site data of the tumor drug-resistant patients.

[0189] In some embodiments, the CpG sites of the tumor-resistant patients include: cg13444964, cg23570590, cg22621867, cg24935773, cg05702851, cg07209546, cg08750951, cg19786920, cg00611789, cg00157477, cg09211399, cg0 9510202, cg13397359, cg06007331, cg19113375, cg17853216, cg01933836, cg21714809, c g05229236, cg16848054, cg04274978, cg25813864, cg10742957, cg19814309, cg23868141.

[0190] Figure 9 This invention provides a system for screening candidate drugs for treating drug-resistant tumor patients, the system comprising:

[0191] 901CR_CIMP phenotype prediction module: includes a classification prediction model, which predicts the CR_CIMP phenotype of tumor-resistant patients based on CpGs site data;

[0192] 902 Candidate Drug Screening Module: Screens candidate drugs for treating drug-resistant tumor patients based on the CR_CIMP phenotype of drug-resistant patients, and obtains candidate drugs for treating drug-resistant tumor patients;

[0193] 903 Output Module: Used to output the filtering results.

[0194] In some embodiments, the CpGs site data of the tumor-resistant patients includes methylation level data of the CpGs sites of the tumor-resistant patients.

[0195] In some embodiments, the CpGs site data of the tumor drug-resistant patients does not include the methylation level data of the CpGs sites of the tumor drug-resistant patients, and the CR_CIMP phenotype prediction module includes a methylation level quantification module for quantifying the methylation level data of the CpGs site data of the tumor drug-resistant patients.

[0196] In some embodiments, the CpG sites of the tumor-resistant patients include: cg13444964, cg23570590, cg22621867, cg24935773, cg05702851, cg07209546, cg08750951, cg19786920, cg00611789, cg00157477, cg09211399, cg0 9510202, cg13397359, cg06007331, cg19113375, cg17853216, cg01933836, cg21714809, c g05229236, cg16848054, cg04274978, cg25813864, cg10742957, cg19814309, cg23868141.

[0197] Figure 10 This invention provides a drug recommendation system for treating drug-resistant tumor patients, the system comprising:

[0198] 1001CR_CIMP phenotype prediction module: includes a classification prediction model, which predicts the CR_CIMP phenotype of tumor drug-resistant patients based on CpGs site data;

[0199] 1002 Drug Recommendation Module: Recommends drugs for treating drug-resistant tumor patients based on the CR_CIMP phenotype of drug-resistant patients, and screens out candidate drugs for treating drug-resistant tumor patients;

[0200] 1003 Output Module: Used to output recommendation results.

[0201] In some embodiments, the CpGs site data of the tumor-resistant patients includes methylation level data of the CpGs sites of the tumor-resistant patients.

[0202] In some embodiments, the CpGs site data of the tumor drug-resistant patients does not include the methylation level data of the CpGs sites of the tumor drug-resistant patients, and the CR_CIMP phenotype prediction module includes a methylation level quantification module for quantifying the methylation level data of the CpGs site data of the tumor drug-resistant patients.

[0203] In some embodiments, the CpG sites of the tumor-resistant patients include: cg13444964, cg23570590, cg22621867, cg24935773, cg05702851, cg07209546, cg08750951, cg19786920, cg00611789, cg00157477, cg09211399, cg0 9510202, cg13397359, cg06007331, cg19113375, cg17853216, cg01933836, cg21714809, c g05229236, cg16848054, cg04274978, cg25813864, cg10742957, cg19814309, cg23868141.

[0204] Figure 11 This invention provides a computer device comprising a first computing device, the first computing device including a memory and a processor.

[0205] Memory: Used to store program instructions;

[0206] Processor: Invokes program instructions, which, when executed, are used to perform the steps of the method for predicting the drug sensitivity of drug-resistant tumor patients as described in the fourth aspect of the present invention, or the steps of the method for predicting the prognosis of drug-resistant tumor patients as described in the fourth aspect of the present invention, or the steps of the method for screening candidate drugs for treating drug-resistant tumor patients as described in the fourth aspect of the present invention.

[0207] In some embodiments, the processor may be one or more.

[0208] In some embodiments, the processor includes, but is not limited to, a controller, an integrated circuit, a microchip, and a computer.

[0209] In some embodiments, the device further includes one or more of a CpGs analyzer, a user interface device, and a user interface device, which are respectively communicatively coupled to the first computing device.

[0210] In this invention, communication coupling refers to the ability of coupled components to exchange data signals with each other, such as electrical signals via a conductive medium, electromagnetic signals via air, and optical signals via an optical waveguide.

[0211] In some embodiments, the CpGs analyzer is used to analyze CpGs data from patients with drug-resistant tumors.

[0212] In some embodiments, the user interface device supports user input of CpGs data from patients with drug-resistant tumors.

[0213] In some embodiments, the user interface device supports receiving analysis results from a CpGs analyzer.

[0214] In some embodiments, the user interface device displays the analysis results received from the CpGs analyzer to the user.

[0215] In some embodiments, the user interface device displays prediction results or filtering results to the user.

[0216] In some embodiments, the user interface device includes, but is not limited to, a local computing device (e.g., a desktop computer, tablet computer, smartphone, etc.) or a remote computing device (e.g., a cloud computing device).

[0217] In some embodiments, the user interface device is communicatively coupled to the first computing device. The user inputs CpGs data of a tumor drug-resistant patient through the user interface device, which is then transmitted to the first computing device. After processing by the first computing device, the prediction results or screening results are displayed by the user interface device.

[0218] In some embodiments, the CpGs analyzer is communicatively coupled to a user interface device, and the user interface device is communicatively coupled to a first computing device. The CpGs analyzer analyzes CpGs data of tumor drug-resistant patients. The analysis results are received by the user interface device and then transmitted to the first computing device. After processing by the first computing device, the prediction results or screening results are displayed by the user interface device.

[0219] Example

[0220] Materials and methods

[0221] Data download and preprocessing:

[0222] TCGA Data Download and Processing: Multi-omics data from TCGA, including DNA methylation profiles, RNA expression profiles, and copy number variation profiles at the gene level, were downloaded from UCSC Xena (https: / / xenabrowser.net / datapages / ). Mutation profiles and patient drug response information from TCGA were also downloaded from GDC (https: / / portal.gdc.cancer.gov / analysis_page?app=Downloads). This invention categorizes patients with drug responses classified as "Stabledisease" and "Clinical progressive disease" as drug-resistant patients. Specifically, a specific cancer type treated with a particular drug is defined as a condition. For example, LGG patients treated with temozolomide are referred to as a condition: LGG_Temozolomide. This invention then retains only the conditions of more than 30 drug-resistant samples for downstream analysis (Table 1). Level 3 DNA methylation profiles (β values) obtained using a 450K array were processed according to the following criteria: 1) CpG sites aligned to the X and Y chromosomes were filtered out; 2) CpG sites with more than 20% missing values ​​were filtered out; 3) Samples with more than 20% missing values ​​were filtered out; 4) Remaining missing values ​​in the data were imputed using the R package impute (v1.76.0). The raw count values ​​from RNA-seq were used for differential expression analysis, and gene expression levels were expressed using TPM values ​​converted to log2(TPM+1).

[0223] Table 1. Number of drug-resistant samples in the three conditions

[0224] Other publicly available data downloads included: 450K manifest probe annotation files downloaded from Illumina (https: / / support.illumina.com / downloads / infinium_humanmethylation450_product_files.html) for annotating CpG sites in gene regions and CpG island-related regions; cancer driver genes (Oncogenes) downloaded from the COSMIC database (https: / / cancer.sanger.ac.uk / census); EMT-related genes downloaded from the dbEMT database (https: / / bioinfo-minzhao.org / dbemt / ); HALLMARK pathway gene sets and cell cycle pathway gene sets were obtained from MSigDB (https: / / www.gsea-msigdb.org / gsea / msigdb); DDR-related gene sets were obtained from studies by Liu et al., Kang et al., and KEGG (https: / / www.genome.jp / kegg / ). Tumor neoantigens, aneuploidy scores, lymphocyte infiltration, and leukocyte percentage were obtained from studies by Thorsson et al. The dataset GSE104293 was obtained from the GEO database and used as an independent validation set for the subtype prediction model. The data sources for all public datasets used in this invention are shown in Table 2.

[0225] Table 2. Sources of all public datasets used in this invention

[0226] Comparison of cancer genomic variations among CR_CIMP subtypes

[0227] Tumor mutation burden (TMB) was calculated using the R package maftools (v2.22.0). Differentially expressed genes were calculated using the R package edgeR (v4.4.1) with raw RNA-seq counts as input. Additionally, gene methylation levels were represented by the mean of CpG sites in gene promoter regions. Gene-level copy number variation was used to measure the overall global copy number variation level of the sample.

[0228] Comparison of biological pathway activities among CR_CIMP subtypes

[0229] The HALLMARK pathway classification was obtained from MsigDB. The R package GSVA (v2.0.5) was used to calculate the single sample gene set enrichment analysis (ssGSEA) score for each sample, representing the activity level of the pathway or functional set. Additionally, this invention uses multiple biomarkers to assess the patient's immunogenicity, including PD-L1 expression, TMB, tumor neoantigens, cytolytic activity (CYT), major histocompatibility complex (MHC) activity, lymphocyte infiltration, and leukocyte percentage. CYT activity was calculated using the geometric mean of GZMA and PRF1 expression levels. MHC activity was calculated using the average gene expression levels of the MHC-I gene set (HLA-A, HLA-B, HLA-C, TAP1, TAP2, NLRC5, PSMB9, PSMB8, and B2M).

[0230] Gene set enrichment analysis

[0231] The KEGG pathway enrichment analysis of HVGs and differentially expressed genes was performed using the R package clusterProfiler (v4.14.4).

[0232] Prediction of drug response in drug-resistant patients

[0233] This invention uses the R package oncoPredict (v1.2) to predict the half-maximal inhibitory concentration (IC50) of a drug in resistant patients. 50 IC 50 The IC50 value is used to reflect the effectiveness of drug treatment. 50 The smaller the value, the more sensitive the patient is to the drug; conversely, the larger the value, the more resistant the patient is. oncoPredict uses cell line gene expression profiles as a training set and, based on tissue sample gene expression profiles from drug-resistant patients, can predict the IC50 of resistant patients for 198 drugs. 50 value.

[0234] Statistical analysis

[0235] All statistical analyses in this invention were performed using R (v4.4.2) software. Comparisons of continuous variables (such as age and various activity scores) among CR_CIMP subtypes were performed using the Wilcoxon rank-sum test. The hypergeometric distribution test was used to assess whether HVCs or CpGs used in the random forest model were significantly enriched in gene promoter regions or CpG island regions. In addition, the hypergeometric distribution test was also used to assess the enrichment of HVGs in two tumor resistance-related functional sets (EMT and Oncogenes). The log-rank test was used to assess prognostic differences among CR_CIMP subtypes, and the Kaplan-Meier (KM) survival curves for the subtypes were plotted using the R package survival (v3.8-3).

[0236] Specific experimental methods:

[0237] Example 1: Method for constructing a tumor drug resistance CpG island methylation phenotype model

[0238] CR_CIMP subtype identification

[0239] For each condition, the standard deviation of the β value for each CpG site across all samples is calculated to reflect the level of variation. Then, the top 1% of CpG sites with high methylation levels (High Variable CpGs, HVCs) are selected for consensus clustering to investigate the overall methylation heterogeneity of drug-resistant samples. The R package ConsensusClusterPlus (v1.70.0) is used to implement consensus clustering with the following parameters: maxK = 6, clusterAlg = "km", distance = "euclidean". The optimal number of clusters K is determined using a combination of the consensus matrix and the cumulative distribution function (CDF). A drug-resistant cluster (subgroup) exhibiting consistent high methylation of HVCs is considered to have the tumor-resistant CpG island methylation phenotype (CR_CIMP+), while the remaining clusters not exhibiting this consistent high methylation pattern are considered CR_CIMP-.

[0240] Based on the 450K manifest file, HVCs are aligned to the promoter regions and CpG island regions of nearby genes. This invention uses the TSS1500, TSS200, 5'UTR, and 1stExon regions as promoter regions. In addition, high variable genes (HVGs) are defined as genes whose promoter regions contain at least one HVC.

[0241] To comprehensively characterize the DNA methylation heterogeneity of drug-resistant tumors, this invention collected drug response information from all patients in the TCGA cohort. Based on the patient's cancer type, a condition was defined as a patient with a specific cancer type resistant to a particular drug. To ensure statistical power, only conditions with more than 30 resistant patients were retained. Ultimately, only three conditions were retained for downstream analysis. For each condition, the top 1% of HVCs were selected for consensus clustering. The final number of clusters, K, was determined using a heatmap of the consensus matrix and the consensus CDF curve. In LGG_Temozolomide, when K=2, the consensus CDF was the flattest, with consensus scores concentrated between 0 and 1, indicating that the clustering results were most stable at this point. Figure 12 A). The heatmap of the consensus matrix also reveals two clear heatmap partitions when K=2 ( Figure 12 B). These results indicate that K=2 is the optimal clustering choice. It is noteworthy that the two clusters identified through consensus clustering exhibit two completely different DNA methylation patterns: C1 shows consistent high methylation of HVCs; in contrast, the methylation levels of these HVCs in C2 are inconsistent, with most showing low methylation. Figure 12 C). Therefore, this invention also defines drug-resistant tumor patients with consistent hypermethylation in HVCs (C1) as CR_CIMP+, and drug-resistant tumor patients (C2) who do not have this pattern of simultaneous hypermethylation at a large number of sites as CR_CIMP-.

[0242] Similar to LGG_Temozolomide, this invention also performed consensus clustering on the other two conditions. In the gemcitabine-resistant pancreatic adenocarcinoma (PAAD) condition (PAAD_Gemcitabine), after selecting the optimal number of clusters K = 2 through consensus CDF (…),… Figure 13 A), the consensus matrix heatmap shows that the distinction between these two categories is very good ( Figure 13 B). Similarly, C2 showed consistent hypermethylation of HVCs ( Figure 13 C). Therefore, in this condition, C2 is defined as CR_CIMP+ and C1 as CR_CIMP-. In the gemcitabine-resistant bladder cancer (BLCA) condition (BLCA_Gemcitabine), C2 is defined as CR_CIMP+ and C1 is defined as CR_CIMP-. Figure 14 ).

[0243] Construction of CR_CIMP subtype prediction model

[0244] In order to classify a new drug-resistant patient into a specific CR_CIMP subtype for LGG_Temozolomide, this invention develops a random forest model to predict the CR_CIMP subtype. Figure 15 Based on the identification of 3,856 HVCs loci for the CR_CIMP subtype, this invention first assessed their importance for CR_CIMP subtype classification using the Boruta method. Boruta shuffles the original features to create shadow features, and then selects features contributing to prediction by comparing the importance of the original and shadow features. Compared to traditional feature selection methods, Boruta can provide a statistical significance level for each feature. Boruta identified 25 loci that are significantly important for CR_CIMP classification (p < 0.05). 0.05): cg13444964, cg23570590, cg22621867, cg24935773, cg05702851, cg0720954 6, cg08750951, cg19786920, cg00611789, cg00157477, cg09211399, cg09510202, cg The selected sites are: 13397359, cg06007331, cg19113375, cg17853216, cg01933836, cg21714809, cg05229236, cg16848054, cg04274978, cg25813864, cg10742957, cg19814309, and cg23868141. Based on these selected sites, a random forest model is then constructed. The model's phenotypic profile was evaluated using triple cross-validation with missing parameters, repeated 10 times. In each repetition, CR_CIMP+ and CR_CIMP- samples were randomly divided into three equal groups. One group of CR_CIMP+ and one group of CR_CIMP- samples were randomly combined as the test set, and the remaining two groups of CR_CIMP+ and CR_CIMP- samples were combined as the training set. The model on the training set used the default parameters (ntree = 500, mtry = sqrt(nfeatures)). The predictions from all test sets during the cross-validation process were combined to calculate the Area Under the Receiver Operating Characteristic Curve (AUROC). The results showed that the AUROC for the CR_CIMP subtype prediction was 1 (…). Figure 16Finally, based on 25 selected CpG sites from all samples, the random forest model was retrained with default parameters as the final prediction model.

[0245] To validate the model's performance on an independent dataset, this invention collected a cohort of temozolomide-treated LGG patients from a public database. This cohort included 63 temozolomide-treated patients, and their pre-treatment methylation profiles, clinical information, and prognostic data were also obtained. Based on post-treatment PFS, 16 patients were identified as temozolomide-resistant (Table 3). Similar to the TCGA samples, this invention also used the top 1% HVCs from this independent validation set for consensus clustering of resistant samples. As expected, two groups of patients with different methylation phenotypes were identified (…). Figure 17 C2 showed consistent hypermethylation of HVCs; therefore, C2 was defined as CR_CIMP+ and C1 as CR_CIMP-. Consistent with TCGA results, patients in the independent validation set with CR_CIMP- showed significantly worse prognosis and significantly older age. Figure 18 This means that CR_CIMP- is a more malignant subtype. To demonstrate the accuracy of the CR_CIMP subtype prediction model, this invention predicts the CR_CIMP subtype of independent datasets based on a developed random forest model. The results show that the CR_CIMP subtype identified by consensus clustering has a high degree of consistency with the CR_CIMP subtype predicted by the prediction model. Figure 19 (Table 3).

[0246] Table 3. CR_CIMP Subtype Identification and Prediction on Independent Datasets

[0247] Example 2: Application of the CR_CIMP phenotypic model in predicting drug sensitivity and prognosis

[0248] To validate the CR_CIMP subtype, this invention first characterized the clinical features of the CR_CIMP subtype. Survival analysis showed that in LGG_Temozolomide, patients with CR_CIMP- had a significantly worse prognosis ( Figure 20 (p < 0.0001), indicating that CR_CIMP- is a more malignant phenotype compared to CR_CIMP+. Further comparison revealed that CR_CIMP- has a higher histological grade ( Figure 21 A) Higher tumor positivity rate ( Figure 21 B) and older ages ( Figure 21C). Furthermore, although both are drug-resistant patients, the two subtypes may still have different responses to the drug. To confirm this hypothesis, oncoPredict was used to predict the drug response of the two subtypes to temozolomide, and the results showed that CR_CIMP- had a larger IC50 value. 50 value( Figure 21 D), indicating that CR_CIMP- is more resistant to drugs. All the above results consistently demonstrate that within LGG_Temozolomide, CR_CIMP- is a more malignant and drug-resistant subtype.

[0249] In BLCA_Gemcitabine, this invention found that CR_CIMP+ patients had higher M stage, N stage, T stage, and Tumor stage ( Figure 22 This indicates that CR_CIMP+ is a more malignant subtype in this condition. The findings of this invention note that the pattern of clinical differences in CR_CIMP subtypes varies across different conditions, depending on the specific cancer type and medication. Correspondingly, in CIMP studies of tumors, different cancer types exhibit different, even completely opposite, conclusions regarding CIMP subtypes.

[0250] The differences in clinical characteristics between the CR_CIMP subtypes prompted this invention to further explore their genomic variant features. For LGG_Temozolomide, the differences between the two subtypes in global genomic variants were first compared, including TMB, tumor neoantigens, and aneuploidy scores. Overall, the CR_CIMP subtype had more TMB, tumor neoantigens, and aneuploidy scores (…). Figure 23 This indicates that CR_CIMP- has more tumor genomic variations, consistent with the conclusion above, suggesting that CR_CIMP- may be a more malignant subtype under this condition. To further investigate subtype-specific gene mutations, this invention compared the differences in mutation frequencies of genes with higher mutation frequencies between the two subtypes. Among the top 10 genes with the highest mutation frequencies, IDH1, TP53, and ATRX were mutated more frequently in CR_CIMIP+; while EGFR and NF1 mutations were significantly enriched in CR_CIMP-. Figure 24In fact, these differentially mutated genes play important roles in the development and progression of gliomas. Specifically, IDH1 mutations can drive DNA methylation variations across the glioma genome, which may be a direct cause of the CIMP phenotype in gliomas. Furthermore, TP53 and ATRX mutations significantly affect temozolomide resistance in glioma drug therapy, providing a new perspective for overcoming drug resistance in glioma treatment and optimizing treatment regimens.

[0251] In addition to LGG_Temozolomide, genomic variation differences at the copy number level were observed in PAAD_Gemcitabine. CR_CIMP+ patients showed more copy number amplification and copy number deletion variants. Figure 25 Consistent with the above, this also demonstrates that CR_CIMP+ is a more malignant phenotype in this condition. Although the two subtypes did not show significant differences in overall mutational burden, this invention explored specific differential mutations in the two subtypes. Among the top 10 most mutated genes, RB1 mutations were significantly enriched in the CR_CIMP+ subtype ( Figure 26 ), and this invention notes that some patients with RB1 and TP53 co-mutations exist in CR_CIMP+ ( Figure 26 Co-mutation of RB1 and TP53 in bladder cancer is a biomarker for immunotherapy, providing potential treatment options for patients resistant to CR_CIMP+.

[0252] Further investigation was conducted into transcriptomic differences between the two subtypes. In LGG_Temozolomide, differential expression analysis revealed 573 genes significantly upregulated in CR_CIMP+ (Benjamini-Hochberg corrected p-value (p.adjust) < 0.05, |log2FC| > 1), while 1753 genes were significantly upregulated in CR_CIMP-. Several genes with the most significant differences, such as MEOX2 and STAC, could serve as biomarkers for glioma prognosis or drug response. Figure 27 These results indicate that the two subtypes not only differ significantly at the overall transcriptome level, but individual key differentially expressed genes may also influence tumor progression and drug response in a particular subtype. To investigate whether MEOX2 expression is regulated by methylation, this invention analyzed the methylation patterns of CpG sites in its promoter region. The results showed that all 15 CpG sites in the MEXO2 promoter region exhibited higher levels of DNA methylation (Δβ > 0.1) in the CR_CIMP+ subtype, with four methylation sites showing a methylation level difference exceeding 0.4 (Δβ > 0.1). Figure 28The consistent high methylation of the promoter region led this invention to hypothesize that the low expression of MEOX2 in CR_CIMP+ might be due to high methylation of its promoter region. Therefore, this invention further calculated the correlation between MEOX2 gene methylation level and expression level, and the results showed a significant negative correlation between MEOX2 expression and methylation. Figure 29 A). These results suggest that CIMP-induced MEOX2 methylation may affect MEOX2 gene expression, thus playing an important role in temozolomide resistance to LGG.

[0253] For the STAC gene, overexpression increases glioma invasion, while inhibiting STAC expression can induce apoptosis, reduce invasion, and suppress EMT, thus making STAC a potential therapeutic target. Similar to MEOX2, STAC also shows a significant negative correlation between methylation levels and gene expression levels. Figure 29 B). Therefore, taking STAC as an example, this invention proposes the hypothesis that abnormal STAC methylation leads to CR_CIMP-resistance (B). Figure 30 In the CR_CIMP- subtype, STAC is highly expressed due to promoter hypomethylation. High STAC expression promotes apoptosis, increases invasion, and activates the EMT process, leading to temozolomide resistance. Therefore, inhibiting STAC expression with STAC inhibitors can suppress apoptosis, reduce invasion, and inhibit the EMT process, thereby reversing drug resistance.

[0254] Furthermore, this invention also performed functional enrichment analysis on differentially expressed genes in the two subtypes. The results showed that genes specific to the CR_CIMP subtype were significantly enriched in tumor-related or drug resistance-related pathways, such as the cAMP signaling pathway and the PI3K / AKT signaling pathway. Figure 31 Besides LGG_Temozolomide, the other two conditions also showed significant gene differential expression among the subtypes, and functional enrichment analysis revealed the enrichment of differentially expressed genes in tumor- or drug resistance-related pathways. These results reveal significant global or specific genomic variations among the CR_CIMP subtypes, thus providing further evidence for the molecular subtyping of CR_CIMP subtypes.

[0255] In LGG_Temozolomide, to investigate the differences in biological pathway activity among CR_CIMP subtypes, the activity of 50 tumor HALLMARK pathways in CR_CIMP subtype patients was scored. The results showed that the activity of most HALLMARK pathways (29 / 50) exhibited significant differences between the two subtypes (p.adjust < 0.05), and most of these activity differences were higher in CR_CIMP-. Based on the HALLMARK pathway classification, this invention found that CR_CIMP- patients had higher activity in developmental, DNA damage, immune, and metabolic-related pathways. Figure 32 In addition, CR_CIMP- patients showed higher activity in the PI3K / AKT / mTOR signaling pathway during hypoxia. These results suggest that CR_CIMP- may be more malignant and more prone to metastasis.

[0256] The activation of immune processes exhibited by CR_CIMP, combined with the higher TMB and more tumor neoantigens mentioned above, prompted this invention to further evaluate the immunophenotypes of these two subtypes. This invention collected several key biomarkers for characterizing immunogenicity, including lymphocyte infiltration, leukocyte percentage, PD-L1 (CD274) expression, MHC activity, CYT activity, TMB, and tumor neoantigens. These key immunogenicity signatures consistently resulted in higher scores in CR_CIMP patients. Figure 33 This suggests that CR_CIMP- may be more effective against immune checkpoint inhibitors (ICIs).

[0257] Furthermore, for PAAD_Gemcitabine, gemcitabine primarily exerts its therapeutic effect by inhibiting DNA synthesis in tumor cells. Therefore, this invention investigated cell cycle activity and DDR-related pathway activity in patients under this condition. Results showed that CR_CIMP+ patients exhibited significantly higher cell cycle and DDR activity (…). Figure 34 This suggests that CR_CIMP+ resistance in PAAD_Gemcitabine may be related to enhanced DDR activity. Previous reports have confirmed that DDR is involved in the gemcitabine resistance process in pancreatic cancer.

[0258] Example 3: Application of the CR_CIMP phenotype model in screening candidate drugs for treating drug-resistant tumor patients.

[0259] The CR_CIMP subtype has been validated from multiple perspectives, including clinical characteristics, genomic variations, and biological activity. The significance of subtype classification lies in its ability to further stratify drug-resistant patients, increasing the demand for precision and personalized medicine. Therefore, this invention further identifies candidate therapeutic agents for each subtype.

[0260] For each condition, the following process is performed to identify candidate drugs: Based on the patient's gene expression profile, this invention utilizes oncoPredict to predict the IC50 of 198 drugs for all patients. 50 The value is then compared between the IC values ​​of the two subtypes. 50 The value has a significantly lower IC value within a certain subtype. 50 Drugs with a p.adjust value (p.adjust < 0.01) are predicted to be effective in treating this subtype. Figure 35 A). For LGG_Temozolomide, four drugs (AZD6482_2169, AZD8055_1059, KU-55933_1030, ZM447439_1050) had lower IC50 in CR_CIMP-. 50 Therefore, these four drugs are considered candidate drugs for treating CR_CIMP- resistant patients. For example, the PI3K inhibitor AZD6482 has been identified as a candidate drug for CR_CIMP-, as it has a lower IC50 value than CR_CIMP+ in CR_CIMP-. 50 value( Figure 35B). AZD6482 inhibits glioma cell proliferation and promotes apoptosis by reducing the activity of the PI3K / AKT / mTOR signaling pathway. Similarly, 34 drugs (Afatinib_1032, AGI-5198_1913, Alisertib_1051, AT13148_2170, BIBR-1532_2043, Dactinomycin_1911, Daporinad_1248, Docetaxel_1007, Eg5_9814_1712, Elephantin_1835, Entinosta) t_1593, Erlotinib_1168, Fludarabine_1813, Foretinib_2040, GDC0810_1925, GNE-317_1926, IGF1 R_3801_1738, Lapatinib_1558, Leflunomide_1578, Linsitinib_1510, Niraparib_1177, Nutlin-3a (-)_1047, NVP-ADW742_1932, Olaparib_1017, Oxaliplatin_1089, Oxaliplatin_1806, P22077_1933, Paclitaxel_1080, PAK_5339_1730, Palbociclib_1054, SB505124_1194, Tamoxifen_1199, Venetoclax_1909, Voronostat_1012) were identified as potential drug candidates effective against CR_CIMP+. For PAAD_Gemcitabine, eight candidate drugs (Acetalax_1804, ERK_6604_1714, PD0325901_1060, SCH772984_1564, Selumetinib_1736, Trametinib_1372, Ulixertinib_1908, and VX-11e_2096) showed an IC50 score of 1 in CR_CIMP+. 50 Lower values, for example, trametinib ( Figure 35C). When the MEK inhibitor trametinib is combined with standard chemotherapy (Nab-paclitaxel and gemcitabine combination), it can inhibit the growth of PAAD cell line AsPC-1 tumor cells. Similarly, in this condition, 41 drugs (AGI-5198_1913, AGI-6780_1634, AMG-319_2045, AZD1208_1449, AZD5991_1720, AZD6482_2169, BIBR-1532_2043, BMS-754807_2171, Cyclophosphamide_1512, CZC24832_1615, Doramapimod_1042, Entinostat_1593, EPZ004777_1) were used. 237, EPZ5676_1563, Fludarabine_1813, GSK2578215A_1927, GSK343_1627, Irinotecan_1088, IWP-2_1576, JAK1_8709_1718, JQ1 _2172, LCL161_1557, LJI308_2107, LY2109761_1852, MIRA-1_1931, Nelarabine_1814, Niraparib_1177, NU7441_1038, Nutlin-3a (-)_1047, OF-1_1853, Olaparib_1017, Oxaliplatin_1806, PCI-34051_1621, Picolinici-acid_1635, PRIMA-1MET_1131, RO-3306_1052, RVX-208_1625, TAF1_5496_1732, Venetoclax_1909, WZ4003_1614, Zoledrnate_1802) are considered candidates for the treatment of CR_CIMP- subtype.

[0261] Furthermore, based on the drug target information provided by Genomics of Drug Sensitivity in Cancer (GDSC), this invention explored the targets and target pathways of these candidate drugs. In LGG_Temozolomide, two of the four drugs predicted to be effective against CR_CIMP- targeted the PI3K / AKT / mTOR signaling pathway. Figure 36A). Consistent with this result, the above study also found that the PI3K / AKT / mTOR signaling pathway is more active in the CR_CIMP- subtype, indicating a potential drug resistance mechanism in this subtype. Combining the targets of these two drugs with the function of the PI3K / AKT / mTOR signaling pathway, this invention proposes a mechanism by which these two drugs may have a potential therapeutic effect on the CR_CIMP- resistant subtype. Figure 37 The drug AZD6482 inhibits PI3K expression, thereby reducing the activity of the PI3K / AKT / mTOR signaling pathway. The drug AZD8055 inhibits this pathway by downregulating mTOR expression. Decreased PI3K / AKT / mTOR signaling pathway activity reduces cell proliferation, increases apoptosis, and ultimately reverses drug resistance in CR_CIMP-. In contrast, among the CR_CIMP+ candidate drugs for PAAD_Gemcitabine, the MAPK / ERK signaling pathway is frequently targeted (7 / 8). Figure 36 B).

[0262] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0263] In the several embodiments provided by this invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative, and in actual implementation, there may be other ways of division, such as multiple modules being combined or integrated into another system, or some features being ignored or not executed.

[0264] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0265] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0266] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0267] The above description of the embodiments is only for understanding the method and core ideas of the present invention. It should be noted that those skilled in the art can make various improvements and modifications to the present invention without departing from the principles of the invention, and these improvements and modifications will also fall within the protection scope of the claims of the present invention.

Claims

1. A method for predicting tumor drug resistance CpG island methylation phenotype (CR_CIMP phenotype), characterized in that, The method includes: Obtain CpG site data from cancer drug-resistant patients; The CpG site data is input into the classification prediction model to obtain the classification results of CR_CIMP+ or CR_CIMP-. The CpG sites of CR_CIMP+ show a consistent high methylation level, while the CpG sites of CR_CIMP- show a non-consistent high methylation level.

2. The prediction method according to claim 1, characterized in that, The training method for the classification prediction model includes: 1) Obtain CpGs site data and classification labels from patients with drug-resistant tumors, including CR_CIMP+ and CR_CIMP-. 2) Feature selection of CpGs site data; 3) Use the selected CpGs site data features to train the model as a CR_CIMP phenotypic classification prediction model; Optionally, the CpG sites in step 3) include: cg13444964, cg23570590, cg22621867, cg24935773, cg05702851, cg07209546, cg08750951, cg19786920, cg00611789, cg00157477, cg09211399, cg09510 202, cg13397359, cg06007331, cg19113375, cg17853216, cg01933836, cg21714809, cg0 5229236, cg16848054, cg04274978, cg25813864, cg10742957, cg19814309, cg23868141.

3. The prediction method according to claim 2, characterized in that, The classification prediction models include random forest model, linear regression model, logistic regression model, Lasso regression model, Ridge regression model, linear discriminant analysis model, nearest neighbor model, Naive Bayes model, decision tree model, perceptron model, neural network model, support vector machine model, AdaBoost model, GBDT model, XGBoost model, LightGBM model, or CatBoost model. Optionally, the classification prediction model is selected from the random forest model; Optionally, the parameters used to train the random forest model are ntree = 500, mtry = sqrt(nfeatures).

4. The prediction method according to claim 2, characterized in that, Feature selection of CpGs site data was performed using the R package Boruta. Optionally, feature selection includes using Boruta to shuffle the original features to form shadow features, and then filtering features that contribute to the prediction by comparing the importance of the original features and the shadow features.

5. The prediction method according to claim 2, characterized in that, The prediction method also includes a method for CR_CIMP phenotype identification, the steps of which are as follows: Obtain patient DNA methylation profiles and drug response information; Patients with drug response of "Stable disease" and "Clinical progressive disease" were classified as drug-resistant patients. A specific cancer type treated with a specific drug was defined as a condition. Conditions with multiple drug-resistant samples were retained for analysis. For each condition, the standard deviation of the β value of each CpG site across all samples is calculated to reflect the level of variation. CpG sites with highly variable methylation levels were selected to conduct consensus clustering to explore the overall methylation heterogeneity of drug-resistant samples and determine the optimal number of clusters K. CpG sites with highly variable methylation levels exhibit a drug-resistant cluster with a consistent high methylation level and are defined as CR_CIMP+, while clusters without this consistent high methylation level are defined as CR_CIMP-. Optionally, conditions of more than 30 drug-resistant samples are retained for analysis; Optionally, select the top 1% of CpG sites with high methylation levels to complete consensus clustering to explore the overall methylation heterogeneity of drug-resistant samples and determine the optimal number of clusters K; Optionally, the R package ConsensusClusterPlus can be used for consensus clustering; Optionally, the consensus matrix and the consensus cumulative distribution function are used to determine the optimal number of clusters K; Optionally, the number of clusters K is 2; Optionally, the β value is processed according to the following criteria: 1) Filter and align to CpG sites on the X and Y chromosomes; 2) Filter CpG sites with missing values ​​exceeding 20%; 3) Filter samples with more than 20% missing values; 4) The remaining missing values ​​in the data are filled in using the R package.

6. A model for predicting the CR_CIMP phenotype, characterized in that, The model is obtained by the training method of the classification prediction model according to any one of claims 2-4.

7. Any of the following applications: (1) The application of the model of claim 6 in predicting drug sensitivity in patients with drug-resistant tumors; (2) The application of the model described in claim 6 in predicting the prognosis of patients with drug-resistant tumors; (3) The application of the model described in claim 6 in screening candidate drugs for treating drug-resistant tumor patients; (4) The application of the model of claim 6 in the recommended treatment of patients with tumor resistance.

8. Any one of the following methods: (1) A method for predicting the sensitivity of tumor drug-resistant patients to drugs, characterized in that, The method is performed by a computer device and includes the following steps: Obtain CR_CIMP phenotype data of tumor drug-resistant patients: Use a classification prediction model to process CpG site data of tumor drug-resistant patients and predict the CR_CIMP phenotype of tumor drug-resistant patients; Predicting drug sensitivity in drug-resistant tumor patients: Predicting drug sensitivity in drug-resistant tumor patients based on the CR_CIMP phenotype; (2) A method for predicting the prognosis of patients with drug-resistant tumors, characterized in that the method is performed by a computer device, and the method includes the following steps: Obtain CR_CIMP phenotype data of tumor drug-resistant patients: Use a classification prediction model to process CpG site data of tumor drug-resistant patients and predict the CR_CIMP phenotype of tumor drug-resistant patients; Predicting the prognosis of drug-resistant tumor patients: Predicting the prognosis of drug-resistant tumor patients based on the CR_CIMP phenotype; (3) A method for screening candidate drugs for treating drug-resistant tumor patients, characterized in that the method is performed by computer equipment, and the method includes the following steps: Obtain CR_CIMP phenotype data of tumor drug-resistant patients: Use a classification prediction model to process CpG site data of tumor drug-resistant patients and predict the CR_CIMP phenotype of tumor drug-resistant patients; Screening candidate drugs for treating drug-resistant tumor patients: Screening candidate drugs for treating drug-resistant tumor patients based on the CR_CIMP phenotype of drug-resistant patients; (4) A method for recommending drugs to treat patients with drug-resistant tumors, characterized in that the method is performed by a computer device and includes the following steps: Obtain CR_CIMP phenotype data of tumor drug-resistant patients: Use a classification prediction model to process CpG site data of tumor drug-resistant patients and predict the CR_CIMP phenotype of tumor drug-resistant patients; Recommended medications for treating drug-resistant tumor patients: Medications for treating drug-resistant tumor patients are recommended based on the CR_CIMP phenotype of drug-resistant patients; Optionally, the classification prediction model is the model described in claim 6; Optionally, in patients with low-grade gliomas treated with temozolomide, candidate or recommended drugs for CR_CIMP- include one or more of AZD6482_2169, AZD8055_1059, KU-55933_1030, and ZM447439_1050. Optionally, for patients with low-grade gliomas treated with temozolomide, candidate or recommended drugs for CR_CIMP+ include Afatinib_1032, AGI-5198_1913, Alisertib_1051, AT13148_2170, BIBR-1532_2043, Dactinomycin_1911, Daporinad_1248, Docetaxel_1007, Eg5_9814_1712, and Elephan. tin_1835, Entinostat_1593, Erlotinib_1168, Fludarabine_1813, Foretinib_2040, GDC0810_1925, GNE-317_ 1926, IGF1R_3801_1738, Lapatinib_1558, Leflunomide_1578, Linsitinib_1510, Niraparib_1177, Nutlin-3a One or more of the following: (-)_1047, NVP-ADW742_1932, Olaparib_1017, Oxaliplatin_1089, Oxaliplatin_1806, P22077_1933, Paclitaxel_1080, PAK_5339_1730, Palbociclib_1054, SB505124_1194, Tamoxifen_1199, Venetoclax_1909, and Voronostat_1012; Optionally, in bladder cancer patients resistant to gemcitabine, candidate or recommended drugs for CR_CIMP- include AGI-5198_1913, AGI-6780_1634, AMG-319_2045, AZD1208_1449, AZD5991_1720, AZD6482_2169, BIBR-1532_2043, BMS-754807_2171, Cyclophosphamide_1512, CZC24832_1615, Doramapimod_1042, Entinostat_1593, and EPZ. 004777_1237, EPZ5676_1563, Fludarabine_1813, GSK2578215A_1927, GSK343_1627, Irinotecan_1088, IWP-2_1576, JAK1_8709_1718 , JQ1_2172, LCL161_1557, LJI308_2107, LY2109761_1852, MIRA-1_1931, Nelarabine_1814, Niraparib_1177, NU7441_1038, Nutlin-3a One or more of the following: (-)_1047, OF-1_1853, Olaparib_1017, Oxaliplatin_1806, PCI-34051_1621, Picolinici-acid_1635, PRIMA-1MET_1131, RO-3306_1052, RVX-208_1625, TAF1_5496_1732, Venetoclax_1909, WZ4003_1614, and Zoledrnate_1802; Optionally, in bladder cancer patients resistant to gemcitabine, candidate or recommended drugs for CR_CIMP+ include one or more of Acetalax_1804, ERK_6604_1714, PD0325901_1060, SCH772984_1564, Selumetinib_1736, Trametinib_1372, Ulixertinib_1908, and VX-11e_2096.

9. Any of the following systems: (1) A system for predicting drug sensitivity in tumor drug-resistant patients, characterized in that, The system includes: CR_CIMP phenotype prediction module: includes a classification prediction model, which predicts the CR_CIMP phenotype of tumor-resistant patients based on CpGs site data; Drug sensitivity prediction module: Predicts the sensitivity of drug-resistant tumor patients to drugs based on the CR_CIMP phenotype of drug-resistant patients, and obtains the classification results of whether drug-resistant tumor patients are sensitive to drugs; Output module: Used to output classification results; (2) A system for predicting the prognosis of patients with drug-resistant tumors, characterized in that the system comprises: CR_CIMP phenotype prediction module: includes a classification prediction model, which predicts the CR_CIMP phenotype of tumor-resistant patients based on CpGs site data; Prognostic prediction module: Predicts the prognosis of drug-resistant tumor patients based on the CR_CIMP phenotype of drug-resistant patients, and obtains the classification results of the prognosis of drug-resistant tumor patients. Output module: Used to output classification results; (3) A system for screening candidate drugs for treating drug-resistant tumor patients, characterized in that the system comprises: CR_CIMP phenotype prediction module: includes a classification prediction model, which predicts the CR_CIMP phenotype of tumor-resistant patients based on CpGs site data; Candidate drug screening module: Based on the CR_CIMP phenotype of drug-resistant patients, candidate drugs for treating drug-resistant tumor patients are screened, and candidate drugs for treating drug-resistant tumor patients are obtained. Output module: Used to output the filtering results; (4) A drug recommendation system for treating drug-resistant tumor patients, characterized in that the system comprises: CR_CIMP phenotype prediction module: includes a classification prediction model, which predicts the CR_CIMP phenotype of tumor-resistant patients based on CpGs site data; Drug recommendation module: Recommends drugs for treating drug-resistant tumor patients based on the CR_CIMP phenotype of drug-resistant patients, and screens out candidate drugs for treating drug-resistant tumor patients; Output module: Used to output recommendation results; Optionally, the classification prediction model is the model described in claim 6.

10. Any one of the following products: (1) A computer device, characterized in that, The computer device includes a first computing device, which includes a memory and a processor: Memory: Used to store program instructions; Processor: invokes program instructions, which, when executed, are used to perform the steps of the method for predicting the sensitivity of a drug to a tumor-resistant patient as described in claim 7 (1), or the steps of the method for predicting the prognosis of a tumor-resistant patient as described in claim 7 (2), or the steps of the method for screening candidate drugs for treating tumor-resistant patients as described in claim 7 (3). (2) A computer-readable storage medium, characterized in that it stores a computer program thereon, which, when executed by a processor, implements the steps of the method for predicting the sensitivity of a drug-resistant tumor patient to a drug as described in claim 7 (1), or the steps of the method for predicting the prognosis of a drug-resistant tumor patient as described in claim 7 (2), or the steps of the method for screening candidate drugs for treating drug-resistant tumor patients as described in claim 7 (3).

Images