Cancer staging method and electronic device

Abstract

Provided are a cancer staging method and an electronic device. A staging prediction model obtained by means of integration of a plurality of classification models performs prediction on cancer staging corresponding to input methylation data sets, ensuring the prediction accuracy and simultaneously improving the universality of cancer staging. The method comprises: acquiring a methylation data set of a target subject; and inputting the methylation data set into a staging prediction model, and outputting a cancer staging value of the target subject, the staging prediction model being obtained by means of integration of n classification models, model parameters of the n classification models being obtained by means of performing optimization by using a grid retrieval method, the n classification models being obtained from m classification models by means of performing screening by using a cross validation method, and the m classification models being obtained by means of performing independent training by using the same data set, wherein m>n, m is an integer greater than 2, and n is an integer greater than 1.

Description

Cancer staging method and electronic device

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to the Chinese patent application filed with the State Intellectual Property Office of the People's Republic of China on June 29, 2023, with application number 202310790387.4 and invention name "A cancer staging method, device and electronic device", the entire contents of which are incorporated by reference into this application. Technical Field

[0003] The present disclosure relates to the field of biological information technology, and in particular to a cancer staging method and electronic device. Background Art

[0004] Cancer staging is a method to determine the extent of cancer development and spread. Accurate cancer staging is conducive to formulating the most reasonable treatment plan for cancer patients and effectively judging the prognosis of cancer.

[0005] In the existing technology, cancer staging is usually performed using methods such as clinical and pathological examinations. However, clinical examinations alone (blood tests, radiological examinations, endoscopy, etc.) provide limited information for cancer staging and have limitations. Although pathological examinations can more accurately determine the stage, surgery is required to obtain pathological sections, and not all tumors require surgical treatment. In addition, some patients may have undergone chemotherapy and radiotherapy before the pathological examination, which may underestimate the true stage of the tumor to a certain extent. Therefore, the method of using pathological examinations still has limitations. In addition, different cancers have different staging systems, and some cancers do not have suitable staging methods. Conventional staging methods are not universal.

[0006] Summary of the Invention

[0007] The present disclosure provides a cancer staging method and electronic device, which are used to predict the cancer stage corresponding to an input methylation dataset through a staging prediction model obtained by integrating multiple classification models, thereby ensuring prediction accuracy while improving the universality of cancer staging.

[0008] In a first aspect, an embodiment of the present disclosure provides a method for cancer staging, the method comprising:

[0009] Obtain a methylation dataset of the target object;

[0010] Inputting the methylation dataset into a staging prediction model to output a cancer staging value of the target subject;

[0011] Among them, the stage prediction model is obtained by integrating n classification models, the model parameters of the n classification models are optimized using the grid retrieval method, the n classification models are screened from m classification models using the cross-validation method, and the m classification models are independently trained using the same data set, where m>n, m is an integer greater than 2, and n is an integer greater than 1.

[0012] As an optional implementation, obtaining a methylation dataset of a target object includes:

[0013] Acquire methylation data corresponding to a target methylation site of a target object, and determine the methylation dataset based on the methylation data corresponding to the target methylation site;

[0014] The target methylation site is determined based on the degree of difference in methylation data of the same methylation site corresponding to different cancer stage values, or the correlation between the methylation data corresponding to the methylation site and different cancer stage values.

[0015] As an optional embodiment, the staging prediction model further includes a standardization model, which is used to standardize the methylation data in the methylation dataset by removing the mean and normalizing the variance.

[0016] As an optional implementation, the staging prediction model further includes a dimensionality reduction model, and the dimensionality reduction model is used to perform dimensionality reduction processing on the methylation data in the methylation dataset according to the PCA method.

[0017] As an optional implementation, the staging prediction model is determined in the following manner:

[0018] Using the data set to independently train the m classification models, and determine the classification accuracy of each of the m classification models;

[0019] According to the classification accuracy, k classification models are selected from the m classification models.

[0020] As an optional implementation manner, after selecting k classification models from m classification models according to the classification accuracy, the method further includes:

[0021] Using a cross-validation method, the n classification models are selected from the k classification models, where m≥k≥n; and / or,

[0022] The model parameters of the screened classification model are adjusted using the grid retrieval method.

[0023] As an optional implementation, the method of using cross-validation to select the n classification models from the k classification models includes:

[0024] Using the data set, cross-validate the k classification models respectively to obtain evaluation index values ​​of the k classification models;

[0025] According to the evaluation index values ​​of the k classification models, n classification models are screened out from the k classification models.

[0026] As an optional implementation manner, after obtaining the data set, the method further includes:

[0027] The methylation data corresponding to the target methylation site is screened out from the methylation data contained in the data set, and the m classification models are independently trained using the screened methylation data.

[0028] As an optional embodiment, the data set includes sample sets of different target subjects, each sample set includes methylation data corresponding to each methylation site, and each sample set corresponds to a cancer stage value; the target methylation site is determined by:

[0029] Dividing the data set into multiple subsets according to the cancer stage values ​​corresponding to the sample sets, wherein each subset corresponds to one cancer stage value, and the sample sets included in different subsets have different cancer stage values ​​corresponding to them;

[0030] For each methylation site, a target methylation site is determined based on the methylation data corresponding to the methylation site in multiple subsets.

[0031] As an optional implementation manner, determining the target methylation site based on the methylation data corresponding to the methylation site in multiple subsets respectively includes:

[0032] If the dataset is divided into two subsets, then for each methylation site, the target methylation site is determined based on the degree of difference in the methylation data corresponding to the methylation site in the two subsets; or

[0033] If the data set is divided into two or more subsets, for each methylation site, a target methylation site is determined based on the correlation between the methylation data corresponding to the methylation site in each subset and the cancer stage value corresponding to the subset.

[0034] As an optional implementation manner, the degree of difference between the methylation data corresponding to the methylation sites in the two subsets is determined by:

[0035] Performing a rank sum test on the methylation data corresponding to the methylation sites in the two subsets, and determining the degree of difference according to the rank sum test value; or

[0036] The correlation between the methylation data corresponding to the methylation sites in each subset and the cancer stage value corresponding to the subset is determined by:

[0037] A chi-square test is performed on the methylation data corresponding to the methylation sites in each subset and the cancer stage value corresponding to the subset, and the correlation is determined according to the chi-square test value.

[0038] As an optional implementation, the multiple subsets include more than two subsets; and the dividing the data set into multiple subsets according to the cancer stage values ​​corresponding to the sample set further includes:

[0039] If, among the multiple divided subsets, there is a subset whose number of sample sets is lower than a threshold, then the subset is merged into another subset based on the cancer stage value corresponding to the subset, and the cancer stage value corresponding to the merged another subset is determined, so as to use the merged subsets to determine the target methylation site and train the classification model; wherein the cancer stage value corresponding to the another subset is closest to the cancer stage value corresponding to the subset.

[0040] As an optional implementation, it also includes:

[0041] For the sample set of the target subject, the methylation data corresponding to the methylation sites of the normal tissue adjacent to the cancer in the sample set and / or the methylation data corresponding to the methylation sites containing missing values ​​are removed.

[0042] As an optional implementation, the stage prediction model is integrated according to the following method:

[0043] Using the outputs of n-1 classification models as the input of the nth classification model, and integrating the n classification models into the stage prediction model;

[0044] The input of the n-1 classification models is the data set, and the output of the nth classification model is the cancer stage value corresponding to the data set.

[0045] In a second aspect, an embodiment of the present disclosure further provides an electronic device, including a processor and a memory, wherein the memory is used to store a program executable by the processor, and the processor is used to read the program in the memory and perform the following steps:

[0046] Obtain a methylation dataset of the target object;

[0047] Inputting the methylation dataset into a staging prediction model to output a cancer staging value of the target subject;

[0048] Among them, the stage prediction model is obtained by integrating n classification models, the model parameters of the n classification models are optimized using the grid retrieval method, the n classification models are screened from m classification models using the cross-validation method, and the m classification models are independently trained using the same data set, where m>n, m is an integer greater than 2, and n is an integer greater than 1.

[0049] As an optional implementation, the processor is specifically configured to execute:

[0050] Acquire methylation data corresponding to a target methylation site of a target object, and determine the methylation dataset based on the methylation data corresponding to the target methylation site;

[0051] The target methylation site is determined based on the degree of difference in methylation data of the same methylation site corresponding to different cancer stage values, or the correlation between the methylation data corresponding to the methylation site and different cancer stage values.

[0052] As an optional embodiment, the staging prediction model further includes a standardization model, which is used to standardize the methylation data in the methylation dataset by removing the mean and normalizing the variance.

[0053] As an optional implementation, the staging prediction model further includes a dimensionality reduction model, and the dimensionality reduction model is used to perform dimensionality reduction processing on the methylation data in the methylation dataset according to the PCA method.

[0054] As an optional implementation manner, the processor is specifically configured to determine the staging prediction model in the following manner:

[0055] Using the data set to independently train the m classification models, and determine the classification accuracy of each of the m classification models;

[0056] According to the classification accuracy, k classification models are selected from the m classification models.

[0057] As an optional implementation manner, after selecting k classification models from m classification models according to the classification accuracy, the processor is further configured to execute:

[0058] Using a cross-validation method, the n classification models are selected from the k classification models, where m≥k≥n; and / or,

[0059] The model parameters of the screened classification model are adjusted using the grid retrieval method.

[0060] As an optional implementation, the processor is specifically configured to execute:

[0061] Using the data set, cross-validate the k classification models respectively to obtain evaluation index values ​​of the k classification models;

[0062] According to the evaluation index values ​​of the k classification models, n classification models are screened out from the k classification models.

[0063] As an optional implementation, after obtaining the data set, the processor is further configured to execute:

[0064] The methylation data corresponding to the target methylation site is screened out from the methylation data contained in the data set, and the m classification models are independently trained using the screened methylation data.

[0065] As an optional embodiment, the data set includes sample sets of different target subjects, the sample sets include methylation data corresponding to each methylation site, and one sample set corresponds to one cancer stage value; the processor is specifically configured to determine the target methylation site by:

[0066] Dividing the data set into multiple subsets according to the cancer stage values ​​corresponding to the sample sets, wherein each subset corresponds to one cancer stage value, and the sample sets included in different subsets have different cancer stage values ​​corresponding to them;

[0067] For each methylation site, a target methylation site is determined based on the methylation data corresponding to the methylation site in multiple subsets.

[0068] As an optional implementation, the processor is specifically configured to execute:

[0069] If the dataset is divided into two subsets, then for each methylation site, the target methylation site is determined based on the degree of difference in the methylation data corresponding to the methylation site in the two subsets; or

[0070] If the data set is divided into two or more subsets, for each methylation site, a target methylation site is determined based on the correlation between the methylation data corresponding to the methylation site in each subset and the cancer stage value corresponding to the subset.

[0071] As an optional implementation manner, the processor is specifically configured to determine the degree of difference between the methylation data corresponding to the methylation sites in the two subsets respectively in the following manner:

[0072] Performing a rank sum test on the methylation data corresponding to the methylation sites in the two subsets, and determining the degree of difference according to the rank sum test value; or

[0073] The correlation between the methylation data corresponding to the methylation sites in each subset and the cancer stage value corresponding to the subset is determined by:

[0074] A chi-square test is performed on the methylation data corresponding to the methylation sites in each subset and the cancer stage value corresponding to the subset, and the correlation is determined according to the chi-square test value.

[0075] As an optional implementation, the multiple subsets include more than two subsets; the data set is divided into multiple subsets according to the cancer stage values ​​corresponding to the sample set, and the processor is further configured to execute:

[0076] If, among the multiple divided subsets, there is a subset whose number of sample sets is lower than a threshold, then the subset is merged into another subset based on the cancer stage value corresponding to the subset, and the cancer stage value corresponding to the merged another subset is determined, so as to use the merged subsets to determine the target methylation site and train the classification model; wherein the cancer stage value corresponding to the another subset is closest to the cancer stage value corresponding to the subset.

[0077] As an optional implementation manner, the processor is further configured to execute:

[0078] For the sample set of the target subject, the methylation data corresponding to the methylation sites of the normal tissue adjacent to the cancer in the sample set and / or the methylation data corresponding to the methylation sites containing missing values ​​are removed.

[0079] As an optional implementation, the stage prediction model is integrated according to the following method:

[0080] Using the outputs of n-1 classification models as the input of the nth classification model, and integrating the n classification models into the stage prediction model;

[0081] The input of the n-1 classification models is the data set, and the output of the nth classification model is the cancer stage value corresponding to the data set.

[0082] In a third aspect, an embodiment of the present disclosure further provides a computer storage medium on which a computer program is stored, and when the program is executed by a processor, it is used to implement the steps of the method described in the first aspect above.

[0083] These and other aspects of the present disclosure will become more readily apparent from the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0085] FIG1 is a flowchart of a specific implementation of a cancer staging method provided by an embodiment of the present disclosure;

[0086] FIG2 is a schematic diagram of a data format after preprocessing of a data set provided by an embodiment of the present disclosure;

[0087] FIG3 is a schematic diagram of a differential analysis p-value of methylation data corresponding to each methylation site provided by an embodiment of the present disclosure;

[0088] FIG4 is a distribution diagram of methylation data before normalization provided by an embodiment of the present disclosure;

[0089] FIG5 is a distribution diagram of methylation data after normalization and dimensionality reduction according to an embodiment of the present disclosure;

[0090] 6A to 6E are schematic diagrams of a cross-validation classification model evaluation provided by an embodiment of the present disclosure;

[0091] FIG7 is a flowchart of a multi-model integration provided by an embodiment of the present disclosure;

[0092] FIG8 is a graph showing an ROC_AUC indicator evaluation curve of a staging prediction model provided by an embodiment of the present disclosure;

[0093] FIG9 is a diagram showing the results of cancer staging prediction for liver cancer patients using a staging prediction model provided by an embodiment of the present disclosure;

[0094] FIG10 is a flowchart illustrating a specific implementation of a method for cancer staging provided by an embodiment of the present disclosure;

[0095] FIG11 is a schematic diagram of a cancer staging device provided by an embodiment of the present disclosure;

[0096] FIG12 is a schematic diagram of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0097] To make the objectives, technical solutions, and advantages of the present disclosure more clear, the present disclosure will be further described in detail below with reference to the accompanying drawings. It is apparent that the embodiments described are only a portion of the embodiments of the present disclosure, rather than all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present disclosure without creative effort are intended to fall within the scope of protection of the present disclosure.

[0098] In the embodiments of the present disclosure, the term "and / or" describes the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates that the associated objects are in an "or" relationship.

[0099] The term "methylation" in the embodiments of the present disclosure is catalyzed by enzymes in biological systems. Methylation involves heavy metal modification, regulation of gene expression, regulation of protein function, and ribonucleic acid (RNA) processing. Gene methylation is the most important form of epigenetic inheritance and can affect the occurrence and development of tumors.

[0100] The term "cancer staging" in the embodiments of the present disclosure is a method for determining the extent of cancer development and spread. Cancer staging affects the patient's treatment plan and the judgment of the cancer prognosis.

[0101] In the embodiments of the present disclosure, the term “PCA (Principal Component Analysis)” is used. PCA is often used for dimensionality reduction of high-dimensional data and can be used to extract the main features of the data.

[0102] The term "grid search" in the embodiments of the present disclosure is a model hyperparameter optimization technology, which is often used to optimize three or fewer hyperparameters and is essentially an exhaustive method.

[0103] In the embodiments of the present disclosure, the term "model integration" refers to combining multiple weakly supervised models in order to obtain a better and more comprehensive strongly supervised model. The underlying idea of ​​the integrated model is that even if a weak classifier makes an incorrect prediction, other weak classifiers can correct the error.

[0104] The application scenarios described in the embodiments of the present disclosure are intended to more clearly illustrate the technical solutions of the embodiments of the present disclosure and do not constitute a limitation on the technical solutions provided by the embodiments of the present disclosure. Persons skilled in the art will appreciate that, as new application scenarios emerge, the technical solutions provided by the embodiments of the present disclosure will also be applicable to similar technical problems. In the description of the present disclosure, unless otherwise specified, "multiple" means two or more.

[0105] Cancer staging is a method for determining the extent of cancer development and spread. Accurate cancer staging facilitates the development of optimal treatment plans for cancer patients and effectively determines their prognosis. Among hematologic malignancies, lymphoma and Hodgkin's lymphoma both use the Ann Arbor staging system. For solid tumors, the TNM (Tumor Node Metastasis) staging system is the most common, but there are many differences depending on the tumor type. Commonly used tumor types include breast cancer, colon cancer, kidney cancer, laryngeal cancer, liver cancer, lung cancer, prostate cancer, skin cancer, and bladder cancer, which are classified into stages I, II, III, and IV based on the severity of the tumor. While existing cancer staging methods often rely on clinical and pathological examinations, the drawback is that when performed solely through clinical examinations (blood tests, radiological examinations, endoscopy, etc.), the information provided for cancer staging is limited, which is a limitation. Although pathological examination can more accurately determine the stage, it requires surgery to obtain pathological sections. Not all tumors require surgical treatment, and some patients may have undergone chemotherapy and radiotherapy before the pathological examination, which may underestimate the true stage of the tumor to a certain extent. Therefore, the method of using pathological examination still has limitations. Secondly, different cancers have different staging systems, and some cancers do not have appropriate staging methods. The staging is highly subjective, and conventional staging methods are not universal. In addition to clinical systems, there are currently methods for cancer staging based on methylation data, but these methods usually only consider a few methylation sites, and the classification model method is relatively simple, failing to effectively utilize most of the methylation data of the genome.

[0106] The cancer staging method provided in this embodiment is used to predict the cancer stage corresponding to an input methylation dataset using a staging prediction model obtained by integrating multiple classification models. The staging prediction model in this embodiment is independently trained using m classification models, and n classification models are screened using a cross-validation method. The model parameters of these n classification models are optimized using a grid search method. This method can effectively capture the complex relationship between methylation data and cancer stage, thereby improving the accuracy of cancer stage prediction. Compared with current clinical staging methods, cancer stage prediction using methylation site information is applicable to the vast majority of cancer types and can effectively distinguish the cancer stage types of different patients, thereby improving the universality of cancer stage prediction.

[0107] As shown in FIG1 , the specific implementation process of a cancer staging method provided in this embodiment is as follows:

[0108] Step 100: Obtain a methylation dataset of a target object;

[0109] The target object in this embodiment includes but is not limited to human tumor tissue.

[0110] In practice, the methylation dataset in this embodiment includes methylation data corresponding to multiple methylation sites, wherein one methylation site in one sample corresponds to one methylation data. The methylation data in this embodiment includes methylation level values.

[0111] In some embodiments, the methylation dataset of the target subject is determined by:

[0112] Acquire methylation data corresponding to a target methylation site of a target object, and determine the methylation dataset based on the methylation data corresponding to the target methylation site;

[0113] The target methylation site is determined based on the degree of difference in methylation data of the same methylation site corresponding to different cancer stage values, or the correlation between the methylation data corresponding to the methylation site and different cancer stage values.

[0114] During implementation, the methylation dataset of the target object in this embodiment includes methylation data corresponding to the target methylation site, and the target methylation site is screened out from each methylation site. Since the methylation data corresponding to the target methylation site can more accurately characterize the characteristics of cancer staging, the methylation dataset formed using the methylation data corresponding to the target methylation site can more accurately predict the cancer staging and improve the accuracy of the prediction.

[0115] It should be noted that the number of target methylation sites is less than or equal to the number of each methylation site contained in the target object itself.

[0116] Step 101: input the methylation dataset into a staging prediction model to output the cancer staging value of the target subject;

[0117] Among them, the stage prediction model is obtained by integrating n classification models, the model parameters of the n classification models are optimized using the grid retrieval method, the n classification models are screened from m classification models using the cross-validation method, and the m classification models are independently trained using the same data set, where m>n, m is an integer greater than 2, and n is an integer greater than 1.

[0118] In practice, the staging prediction model in this embodiment is used to predict the cancer staging of the input methylation data set. On the one hand, since the staging prediction model in this embodiment is integrated by multiple classification models, and model integration refers to combining multiple weakly supervised models in order to obtain a better and more comprehensive strong supervised model, the underlying idea of ​​the integrated model is that even if a certain weak classifier makes an incorrect prediction, the other weak classifiers can correct the error. Therefore, the accuracy of the staging prediction model prediction is improved by model integration. On the other hand, the n classification models included in the staging prediction model in this embodiment are obtained by screening from m classification models using a cross-validation method, and cross-validation can evaluate whether the m classification models perform well on the entire data set, and is used to evaluate the stability and generalization ability of the classification model. For example, when one of the m classification models fluctuates greatly on the AUC indicator of all folds of cross-validation, the classification model will be excluded, that is, it will not be selected as one of the n classification models. Finally, the model parameters of the n classification models finally selected in this embodiment are optimized using the grid retrieval method, and network retrieval is a model hyperparameter optimization technology that is often used to optimize three or fewer hyperparameters. It is essentially an exhaustive method. The use of grid retrieval to adjust the model parameters of the n classification models is to select the model parameters with the best classification model performance based on the n classification models that have been selected. This embodiment optimizes the accuracy, stability and predictive ability of the stage prediction model for new data sets through the coordinated execution of independent training, cross-validation and grid retrieval parameter adjustment of multiple classification models.

[0119] In some embodiments, the staging prediction model further includes a standardization model, which is used to standardize the methylation data in the methylation dataset by removing the mean and normalizing the variance.

[0120] In practice, since the distribution of methylation data for multiple target methylation sites is skewed, in order to unify the methylation rates of the methylation data for multiple target methylation sites, the methylation data for each target methylation site can be normalized so that the methylation data for each target methylation site is as close to a normal distribution as possible. This allows for more accurate predictions based on the methylation data for each target methylation site.

[0121] In some embodiments, the staging prediction model further includes a dimensionality reduction model, which is used to perform dimensionality reduction processing on the methylation data in the methylation dataset according to a PCA method.

[0122] In practice, PCA transforms raw data into a set of linearly independent representations across all dimensions through a linear transformation. This can be used to extract the data's primary characteristic components and is commonly used for dimensionality reduction in high-dimensional data. PCA extracts the primary features of the methylation dataset formed by each target methylation data point, thereby generating a methylation dataset containing these key features. This reduces the amount of data required for subsequent predictions based on this methylation dataset, improving the algorithm's prediction speed and efficiency.

[0123] Optionally, the stage prediction model in this embodiment includes, but is not limited to, normalizing and reducing the methylation dataset, and performing prediction using the normalized and reduced methylation dataset. This embodiment does not impose any restrictions on the order of normalization and dimensionality reduction.

[0124] In some embodiments, this embodiment determines the staging prediction model in the following manner:

[0125] Step 1: Using the data set, the m classification models are independently trained to determine the classification accuracy of each of the m classification models;

[0126] In implementation, the data set includes sample sets of different target objects, and the sample sets include methylation data corresponding to each methylation site; one sample set corresponds to one cancer stage value; one target object corresponds to one sample set, and one sample set includes methylation data corresponding to multiple methylation sites.

[0127] In implementation, during the training of the classification model, the training set in the data set is used for training, and the training set includes a sample set.

[0128] Optionally, after obtaining the dataset and before using the dataset to train the classification model, the methylation data in the dataset may be preprocessed. The preprocessing method includes but is not limited to any one or a combination of the following methods:

[0129] Method 1a: After obtaining the sample set, the methylation data corresponding to the methylation sites of the normal tissue adjacent to the cancer in the sample set may be removed for the sample set of the target subject;

[0130] Mode 1b: After obtaining the sample set, the methylation data corresponding to the methylation sites containing missing values ​​(NA) in the sample set may be removed for the sample set of the target object.

[0131] During implementation, a sample set can be obtained from the public database TCGA, where each sample set corresponds to a cancer stage value, and the methylation data corresponding to the methylation sites of normal tissue adjacent to the cancer and the methylation data corresponding to the methylation sites of tumor tissue containing missing values ​​are removed.

[0132] In some embodiments, after obtaining the dataset, the present embodiment may further screen the training set in the dataset to screen out methylation data corresponding to target methylation sites that characterize cancer staging characteristics, thereby improving prediction accuracy. The specific method is as follows:

[0133] Method 1c: Filtering methylation data corresponding to the target methylation site from the methylation data included in the training set of the dataset, and independently training the m classification models using the filtered methylation data. The filtered methylation data refers to the methylation data corresponding to the filtered target methylation site.

[0134] During implementation, after obtaining a sample set and removing the methylation data corresponding to the methylation sites of normal tissue adjacent to the cancer and the methylation data corresponding to the methylation sites containing missing values, the methylation data corresponding to the target methylation sites can be screened out from the removed sample set, thereby forming a data set using the screened sample set.

[0135] In some embodiments, the target methylation site is determined based on the training set in the dataset, and the target methylation site can be determined by the following process:

[0136] Process c1) dividing the data set into multiple subsets according to the cancer stage values ​​corresponding to the sample sets, wherein each subset corresponds to one cancer stage value, and the sample sets included in different subsets correspond to different cancer stage values;

[0137] During implementation, the data set can be divided into two or more subsets. For example, one subset corresponds to one cancer stage value. The sample set of stage I is divided into one subset, the sample set of stage II is divided into one subset, the sample set of stage III is divided into one subset, and the sample set of stage IV is divided into one subset, so that the data set is finally divided into four subsets. The sample sets of stages I and II can also be divided into one subset, and the sample sets of stages III and IV can be divided into one subset, that is, the data set is divided into two subsets, and the cancer stage value corresponding to a subset containing the sample sets of stages I and II is set to the early stage, and the cancer stage value corresponding to a subset containing the sample sets of stages III and IV is set to the late stage. This embodiment does not impose too many restrictions on the division of the data set, and it can be adjusted according to actual conditions.

[0138] Optionally, after the dataset is divided into multiple subsets according to the cancer stage values ​​corresponding to the sample set, if the multiple subsets include more than two subsets, the subsets may be merged. The specific steps are as follows:

[0139] If, among the multiple divided subsets, there is a subset whose number of sample sets is lower than a threshold, then the subset is merged into another subset based on the cancer stage value corresponding to the subset, and the cancer stage value corresponding to the merged another subset is determined, so as to use the merged subsets to determine the target methylation site and train the classification model; wherein the cancer stage value corresponding to the another subset is closest to the cancer stage value corresponding to the subset.

[0140] In practice, when there are more than two subsets, if there is an imbalance in the amount of methylation data between different subsets, the different subsets can be merged. For example, a dataset includes sample sets of different liver cancer patients and their corresponding cancer stage values. The number of stage I cases (the number of sample sets) is 212, the number of stage II cases is 107, and the total number of stages III and IV is 150. Methylation sites with missing methylation data are removed, and methylation data from normal samples are removed. The sample sets are then grouped according to the cancer stage value, i.e., the dataset is split. Because the number of stage III and stage IV cases is dispersed, and the number of cases in some stages is relatively small, there is an imbalance in the sample sets. Therefore, the stage III and stage IV cases are merged into the late stage cancer, and the stage I and stage II cases are merged into the early stage cancer. At this point, the early stage cancer contains 265 cases, and the late stage cancer contains 115 cases.

[0141] Optionally, after dividing the dataset into multiple subsets, the format of the dataset can be adjusted to facilitate execution of the next process. For example, the dataset can be transposed so that the first column contains cancer stage values, and the second and subsequent columns contain methylation data corresponding to the methylation sites. As shown in Figure 2, this embodiment provides a data format after dataset preprocessing.

[0142] After preprocessing the dataset, the preprocessed dataset is split into a training set and a test set in a ratio of 8:2. In the specific implementation, the dataset is first divided into multiple subsets, and then the dataset is split into a training set and a test set. The training set after the split includes the multiple subsets, and the test set also includes the multiple subsets.

[0143] Process c2) For each methylation site, a target methylation site is determined based on the methylation data corresponding to the methylation site in multiple subsets.

[0144] Based on the number of subsets into which the training set is divided, this embodiment provides different methods for determining target methylation sites, as shown below:

[0145] Method c1: The training set includes two subsets, and the target methylation sites are determined by the following method:

[0146] For each methylation site, the target methylation site is determined according to the degree of difference in the methylation data corresponding to the methylation site in the two subsets.

[0147] In implementation, the degree of difference of the methylation data corresponding to the methylation sites in the two subsets is determined by:

[0148] A rank sum test is performed on the methylation data corresponding to the methylation sites in the two subsets, and the degree of difference is determined according to the rank sum test value.

[0149] In the implementation, according to the cancer stage, the training set includes subset 1 corresponding to the early stage of cancer and subset 2 corresponding to the late stage. The methylation data corresponding to the same methylation site in the two subsets are subjected to a rank sum test to obtain the rank sum test value p. The formula is as follows: P n =w(G1,G2) Formula 1;

[0150] In formula 1, P n is the rank sum test value (significance test value) obtained after the rank sum test for the nth methylation site, w is the rank sum test function, G1 represents the methylation data of the nth methylation site in subset 1, and G2 represents the methylation data of the nth methylation site in subset 2.

[0151] Then, the researchers sorted the methylation sites by p-value from smallest to largest, selecting the top 1,000 sites with the highest degree of difference (i.e., the smallest p-value) as target methylation sites. In theory, the methylation data corresponding to these 1,000 sites can more accurately distinguish target subjects at different cancer stages.

[0152] For example, the pre-processed data set is classified according to early cancer (identified as 0) and late cancer (identified as 1) to obtain two subsets, one subset corresponding to early cancer and the other subset corresponding to late cancer. The methylation data on each methylation site are differentially analyzed using a rank sum test to calculate the p value. The p value is sorted from small to large, and the first 1000 methylation sites are selected as target methylation sites. As shown in Figure 3, a schematic diagram of a differential analysis p value of the methylation data corresponding to each methylation site provided in the present embodiment is shown, wherein the leftmost side represents the index value of the methylation site, the column corresponding to cg represents the identifier of the methylation site, and pvalue represents the rank sum test p value calculated for the methylation data corresponding to the methylation site.

[0153] When analyzing methylation data, the rank sum test can be used to detect differentially methylated sites. Since the number of sample sets under different cancer stages is usually different and the methylation rate distribution has no obvious pattern, the rank sum test can be used to perform significance tests under conditions where the data does not conform to the normal distribution and the number of sample sets is different. The methylation data corresponding to the top 1000 most significant target methylation sites are obtained.

[0154] Method c2: The training set includes two or more subsets, and the target methylation sites are determined by the following method:

[0155] For each methylation site, a target methylation site is determined based on the correlation between the methylation data corresponding to the methylation site in each subset and the cancer stage value corresponding to the subset.

[0156] In implementation, the correlation between the methylation data corresponding to the methylation sites in each subset and the cancer stage value corresponding to the subset is determined in the following manner:

[0157] A chi-square test is performed on the methylation data corresponding to the methylation sites in each subset and the cancer stage value corresponding to the subset, and the correlation is determined according to the chi-square test value.

[0158] The chi-square test measures the degree of deviation between the actual observed values ​​and the theoretical inference values ​​of the statistical sample. The degree of deviation between the actual observed values ​​and the theoretical inference values ​​determines the size of the chi-square value.

[0159] In practice, if the training set includes three or more subsets, each subset corresponds to a cancer stage. The chi-square test can be used to calculate the correlation between each methylation site and the cancer stage value, that is, the chi-square statistic, and the 1000 methylation sites with the highest chi-square statistic are selected as target methylation sites. The formula for calculating the chi-square test is as follows: S n =chi2(X n ,y) Formula 2;

[0160] In formula 2, S n represents the chi-square test value obtained after the chi-square test of the nth methylation site, chi2 represents the chi-square test function, X n represents the methylation data corresponding to the nth methylation site in each subset, and y represents the cancer stage value corresponding to each subset.

[0161] When the training set is divided into three subsets, the values ​​of y are 0, 1, and 2; when the training set is divided into four subsets, the values ​​of y are 0, 1, 2, and 3.

[0162] Optionally, after the target methylation sites are determined as described above, the training set can be pre-processed by standardization and dimensionality reduction. For example, after 1000 target methylation sites are determined, the methylation data corresponding to the 1000 methylation sites are normalized. The standardization process uses mean removal and variance normalization, and the formula is as follows:

[0163] In formula 3, Represents the methylation data corresponding to the target methylation site after normalization, M n represents the methylation data corresponding to the target methylation site, μ represents the mean of the methylation data corresponding to the target methylation site in each subset, and σ represents the standard deviation of the methylation data corresponding to the target methylation site in each subset.

[0164] For example, statistics are collected for the methylation data corresponding to 1,000 target methylation sites, and a distribution diagram is obtained as shown in FIG4 . This embodiment provides a distribution diagram of methylation data before normalization. The distribution of methylation data corresponding to each target methylation site in the diagram is skewed, which is not conducive to model training. Therefore, the skewed data is standardized using the mean removal and variance normalization methods to obtain methylation data that is as close to a normal distribution as possible for training the classification model.

[0165] Optionally, after the methylation data corresponding to the 1000 methylation sites are normalized, the PCA dimensionality reduction method can be used to reduce the methylation data corresponding to the 1000 methylation sites to 10 feature data. The formula is as follows: X * =PCA(n).fit_transform(X) Formula 4;

[0166] In formula 4, X * represents the methylation data after dimensionality reduction, X * It can be understood as a matrix. PCA(n) represents the construction of a PCA function with a dimensionality reduction result of n. n represents the dimension of the methylation data after dimensionality reduction, such as n = 10. The function fits and transforms X, where X represents the methylation data before dimensionality reduction.

[0167] During implementation, since too many features will affect the model's search for inherent rules, and the methylation data features in the training set in this embodiment are extremely complex, the PCA dimensionality reduction method can be used to set the dimension to 10, and reduce the high-dimensional 1000 feature vectors (methylation data corresponding to 1000 methylation sites) to low-dimensional 10 feature vectors (methylation data corresponding to 10 methylation sites). While accelerating the training speed of the classification model, the characteristics of the metadata can also be guaranteed. As shown in Figure 5, this embodiment provides a distribution diagram of the methylation data after standardization and dimensionality reduction processing. The distribution of the processed methylation data is close to the normal distribution, which is more conducive to the training of the classification model.

[0168] Optionally, to adapt to specific data set analysis requirements, the order of the rank sum test / chi-square test, normalization, and PCA dimensionality reduction can be freely swapped. When processing larger data sets, such as whole-genome methylation sequencing data, since the number of methylation sites in the genome is generally in the tens of thousands, PCA dimensionality reduction can be prioritized to reduce the dimensionality of the data and reduce the time required for the rank sum test / chi-square test and normalization. The time saved is more significant when the number of methylation sites is greater. In addition, if the data being processed is targeted methylation sequencing data containing only dozens of genes, normalization can be performed first, followed by the rank sum test / chi-square test and PCA dimensionality reduction. After normalization, the data distribution is more even. When performing significance testing, differentially methylated sites can be more easily detected because they are not affected by the numerical size of the data itself, improving the sensitivity of the rank sum test / chi-square test. Similarly, the effect of PCA dimensionality reduction can also be improved.

[0169] After the training set is processed by the above process, the processed training set is input into the traversal fitting of the classification model algorithm, the training set is traversed and input into m classification models, and the model performance of each classification model is calculated. Optionally, the m classification models in this embodiment include but are not limited to logistic regression (LogisticRegression), multi-layer perceptron (MLPClassifier), K nearest neighbor (KNeighborsClassifier), support vector machine (SVC), Gaussian process (GaussianProcessClassifier), decision tree (DecisionTreeClassifier), Gaussian naive Bayes (GaussianNB), random forest (RandomForestClassifier), discriminant analysis (DiscriminantAnalysis), multi-layer perceptron (MLPClassifier) ​​and other models. The classification accuracy of each classification model is calculated as an indicator for judging the performance of the model.

[0170] Step 2: According to the classification accuracy, k classification models are selected from the m classification models.

[0171] During implementation, the top k classification models are selected from the m classification models in descending order of classification accuracy for subsequent cross-validation and / or grid retrieval (adjustment of model parameters).

[0172] Optionally, after selecting k classification models from the m classification models according to the classification accuracy, any one or more of the following steps may be performed:

[0173] a) using a cross-validation method to select the n classification models from the k classification models, where m≥k≥n;

[0174] b) Using the grid retrieval method, the model parameters of the screened classification model are adjusted.

[0175] The above cross-validation and grid search can be performed simultaneously, but the order of execution can be customized according to actual needs. When the data set is large, you can first perform cross-validation to screen out n classification models, and then use the grid search method to adjust the model parameters of the n classification models screened out. When the data set is small, you can first use the grid search method to adjust the model parameters of the k classification models screened out, and then use the cross-validation method to screen out the n classification models from the k classification models.

[0176] In implementation, the cross-validation method is used to screen the n classification models from the k classification models. The cross-validation formula is as follows: k =T(X train ,y train ) Formula 5a; Score k =C k (X test ,y test ) Formula 5b;

[0177] Among them, X train ,y train Respectively represent the methylation data in the training set and the cancer stage value corresponding to the training set; X test ,y test Represent the methylation data in the test set and the cancer stage value corresponding to the test set respectively; the training set and the test set are the data set segmentation at the kth fold of the data set, T represents the multiple classification models mentioned in this embodiment, C k Indicates the classification model generated by fitting under the k-th fold, Score k C is the k-th fold classification model k According to X test ,ytest Calculate the model AUC (evaluation index value).

[0178] Optionally, perform cross-validation and screening as follows:

[0179] Step 3a) using the data set, cross-validating the k classification models to obtain evaluation index values ​​for the k classification models;

[0180] In implementation, the model accuracy of the classification model can be used as an evaluation index value.

[0181] Step 3b) n classification models are selected from the k classification models according to the evaluation index values ​​of the k classification models.

[0182] In implementation, the above Score k The AUC averages are used to rank the classification models from high to low, and the top n classification models are selected.

[0183] For example, to prevent overfitting, a 5-fold cross-validation with K=5 is performed on the data set, and ROC_AUC is selected as the verification evaluation method. A plot is performed, and the top 5 classification models with better cross-validation are selected, including logistic regression, linear discriminant analysis, extreme random trees, Gaussian processes, and K-nearest neighbor classification models with better fitting methods, as shown in Figures 6A to 6E. This embodiment provides a schematic diagram of a cross-validated classification model evaluation, wherein Figures 6A to 6E respectively represent ROC_AUC graphs of logistic regression, linear discriminant analysis, extreme random trees, Gaussian processes, and K-nearest neighbor classification models.

[0184] The following parameter optimization process is also performed:

[0185] In some embodiments, the grid retrieval method is used to adjust the model parameters of the screened classification model to obtain the adjusted classification model. Taking cross-validation followed by grid retrieval as an example, after the n classification models are screened out from the k classification models using the cross-validation method, the grid retrieval method is used to adjust the model parameters of the classification models in the n classification models, and the final staging prediction model is determined based on the adjusted classification model.

[0186] During implementation, after selecting n classification models with good cross-validation, grid retrieval and parameter adjustment are performed on the model parameters of the n classification models. The model parameters involved in the n classification models are arranged, combined, and configured, and the optimal combination of model parameter configuration is obtained based on the calculation results of the model accuracy.

[0187] For example, we used a grid search method to adjust the parameters of the aforementioned classification models: logistic regression, linear discriminant analysis, extreme random trees, and Gaussian processes, and evaluated them using the ROC_AUC metric. The four classification models (logistic regression, linear discriminant analysis, Gaussian processes, and extreme random trees) that had been adjusted were then integrated.

[0188] In some embodiments, the staging prediction model is integrated according to the following method:

[0189] Using the outputs of n-1 classification models as the input of the nth classification model, and integrating the n classification models into the stage prediction model;

[0190] The input of the n-1 classification models is the training set, and the output of the nth classification model is the cancer stage value corresponding to the training set.

[0191] During implementation, the n classification models with adjusted parameters are integrated to generate the final stage prediction model. For example, when n = 4, the features output by the three classification models are used as the input data of the fourth classification model for fitting to obtain the final stage prediction model. As shown in Figure 7, this embodiment provides a flowchart of multi-model integration, in which the methylation data in the training set are input into multiple classification models for fitting training, the multiple classification models are evaluated according to the model evaluation index, and n classification models are selected for model integration. The model in the figure refers to the classification model.

[0192] The formula for model integration is as follows: C combine =C4(S C1 ,S C2 ,S C3 ) Formula 6;

[0193] In formula 6, C combine represents the integrated stage prediction model, S C1 ,S C2 ,S C3 Represent the predicted probability scores generated by classification model C1, classification model C2 and classification model C3 respectively. C1 ,S C2 ,S C3 The output data of is input into the fourth classification model C4 for model construction to obtain the final integrated model.

[0194] For example, four classification models with tuned parameters (logistic regression, linear discriminant analysis, Gaussian process, and extreme random tree) were selected and the stacking method in mlxtend was used for model integration. The output data of the first three classification models were used as the input data of the last classification model for fitting. The ROC_AUC indicator was used for evaluation to generate a staging prediction model. The ROC_AUC indicator evaluation curve of the staging prediction model is shown in Figure 8.

[0195] After the staging prediction model is integrated using the above method, in the use stage, the methylation data of patients with unknown liver cancer staging are first obtained, and then 1000 target methylation sites corresponding to the staging prediction model are selected. Finally, the methylation data corresponding to the 1000 target methylation sites are standardized and dimensionally reduced, and the methylation data after standardization and dimension reduction are input into the staging prediction model for staging prediction, and the corresponding prediction results are output. As shown in Figure 9, this embodiment provides a result graph of cancer staging prediction for liver cancer patients using the staging prediction model. The graph includes the predicted staging of each liver cancer patient and indicates the specific staging probability.

[0196] This embodiment uses the traversal fitting classification algorithm (using the training set to independently train m classification models), cross-validation, and grid retrieval parameter adjustment in a coordinated manner. Among them, traversal fitting can preliminarily select a candidate for a better classification algorithm from all classification algorithms. Cross-validation can evaluate whether the aforementioned candidate classification algorithm performs well on the entire data set, which is used to evaluate the stability and generalization ability of the classification model. If the candidate classification model selected by traversal fitting fluctuates greatly in the AUC index of all cross-validation folds, then the classification algorithm used by the classification model will be excluded. Model parameter adjustment is to select the parameters with the best model performance based on the n selected classification models. After the traversal fitting classification algorithm, cross-validation, and grid retrieval parameter adjustment, the classification model has the best accuracy, stability, and predictive ability for new data sets. Model integration is used to further integrate the aforementioned n better models to obtain a final stage prediction model. The performance of the final stage prediction model is higher than that of any of the n models.

[0197] Alternatively, if the data volume is small, such as when processing targeted methylation sequencing data containing only dozens of genes, a model can be built by traversing and fitting the classification model, performing grid retrieval, and then cross-validating. This operation can evaluate all classification algorithms and their associated parameters. In theory, it considers the most sufficient conditions and the most comprehensive considerations when screening models. However, because the computational complexity of considering all classification algorithms and their associated parameters increases exponentially, it consumes the longest time. If the data volume is large, such as when processing whole-genome methylation sequencing data, a model can be built by first traversing and fitting the classification model, then cross-validating, and finally grid retrieval. Traversing and fitting the classification model can preliminarily select the best candidate classification algorithm from all classification algorithms. Cross-validation is used to evaluate the stability and generalization ability of the classification model. Grid retrieval (i.e., model parameter adjustment) is to select the parameters that achieve the best model performance based on the n selected classification models. After traversing and fitting the classification algorithm, cross-validation, and grid retrieval parameter adjustment, the classification model has the best accuracy, stability, and predictive ability for new data sets.

[0198] The idea of ​​model integration in the present embodiment is: the output data of the first three classification models are used as the input data of the last classification model for fitting, and the final stage prediction model is obtained. Specifically, the methylation data after standardization and dimensionality reduction are input into the three classification models respectively, and the prediction results of the prediction probability of each class of the three classification models output are used as the original feature data, and are input into the fourth classification model again for modeling. The method of modeling using the features generated by the first three classifiers as the input data of the last classifier, due to the use of the specific numerical values ​​of the probability of each classifier under different classifications, and the probability numerical values ​​are used as the original data set for model construction evaluation, compared with the method of voting weighted integration, it is more perfect to integrate different classification models, and the weight of each classifier in the prediction result can be objectively considered, which theoretically improves the accuracy of the final model. The method of voting weighted integration only makes the final judgment from the statistics of the classification results of different classifiers. When the weights of different classifiers are not considered, the influence of the poorly performing classifier on the final result is aggravated, and some noise is introduced. When the weights of different classifiers are considered small, the setting of their weights is more subjective.

[0199] As shown in FIG10 , the specific implementation process of a cancer staging method provided in this embodiment is as follows:

[0200] Step 1000: Obtain a data set, remove the methylation data corresponding to the methylation sites of normal tissue adjacent to the cancer, and the methylation data corresponding to the methylation sites containing missing values;

[0201] Step 1001: Divide the data set into multiple subsets according to the cancer stage value, each subset corresponding to a cancer stage value;

[0202] Step 1002: for each methylation site, determine the target methylation site based on the methylation data corresponding to the methylation site in multiple subsets of the training set;

[0203] Step 1003: performing normalization and dimensionality reduction processing on the methylation data corresponding to the target methylation site in the training set to obtain a processed training set;

[0204] Step 1004: Input the processed training set into m classification models respectively, train the m classification models independently, and select k classification models according to the classification accuracy of each of the m classification models;

[0205] Step 1005: Using a cross-validation method, select the n classification models from the k classification models, and using a grid search method, adjust the model parameters of the classification models in the n classification models to obtain the adjusted n classification models;

[0206] Step 1006: using the outputs of n-1 classification models as the inputs of the nth classification model, and integrating the n classification models into the stage prediction model;

[0207] Step 1007: Acquire a methylation dataset of the target object, input the methylation dataset into a staging prediction model, and output a cancer staging value of the target object.

[0208] The cancer staging method provided in this embodiment obtains methylation data and corresponding cancer stage values ​​from cancer patients to form a dataset, preprocesses the dataset and partitions it into subsets. The dataset is then split into a training set and a test set in an 8:2 ratio. The training set is subjected to differential methylation analysis, and multiple target methylation sites with the highest degree of differential methylation are selected. The methylation data corresponding to these target methylation sites are then dimensionality reduced using the PCA method. Multiple classification algorithms are then traversed and fitted to the training set. The fitted classification models are then tested on the test set. N classification models with good test results are selected for grid retrieval and parameter adjustment. The adjusted classification models are then integrated to obtain a final staging prediction model. The final staging prediction model is then used to predict the patient's cancer stage. This embodiment uses methylation site information to predict cancer stage. Compared to current clinical staging methods, this method is applicable to most cancer types and can effectively distinguish cancer stage types among different patients. Combining differential methylation site screening, feature dimensionality reduction, and multi-model integration, the classification algorithm effectively captures the complex relationship between methylation data and cancer stage, improving the accuracy of cancer stage prediction.

[0209] Based on the same inventive concept, the embodiment of the present disclosure also provides a cancer staging device. Since the device is the device in the method in the embodiment of the present disclosure, and the principle of solving the problem by the device is similar to that of the method, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0210] As shown in FIG11 , the device includes:

[0211] A data acquisition module 1100 is used to acquire a methylation dataset of a target object;

[0212] The staging prediction module 1101 is used to input the methylation dataset into a staging prediction model and output the cancer staging value of the target object; wherein the staging prediction model is obtained by integrating n classification models, the model parameters of the n classification models are optimized using a grid retrieval method, the n classification models are screened from m classification models using a cross-validation method, and the m classification models are independently trained using the same dataset, wherein m>n, m is an integer greater than 2, and n is an integer greater than 1.

[0213] As an optional implementation, the data acquisition module 1100 is specifically configured to:

[0214] Acquire methylation data corresponding to a target methylation site of a target object, and determine the methylation dataset based on the methylation data corresponding to the target methylation site;

[0215] The target methylation site is determined based on the degree of difference in methylation data of the same methylation site corresponding to different cancer stage values, or the correlation between the methylation data corresponding to the methylation site and different cancer stage values.

[0216] As an optional embodiment, the staging prediction model further includes a standardization model, which is used to standardize the methylation data in the methylation dataset by removing the mean and normalizing the variance.

[0217] As an optional implementation, the staging prediction model further includes a dimensionality reduction model, and the dimensionality reduction model is used to perform dimensionality reduction processing on the methylation data in the methylation dataset according to the PCA method.

[0218] As an optional implementation, the stage prediction module 1101 is specifically configured to determine the stage prediction model in the following manner:

[0219] Using the data set to independently train the m classification models, and determine the classification accuracy of each of the m classification models;

[0220] According to the classification accuracy, k classification models are selected from the m classification models.

[0221] As an optional implementation, after selecting k classification models from m classification models according to the classification accuracy, the stage prediction module 1101 is further configured to:

[0222] Using a cross-validation method, the n classification models are selected from the k classification models, where m≥k≥n; and / or,

[0223] The model parameters of the screened classification model are adjusted using the grid retrieval method.

[0224] As an optional implementation, the stage prediction module 1101 is specifically configured to:

[0225] Using the data set, cross-validate the k classification models respectively to obtain evaluation index values ​​of the k classification models;

[0226] According to the evaluation index values ​​of the k classification models, n classification models are screened out from the k classification models.

[0227] As an optional implementation, after obtaining the data set, the data obtaining module 1100 is further configured to:

[0228] The methylation data corresponding to the target methylation site is screened out from the methylation data contained in the data set, and the m classification models are independently trained using the screened methylation data.

[0229] As an optional embodiment, the data set includes sample sets of different target subjects, each sample set includes methylation data corresponding to each methylation site, and each sample set corresponds to a cancer stage value; the data acquisition module 1100 is specifically configured to determine the target methylation site by:

[0230] Dividing the data set into multiple subsets according to the cancer stage values ​​corresponding to the sample sets, wherein each subset corresponds to one cancer stage value, and the sample sets included in different subsets have different cancer stage values ​​corresponding to them;

[0231] For each methylation site, a target methylation site is determined based on the methylation data corresponding to the methylation site in multiple subsets.

[0232] As an optional implementation, the data acquisition module 1100 is specifically configured to:

[0233] If the dataset is divided into two subsets, then for each methylation site, the target methylation site is determined based on the degree of difference in the methylation data corresponding to the methylation site in the two subsets; or

[0234] If the data set is divided into two or more subsets, for each methylation site, a target methylation site is determined based on the correlation between the methylation data corresponding to the methylation site in each subset and the cancer stage value corresponding to the subset.

[0235] As an optional implementation, the data acquisition module 1100 is specifically configured to determine the degree of difference between the methylation data corresponding to the methylation sites in the two subsets in the following manner:

[0236] Performing a rank sum test on the methylation data corresponding to the methylation sites in the two subsets, and determining the degree of difference according to the rank sum test value; or

[0237] The correlation between the methylation data corresponding to the methylation sites in each subset and the cancer stage value corresponding to the subset is determined by:

[0238] A chi-square test is performed on the methylation data corresponding to the methylation sites in each subset and the cancer stage value corresponding to the subset, and the correlation is determined according to the chi-square test value.

[0239] As an optional implementation manner, the multiple subsets include more than two subsets; and the data acquisition module 1100 is further configured to:

[0240] If, among the multiple divided subsets, there is a subset whose number of sample sets is lower than a threshold, then the subset is merged into another subset based on the cancer stage value corresponding to the subset, and the cancer stage value corresponding to the merged another subset is determined, so as to use the merged subsets to determine the target methylation site and train the classification model; wherein the cancer stage value corresponding to the another subset is closest to the cancer stage value corresponding to the subset.

[0241] As an optional implementation, the data acquisition module 1100 is further configured to:

[0242] For the sample set of the target subject, the methylation data corresponding to the methylation sites of the normal tissue adjacent to the cancer in the sample set and / or the methylation data corresponding to the methylation sites containing missing values ​​are removed.

[0243] As an optional implementation, the stage prediction model is integrated according to the following method:

[0244] Using the outputs of n-1 classification models as the input of the nth classification model, and integrating the n classification models into the stage prediction model;

[0245] The input of the n-1 classification models is the data set, and the output of the nth classification model is the cancer stage value corresponding to the data set.

[0246] Based on the same inventive concept, the embodiment of the present disclosure also provides an electronic device. Since the electronic device is the electronic device in the method in the embodiment of the present disclosure, and the principle of solving the problem by the electronic device is similar to that of the method, the implementation of the electronic device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0247] As shown in FIG12 , the electronic device includes a processor 1200 and a memory 1201 . The memory 1201 is used to store programs executable by the processor 1200 . The processor 1200 is used to read the programs in the memory 1201 and perform the following steps:

[0248] Obtain a methylation dataset of the target object;

[0249] Inputting the methylation dataset into a staging prediction model to output a cancer staging value of the target subject;

[0250] Among them, the stage prediction model is obtained by integrating n classification models, the model parameters of the n classification models are optimized using the grid retrieval method, the n classification models are screened from m classification models using the cross-validation method, and the m classification models are independently trained using the same data set, where m>n, m is an integer greater than 2, and n is an integer greater than 1.

[0251] As an optional implementation manner, the processor 1200 is specifically configured to execute:

[0252] Acquire methylation data corresponding to a target methylation site of a target object, and determine the methylation dataset based on the methylation data corresponding to the target methylation site;

[0253] The target methylation site is determined based on the degree of difference in methylation data of the same methylation site corresponding to different cancer stage values, or the correlation between the methylation data corresponding to the methylation site and different cancer stage values.

[0254] As an optional embodiment, the staging prediction model further includes a standardization model, which is used to standardize the methylation data in the methylation dataset by removing the mean and normalizing the variance.

[0255] As an optional implementation, the staging prediction model further includes a dimensionality reduction model, and the dimensionality reduction model is used to perform dimensionality reduction processing on the methylation data in the methylation dataset according to the PCA method.

[0256] As an optional implementation, the processor 1200 is specifically configured to determine the staging prediction model in the following manner:

[0257] Using the data set to independently train the m classification models, and determine the classification accuracy of each of the m classification models;

[0258] According to the classification accuracy, k classification models are selected from the m classification models.

[0259] As an optional implementation manner, after selecting k classification models from m classification models according to the classification accuracy, the processor 1200 is further configured to execute:

[0260] Using a cross-validation method, the n classification models are selected from the k classification models, where m≥k≥n; and / or,

[0261] The model parameters of the screened classification model are adjusted using the grid retrieval method.

[0262] As an optional implementation manner, the processor 1200 is specifically configured to execute:

[0263] Using the data set, cross-validate the k classification models respectively to obtain evaluation index values ​​of the k classification models;

[0264] According to the evaluation index values ​​of the k classification models, n classification models are screened out from the k classification models.

[0265] As an optional implementation, after obtaining the data set, the processor 1200 is further configured to execute:

[0266] The methylation data corresponding to the target methylation site is screened out from the methylation data contained in the data set, and the m classification models are independently trained using the screened methylation data.

[0267] As an optional embodiment, the data set includes sample sets of different target subjects, the sample sets include methylation data corresponding to each methylation site, and one sample set corresponds to one cancer stage value; the processor 1200 is specifically configured to determine the target methylation site by:

[0268] Dividing the data set into multiple subsets according to the cancer stage values ​​corresponding to the sample sets, wherein each subset corresponds to one cancer stage value, and the sample sets included in different subsets have different cancer stage values ​​corresponding to them;

[0269] For each methylation site, a target methylation site is determined based on the methylation data corresponding to the methylation site in multiple subsets.

[0270] As an optional implementation manner, the processor 1200 is specifically configured to execute:

[0271] If the dataset is divided into two subsets, then for each methylation site, the target methylation site is determined based on the degree of difference in the methylation data corresponding to the methylation site in the two subsets; or

[0272] If the data set is divided into two or more subsets, for each methylation site, a target methylation site is determined based on the correlation between the methylation data corresponding to the methylation site in each subset and the cancer stage value corresponding to the subset.

[0273] As an optional implementation manner, the processor 1200 is specifically configured to determine the degree of difference between the methylation data corresponding to the methylation sites in the two subsets in the following manner:

[0274] Performing a rank sum test on the methylation data corresponding to the methylation sites in the two subsets, and determining the degree of difference according to the rank sum test value; or

[0275] The correlation between the methylation data corresponding to the methylation sites in each subset and the cancer stage value corresponding to the subset is determined by:

[0276] A chi-square test is performed on the methylation data corresponding to the methylation sites in each subset and the cancer stage value corresponding to the subset, and the correlation is determined according to the chi-square test value.

[0277] As an optional implementation, the multiple subsets include more than two subsets; the data set is divided into multiple subsets according to the cancer stage values ​​corresponding to the sample set, and the processor 1200 is further configured to execute:

[0278] If, among the multiple divided subsets, there is a subset whose number of sample sets is lower than a threshold, then the subset is merged into another subset based on the cancer stage value corresponding to the subset, and the cancer stage value corresponding to the merged another subset is determined, so as to use the merged subsets to determine the target methylation site and train the classification model; wherein the cancer stage value corresponding to the another subset is closest to the cancer stage value corresponding to the subset.

[0279] As an optional implementation manner, the processor 1200 is further configured to execute:

[0280] For the sample set of the target subject, the methylation data corresponding to the methylation sites of the normal tissue adjacent to the cancer in the sample set and / or the methylation data corresponding to the methylation sites containing missing values ​​are removed.

[0281] As an optional implementation, the stage prediction model is integrated according to the following method:

[0282] Using the outputs of n-1 classification models as the input of the nth classification model, and integrating the n classification models into the stage prediction model;

[0283] The input of the n-1 classification models is the data set, and the output of the nth classification model is the cancer stage value corresponding to the data set.

[0284] Based on the same inventive concept, embodiments of the present disclosure provide a computer storage medium comprising computer program code. When executed on a computer, the computer program code causes the computer to perform any of the cancer staging methods discussed above. Because the principles underlying the problems solved by the computer storage medium are similar to those of the cancer staging methods, the implementation of the computer storage medium can be referenced to the implementation of the methods, and any repetitive details will not be repeated.

[0285] In a specific implementation process, computer storage media may include: Universal Serial Bus Flash Drive (USB), mobile hard disk, Read-Only Memory (ROM), Random Access Memory (RAM), magnetic disk or optical disk, and other storage media that can store program code.

[0286] Based on the same inventive concept, embodiments of the present disclosure further provide a computer program product comprising computer program code that, when executed on a computer, causes the computer to perform any of the cancer staging methods discussed above. Because the principles underlying the problems solved by the computer program products are similar to those of the cancer staging methods, the implementation of the computer program products can be referenced to the implementation of the methods, and any repetitive details will not be repeated.

[0287] The computer program product can employ any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0288] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0289] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0290] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0291] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0292] Obviously, those skilled in the art may make various changes and modifications to the present disclosure without departing from the spirit and scope of the present disclosure. Thus, if these modifications and variations of the present disclosure fall within the scope of the claims of the present disclosure and their equivalents, the present disclosure is intended to include these modifications and variations.

Claims

1. A method for staging cancer, wherein: The method includes: Obtain a methylation dataset of a target object; Inputting the methylation dataset into a staging prediction model, and outputting a cancer staging value of the target object; Among them, the stage prediction model is obtained by integrating n classification models, the model parameters of the n classification models are optimized by using the grid retrieval method, the n classification models are screened from m classification models by using the cross-validation method, and the m classification models are independently trained using the same data set, and m>n, m is an integer greater than 2, and n is an integer greater than 1.

2. The method according to claim 1, wherein: Obtain the methylation dataset of the target object, including: Acquire methylation data corresponding to a target methylation site of a target object, and determine the methylation data set according to the methylation data corresponding to the target methylation site; The target methylation site is determined based on the degree of difference in methylation data of the same methylation site corresponding to different cancer stage values, or the correlation between the methylation data corresponding to the methylation site and different cancer stage values.

3. The method according to claim 1, wherein: The staging prediction model also includes a standardization model, which is used to standardize the methylation data in the methylation data set by removing the mean and normalizing the variance.

4. The method according to claim 1, wherein: The staging prediction model also includes a dimensionality reduction model, which is used to perform dimensionality reduction processing on the methylation data in the methylation data set according to the PCA method.

5. The method according to claim 1, wherein: The staging prediction model is determined by: The m classification models are trained independently using the data set to determine the m classifications. The classification accuracy of each model; According to the classification accuracy, k classification models are selected from the m classification models.

6. The method according to claim 5, wherein: After selecting k classification models from m classification models according to the classification accuracy, the method further includes: Using a cross-validation method, the n classification models are selected from the k classification models, wherein m≥k≥n; and / or, The model parameters of the screened classification model are adjusted using the grid retrieval method.

7. The method according to claim 6, wherein: The method of using cross validation to select the n classification models from the k classification models includes: Using the data set, cross-validating the k classification models respectively to obtain evaluation index values ​​of the k classification models; According to the evaluation index values ​​of the k classification models, n classification models are selected from the k classification models.

8. The method according to claim 5, wherein: After obtaining the data set, the method further includes: The methylation data corresponding to the target methylation site is screened out from the methylation data contained in the data set, and the m classification models are independently trained using the screened methylation data.

9. The method according to claim 2 or 8, wherein: The data set includes sample sets of different target objects, the sample sets include methylation data corresponding to each methylation site, and one sample set corresponds to one cancer stage value; the target methylation site is determined by: According to the cancer staging values ​​corresponding to the sample sets, the data set is divided into a plurality of subsets, wherein each subset corresponds to a cancer staging value, and the sample sets included in different subsets correspond to different cancer staging values; For each methylation site, a target methylation site is determined according to the methylation data in multiple subsets respectively corresponding to the methylation site.

10. The method according to claim 9, wherein: Determining the target methylation site according to the methylation data respectively corresponding to the methylation site in the plurality of subsets includes: If the data set is divided into two subsets, for each methylation site, the target methylation site is determined according to the difference between the methylation data corresponding to the methylation site in the two subsets. locus; or, If the data set is divided into two or more subsets, for each methylation site, a target methylation site is determined according to the correlation between the methylation data corresponding to the methylation site in each subset and the cancer stage value corresponding to the subset.

11. The method according to claim 10, wherein: The difference degree of the methylation data corresponding to the methylation sites in the two subsets is determined by: Performing a rank sum test on the methylation data corresponding to the methylation sites in the two subsets respectively, and determining the degree of difference according to the rank sum test value; or, The correlation between the methylation data corresponding to the methylation site in each subset and the cancer stage value corresponding to the subset is determined in the following manner: A chi-square test is performed on the methylation data corresponding to the methylation site in each subset and the cancer stage value corresponding to the subset, and the correlation is determined according to the chi-square test value.

12. The method according to claim 9, wherein: The plurality of subsets include more than two subsets; The step of dividing the data set into a plurality of subsets according to the cancer staging values ​​corresponding to the sample set further includes: If, among the multiple divided subsets, there is a subset whose number of sample sets is lower than a threshold, the subset is merged into another subset according to the cancer stage value corresponding to the subset, and the cancer stage value corresponding to the merged another subset is determined, so as to use the merged subsets to determine the target methylation site and train the classification model; wherein the cancer stage value corresponding to the another subset is closest to the cancer stage value corresponding to the subset.

13. The method according to claim 9, wherein: Also includes: For the sample set of the target object, the methylation data corresponding to the methylation sites of the adjacent normal tissues in the sample set and / or the methylation data corresponding to the methylation sites containing missing values ​​are removed.

14. The method according to claim 1, wherein: The stage prediction model is integrated according to the following method: Using the outputs of n-1 classification models as the input of the nth classification model, and integrating the n classification models into the stage prediction model; The input of the n-1 classification models is the data set, and the output of the nth classification model is the cancer stage value corresponding to the data set.

15. An electronic device, wherein: The electronic device comprises a processor and a memory, wherein the memory is used to store a program executable by the processor, and the processor is used to read the program in the memory and execute the steps of any one of the methods of claims 1 to 14.

16. A computer storage medium having a computer program stored thereon, wherein: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 14 are implemented.

Images