A method for screening protein markers for breast cancer based on the estrogen receptor-epigenetic factor interaction network

Abstract

The present invention provides a method for screening protein markers for breast cancer based on an estrogen receptor-epigenetic factor interaction network, and belongs to the technical field of screening protein markers for breast cancer. The technical problem to be solved is: to provide an improvement on the method for screening protein markers for breast cancer based on an estrogen receptor-epigenetic factor interaction network. The technical scheme adopted to solve the technical problem is: to collect multiple groups of mass spectrometry data of protein interaction groups of estrogen receptors and epigenetic factors from a public database, to perform qualitative and quantitative analysis on the protein group using MaxQuant software, and to perform peptide library search and matching; to perform matching and scoring on peptide-spectrum using an algorithm based on binomial distribution probability, to further sort the peptides by the scoring, to determine peptide modifications, and to calculate the scores including neutral losses and excluding neutral losses each time. The present invention is applied to screening protein markers for breast cancer.

Description

Technical Field

[0001] The present invention provides a method for screening protein markers of breast cancer based on an estrogen receptor-epigenetic factor interaction network, belonging to the technical field of screening protein markers of breast cancer. Background Art

[0002] Breast cancer, as a common female malignancy, shows a high degree of heterogeneity, including differences in molecular subtypes, clinical pathological characteristics and treatment response. In clinical practice, accurate classification and prognostic judgment are crucial for the selection of individualized treatment strategies. Current breast cancer classification methods often rely on technologies such as genomics and transcriptomics. However, these methods ignore changes at the protein level, especially the role of estrogen receptors and their interacting proteins in the occurrence and development of breast cancer, resulting in low classification accuracy. In addition, traditional breast cancer marker detection technologies are mostly limited to the study of single molecules, and there are problems such as insufficient marker specificity and cumbersome screening process leading to low detection efficiency, and lack of comprehensive consideration of protein interactions within breast cancer cells. Summary of the invention

[0003] In order to overcome the deficiencies in the prior art, the present invention aims to solve the technical problem of providing an improved method for screening protein markers for breast cancer based on an estrogen receptor-epigenetic factor interaction network.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for screening breast cancer protein markers based on the estrogen receptor-epigenetic factor interaction network, comprising the following screening steps:

[0005] Step 1: Construct a comprehensive estrogen receptor and epigenetic factor interaction network by mining and analyzing proteomic data:

[0006] Step 1.1: Collect multiple groups of protein interaction mass spectrometry data of estrogen receptor ERα / β and epigenetic factors METTL3 / 14 and TET2 from the public databases ProteomeXchange, UniProt, and NCBI, and use MaxQuant software to perform qualitative and quantitative analysis of the proteome and perform peptide library search and matching;

[0007] Step 1.2: Use an algorithm based on binomial distribution probability to score the peptide-spectrum match, and further sort the peptides by score to determine peptide modifications:

[0008] First, calculate the number of matches between the theoretical fragment masses and the peaks in the mass spectrum k , then by calculating at least k The probability of a match is calculated and its negative logarithm is multiplied by 10 to obtain the score s of the corresponding spectrum (q , loss ), which reflects the similarity between the theoretical fragment mass list and the mass spectrum mass list, and is calculated as:

[0009] ;

[0010] Where: n represents the total number of theoretical ions, k Indicates the number of matches between the theoretical fragment masses and the peaks in the mass spectrum, q Indicates the number of peaks allowed to appear in the mass interval, which is used to calculate the probability of a single random match;

[0011] In mass spectrometry analysis, compound molecules undergo a fragmentation process to generate different fragment ions and produce neutral losses. Each time, the scores including neutral losses and excluding neutral losses are calculated, and the maximum of the two scores is selected. The same scoring operation is used for the sequence from the reverse database. The calculation formula is:

[0012] ;

[0013] Where: s ( q ) represents the maximum value of the scores including neutral loss and excluding neutral loss. The best score is selected by calculating the scores under different q values;

[0014] Based on the above scoring operation, nuclear receptors and epigenetic factors were screened from the protein list;

[0015] Step 1.3: Collect the published evidence of the interaction between estrogen receptor and epigenetic factors in the protein interaction databases STRING and BioGRID, integrate the results with the nuclear receptors and epigenetic factors screened in step 1.2, and use Cytoscape software to draw a protein interaction network diagram. The drawn protein interaction network diagram includes the interaction between estrogen receptor and nuclear receptor and various epigenetic modifiers;

[0016] Step 1.4: For breast cancer proteome data, use Limma package software to perform differential analysis between breast cancer samples and adjacent normal tissue samples, and screen out proteins that are differentially expressed in breast cancer tissues. These differentially expressed proteins are related to the occurrence and development of breast cancer.

[0017] Step 2: Use machine learning methods to identify new types of breast cancer and then screen out protein markers:

[0018] Step 2.1: Determine new breast cancer classification based on consensus clustering algorithm:

[0019] Nuclear receptors and epigenetic factors that interact with estrogen receptors were screened out from the differentially expressed proteins as candidates for classification and markers. The expression data of these differentially expressed proteins in breast cancer samples were used to perform unsupervised clustering of breast cancer using a consensus clustering algorithm.

[0020] The PAM algorithm of center point partitioning is used for unsupervised clustering. By selecting representative samples as center points, the samples in the data set are divided into the center points closest to them, and the optimal center point configuration is found so that the total distance between all samples and their corresponding center points is minimized.

[0021] Step 2.2: Design an integrated classifier model PEC based on proteomic data, take the protein expression level in breast cancer samples as features, take breast cancer samples with classification labels as target variables, and fuse multiple separate classifiers to predict and screen protein markers. The specific design method of the integrated classifier model PEC is as follows:

[0022] The grid search method is used to determine the best parameters of each classifier by exhaustively enumerating all possible combinations in the parameter space, thereby optimizing the performance of individual classifiers. In this process, a soft voting strategy is used to perform a weighted average of the prediction results of each classifier, where the weight depends on the confidence of each classifier.

[0023] The individual classifiers are combined into a voting classifier, and the prediction results of multiple classifiers are combined to determine the final classification decision.

[0024] The specific method for peptide library matching in step 1.1 is:

[0025] Step 1.1.1: Sample preparation and protein extraction;

[0026] Step 1.1.2: Separate proteins by polyacrylamide gel electrophoresis, separate the proteins according to their isoelectric points, and further separate them according to their molecular weight to obtain a protein map;

[0027] Step 1.1.3: Digest the protein and use trypsin to hydrolyze the protein into peptides. Then use a mass spectrometer to analyze the digested protein and collect peptide mass spectrometry data.

[0028] Step 1.1.4: Match the peptide mass spectrum measured in the above experiment with the theoretical peptide mass spectrum in the database. The peptide mass spectrum matching strategy uses the peptide matching index to determine the protein identification result by comparing the mass spectrum peak in the peptide mass spectrum with the theoretical mass spectrum peak in the database;

[0029] Step 1.1.5: Detect protein expression levels and protein interactions through protein immunoblotting experiments, verify the proteins identified by mass spectrometry analysis, and use biological experiments to verify the functions and effects of the identified proteins in biological processes.

[0030] The specific calculation steps of the PAM algorithm used in step 2.1 are:

[0031] Step 2.1.1: Initialization: There are n samples in total, and k samples are randomly selected as center points;

[0032] Step 2.1.2: Calculate distance and assign samples: Calculate the distance between each sample and the center point, and assign each sample to the category of the center point closest to it;

[0033] Step 2.1.3: Update the center point: For all other points in the i-th class except the center point, calculate the sum of the distances between each sample and other samples in order, that is, the value of the criterion function, and the calculation formula is:

[0034] ;

[0035] Where: E represents the sum of squares of errors, that is, the sum of squares of distances from all points to their cluster centers. represents the number of cluster centers, that is, the number of cluster categories, p represents a point in the data set, that is, a data sample, o i Represents the i-th category C i The center point, dist(p, o i ) represents the data point p to the cluster center o i distance;

[0036] Step 2.1.4: Iterative update: Repeat the process of assigning samples and updating center points until convergence, that is, the center points no longer change or the maximum number of iterations has been reached;

[0037] Step 2.1.5: Output results: Get the final number of cluster centers .

[0038] The specific method for designing the integrated classifier model PEC for data processing in step 2.2 is:

[0039] Step 2.2.1: Input the protein expression matrix in the sample, where each row represents a sample, each column represents the expression level of a protein, and the classification information column represents the category to which each sample belongs;

[0040] Step 2.2.2: Separate the feature matrix X and label vector y from the input data, divide it into training set and test set, and use 70% of the data as training set and 30% of the data as test set;

[0041] Step 2.2.3: Initialize the model: Use the grid search method, for each parameter combination, use 10-fold cross validation on the training set, and select the preset parameter combination as the parameters of the final model;

[0042] Step 2.2.4: Combine the initialized models into a voting classifier, using a soft voting strategy;

[0043] Step 2.2.5: Fit the voting classifier using the training set data;

[0044] Step 2.2.6: Output protein marker contribution value:

[0045] The voting classifier performs weighted voting on the prediction results of each initialization model to obtain the final prediction result;

[0046] Traverse each model in the voting classifier, extract its feature importance, calculate the average of all model feature importances, and get the average contribution of each feature;

[0047] To calculate the contribution of proteins in different groups, a voting classifier was used to predict the training set to obtain the probability that each sample belonged to each category, and these probabilities were used as the contribution of each feature in different groups.

[0048] The beneficial effects of the present invention compared with the prior art are as follows: in order to solve the defects in the current breast cancer typing and marker screening methods, the present invention constructs an estrogen receptor-epigenetic factor interaction network, comprehensively considers the protein interactions inside breast cancer cells, improves the typing accuracy, and optimizes the marker screening process. The present invention can more accurately identify key protein markers in breast cancer, effectively improve screening efficiency and accuracy, and provide strong support for subsequent precision medicine for breast cancer. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The present invention will be further described below in conjunction with the accompanying drawings:

[0050] Figure 1 A flowchart of the steps of protein library search and identification in the present invention;

[0051] Figure 2 This is a flow chart of the steps of the PEC algorithm used in the present invention. DETAILED DESCRIPTION

[0052] like Figure 1 and Figure 2As shown, the present invention specifically provides a method for screening protein markers for breast cancer based on an estrogen receptor-epigenetic factor interaction network, which involves analysis and identification of proteomic data. The method mainly constructs an estrogen receptor-epigenetic factor interaction network, performs differential analysis on the proteomic data of breast cancer samples relative to normal samples, screens differentially expressed proteins, and then selects nuclear receptors and epigenetic factors that interact with estrogen receptors from the differentially expressed proteins. The expression data of these proteins in breast cancer samples are used to apply machine learning methods to type breast cancer samples, and protein markers in different types are screened, thereby effectively improving the specificity and screening efficiency of the markers, more accurately identifying key protein markers in breast cancer, and providing a new basis for the typing of breast cancer.

[0053] The technical solution adopted by the present invention mainly includes:

[0054] First, by mining and analyzing proteomic data, a comprehensive estrogen receptor and epigenetic factor interaction network (ER-Epifactor) was constructed.

[0055] 2. The use of machine learning methods revealed new types of breast cancer and screened out protein markers.

[0056] The specific steps are as follows:

[0057] (1) First, 17 groups of protein interaction group mass spectrometry data of estrogen receptors (ERα / β) and epigenetic factors (METTL3 / 14, TET2) were collected from the public databases ProteomeXchange, UniProt, and NCBI, as shown in Table 1. MaxQuant software was used to perform qualitative and quantitative analysis on the protein interaction group mass spectrometry data. The peptide library search and matching process is as follows: Figure 1 shown.

[0058] The specific method for peptide library matching is as follows:

[0059] 1. Sample preparation and protein extraction. Collect biological samples, which can be cell cultures, tissue samples, serum, urine, etc. For cell or tissue samples, it is usually necessary to first disrupt the cells to release proteins, and then use appropriate buffers to extract proteins from the samples.

[0060] 2. Protein separation. Polyacrylamide gel electrophoresis (SDS-PAGE) is used to separate proteins. First, the proteins are separated according to their isoelectric point (pI), and then further separated according to their molecular weight to obtain a higher resolution protein map.

[0061] 3. Obtain peptides using mass spectrometry. First, digest the protein and use trypsin to hydrolyze the protein into peptides. Then use a mass spectrometer (liquid chromatography-mass spectrometry LC-MS / MS) to analyze the digested protein and collect peptide mass spectrometry data.

[0062] 4. Peptide mass spectrum matching. Match the experimentally measured peptide mass spectrum with the theoretical peptide mass spectrum in the database. The peptide mass spectrum matching strategy can use peptide matching indicators, such as general spectrum matching, library matching, template matching, etc., to determine the protein identification result by comparing the mass spectrum peaks in the peptide mass spectrum with the theoretical mass spectrum peaks in the database.

[0063] 5. Result verification. Use Western Blot to detect protein expression levels and protein interactions, and verify the proteins identified by mass spectrometry. Use biological experiments (such as cell or mouse experiments) to verify the functions and effects of the identified proteins in biological processes.

[0064]

[0065] Table 1 Data collection of estrogen receptor and epigenetic factor interaction groups

[0066] (2) Use an algorithm based on binomial distribution probability to score the peptide-spectrum match, and further sort the peptides by score to determine peptide modifications, etc. In mass spectrometry analysis, the molecules of the compound will undergo a fragmentation process to generate different fragment ions. When some ions in these fragmented fragments lose one or more neutral molecules (such as H2O, NH3), atoms or modification groups (such as phosphate groups, sugar groups, etc.), neutral loss will occur. The calculation formula of the scoring model algorithm is shown in formula (1), where n represents the total number of theoretical ions, k Indicates the number of matches between the theoretical fragment masses and the peaks in the mass spectrum, q Represents the number of peaks allowed to appear in the mass interval, which is used to calculate the probability of a single random match; first, the number of matches k between the theoretical fragment mass and the peaks in the mass spectrum is calculated, and then the probability of at least k matches is calculated and its negative logarithm is multiplied by 10 to obtain the score s(q,loss) of the corresponding spectrum. This score reflects the similarity between the theoretical fragment mass list and the mass spectrum mass list.

[0067] (1);

[0068] Each time, the score including neutral loss and excluding neutral loss is calculated, and the maximum of the two scores is selected. The calculation formula is shown in formula (2). For sequences from the reverse database, all scoring processes operate in the same way, so no bias is introduced.

[0069] (2);

[0070] Where: s(q) represents the maximum value of the score including neutral loss and excluding neutral loss. By calculating the score at different q values ​​(number of peaks per 100 Th) and selecting the best score, the intensity of the peak in the mass spectrum is indirectly taken into account. This means that even if the n and k values ​​are the same, different peak intensities will lead to different scores. The final score s represents the optimization of the number of highest intensity peaks considered per 100 Da m / z interval, as well as the inclusion of specific modification neutral losses. This means that when calculating the matching score, the model will give priority to those peaks with the highest intensity in a specific mass interval, and will pay special attention to those neutral loss peaks associated with specific protein modifications. In this way, the model is able to more accurately match experimental mass spectrometry data with theoretical peptide fragmentation patterns, thereby improving the accuracy of protein identification.

[0071] Based on the above scoring operation, nuclear receptors and epigenetic factors were screened out from the protein list.

[0072] (3) Collect the published evidence of the interaction between estrogen receptor and epigenetic factors in the protein interaction databases STRING and BioGRID, integrate the results with the nuclear receptors and epigenetic factors screened in step (2), and use Cytoscape to draw a protein interaction network diagram. This network covers the interaction between estrogen receptor (ER) and other nuclear receptors (such as androgen receptor, progesterone receptor, etc.) and various epigenetic modifiers (such as histone modification enzymes, DNA methylation modification enzymes, etc.).

[0073] (4) Differential analysis and screening of differentially expressed proteins. Based on the breast cancer proteome data, the Limma package was used to compare and analyze breast cancer samples with adjacent normal tissue samples to screen out proteins that are differentially expressed in breast cancer tissues. These differentially expressed proteins may be closely related to the occurrence and development of breast cancer.

[0074] (5) Consistent clustering reveals new types of breast cancer. Nuclear receptors and epigenetic factors that interact with estrogen receptors were selected from the differentially expressed proteins as candidates for classification and markers. The expression data of these proteins in breast cancer samples were used to perform unsupervised clustering of breast cancer using the consistent clustering algorithm. The partitioning around medoids (PAM) algorithm was used for unsupervised clustering. Its core principle is K-medoids. By selecting representative samples as medoids, the samples in the data set are divided into the medoids closest to them, that is, an optimal medoid configuration is found so that the total distance between all samples and their corresponding medoids is minimized.

[0075] The PAM algorithm calculation process used mainly includes the following steps:

[0076] 1. Initialization: There are n samples in total (n=392 in this study), and k samples are randomly selected as center points.

[0077] 2. Calculate distance and assign samples: Calculate the distance between each sample and the center point, and assign each sample to the category of the center point closest to it.

[0078] 3. Update the center point: For all other points in the i-th class except the center point, calculate the sum of the distances between each sample and other samples in order, that is, the value of the criterion function. The calculation formula is shown in formula (3), where represents the i-th classification C i The center point of the criterion function is selected, and the point corresponding to the minimum value of the criterion function is selected as the new center point.

[0079] (3);

[0080] Where: E represents the sum of squares of errors, that is, the sum of squares of distances from all points to the centers of their clusters. The smaller this value is, the better the clustering effect is. represents the number of cluster centers, that is, the number of cluster categories. p represents a point in the data set, that is, a data sample. i Represents the i-th category C i The center point, dist(p, o i ) represents the data point p to the cluster center o i The distance metric used in this study is the Pearson correlation coefficient, which measures the similarity between data points. Finally, the point corresponding to the minimum criterion function value is selected as the new center point.

[0081] 4. Iterative update: Repeat the process of allocating samples and updating center points until convergence, that is, the center points no longer change or the set maximum number of iterations has been reached.

[0082] 5. Output results: get the final number of cluster centers .

[0083] (6) Screening protein markers through machine learning fusion models. The present invention proposes a protein-ensemble-classifier model (PEC) based on proteomic data. It is a supervised machine learning model that uses the protein expression level in breast cancer samples as features and breast cancer samples with classification labels as target variables. It integrates multiple separate classifiers to predict protein markers. The design method for the PEC model includes:

[0084] The grid search method is used to determine the optimal parameters of each classifier by exhaustively enumerating all possible combinations in the parameter space, thereby optimizing the performance of individual classifiers and improving the performance and generalization ability of the overall model. The optimal parameter settings for each classifier are shown in Table 2.

[0085]

[0086] Table 2 Optimal parameter settings for the basic model

[0087] Combining individual classifiers into a voting classifier means combining the predictions of multiple classifiers to determine the final classification decision. In this process, a soft voting strategy is adopted, which is to weight the prediction results of each classifier, where the weight depends on the confidence of each classifier. This approach aims to comprehensively utilize the prediction capabilities of multiple classifiers and improve the prediction accuracy and stability of the integrated model.

[0088] For the specific PEC algorithm process, see Figure 2 The specific method of using this model to process data is:

[0089] 1. Input the protein expression matrix in the sample, where each row represents a sample, each column represents the expression level of a protein, and the classification information column represents the category to which each sample belongs.

[0090] 2. Data preprocessing: Separate the feature matrix X (protein expression matrix) and label vector y (sample classification label) from the input data. Divide the data into training set (X_train, y_train) and test set (X_test, y_test), using 70% of the data as the training set and 30% of the data as the test set.

[0091] 3. Initialize the model: Use the grid search method, and for each parameter combination, use 10-fold cross validation on the training set to select the parameter combination with the best performance as the parameters of the final model. The optimal parameter settings for each basic model are shown in Table 2 above. Determining the optimal parameter combination through the grid search method can make the integrated model more adaptable to specific data sets and task requirements, thereby achieving better prediction results.

[0092] 4. Create a voting classifier: Combine the initialized models into a voting classifier (VotingClassifier) ​​using a soft voting strategy.

[0093] 5. Fit the voting classifier: Fit the voting classifier using the training set data (X_train, y_train).

[0094] 6. Output protein marker contribution value: The voting classifier performs weighted voting on the prediction results of each initialization model to obtain the final prediction result. Traverse each model in the voting classifier, extract its feature importance (FeatureImportances), calculate the average of all model feature importances, and obtain the average contribution of each feature.

[0095] To calculate the contribution of proteins in different groups, a voting classifier was used to predict the training set to obtain the probability that each sample belonged to each category, and these probabilities were used as the contribution of each feature in different groups.

[0096] This invention constructs a comprehensive estrogen receptor-epigenetic factor interaction network map for the first time. The network covers 186 proteins, of which 173 proteins interact with ERα, 133 proteins interact with ERβ, and 120 proteins interact with both ERα and ERβ, which fills the gap in estrogen receptor-epigenetic factor interaction network data in public databases (such as STRING).

[0097] The present invention applies the estrogen receptor-epigenetic factor network to the research on breast cancer classification and prognosis and treatment, and uses a consistent clustering algorithm to reveal four new classifications of breast cancer.

[0098] The present invention designs an integrated classifier model (PEC) based on proteomic data for screening protein markers. By integrating five basic classifier models (random forest, gradient boosting algorithm, adaptive boosting algorithm, logistic regression and support vector machine), a soft voting strategy is adopted to achieve the prediction ability of the integrated model. By comparing with the five individual models, the PEC model shows better classification and prediction performance, with the accuracy and specificity increased by 0.74% and 1.64% respectively.

[0099] Further screening of 16 significant proteins based on the contribution values ​​of proteins in different subtypes, and functional enrichment analysis of these protein markers revealed their roles in different biological processes. These findings will help to deepen our understanding of the development and prognosis of breast cancer and provide patients with more personalized and precise treatment plans.

[0100] The present invention provides a new method for breast cancer typing and marker screening by integrating the estrogen receptor interaction network and epigenetic information. Compared with the traditional method based on gene expression profile, the present invention pays more attention to the changes at the protein level, which can more truly reflect the biological characteristics of breast cancer and more comprehensively reveal the occurrence and development mechanism of breast cancer. In addition, a new typing of breast cancer samples is proposed through consistency clustering, and a machine learning fusion model PEC is proposed to screen protein markers. As shown in Table 3, compared with the basic classifier alone, the PEC model shows better classification and prediction performance, with an accuracy and specificity of 0.74% and 1.64% respectively. By making full use of big data analysis technology, the accuracy and specificity of breast cancer marker screening are improved, and markers with biological significance can be screened more effectively, which is expected to provide strong support for the early diagnosis, molecular subtype classification, prognosis and treatment strategy selection of breast cancer, and has broad clinical application prospects.

[0101]

[0102] Table 3 Performance comparison of PEC model with other models

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for screening breast cancer protein markers based on the estrogen receptor-epigenetic factor interaction network, characterized in that: The screening steps include the following: Step 1: Construct a comprehensive interaction network between estrogen receptor and epigenetic factors by mining and analyzing proteomic data: Step 1.1: Collect multiple groups of protein interaction mass spectrometry data of estrogen receptor ERα / β and epigenetic factors METTL3 / 14 and TET2 from the public databases ProteomeXchange, UniProt, and NCBI, and use MaxQuant software to perform qualitative and quantitative analysis of the proteome and perform peptide library search and matching; Step 1.2: Use an algorithm based on binomial distribution probability to score the peptide-spectrum match, and further sort the peptides by score to determine peptide modifications: First, calculate the number of matches between the theoretical fragment masses and the peaks in the mass spectrum k , then by calculating at least k The probability of a match is calculated and its negative logarithm is multiplied by 10 to obtain the score s of the corresponding spectrum ( q , loss ), which reflects the similarity between the theoretical fragment mass list and the mass spectrum mass list, and is calculated as: ； Where: n represents the total number of theoretical ions, k Indicates the number of matches between the theoretical fragment masses and the peaks in the mass spectrum, q Indicates the number of peaks allowed to appear in the mass interval, which is used to calculate the probability of a single random match; In mass spectrometry analysis, compound molecules undergo a fragmentation process to generate different fragment ions and produce neutral losses. Each time, the scores including neutral losses and excluding neutral losses are calculated, and the maximum of the two scores is selected. The same scoring operation is used for the sequence from the reverse database. The calculation formula is: ； Where: s ( q ) represents the maximum value of the scores including neutral loss and excluding neutral loss. The best score is selected by calculating the scores under different q values; Based on the above scoring operation, nuclear receptors and epigenetic factors were screened from the protein list; Step 1.3: Collect the published evidence of the interaction between estrogen receptor and epigenetic factors in the protein interaction databases STRING and BioGRID, integrate the results with the nuclear receptors and epigenetic factors screened in step 1.2, and use Cytoscape software to draw a protein interaction network diagram. The drawn protein interaction network diagram includes the interaction between estrogen receptor and nuclear receptor and various epigenetic modifiers; Step 1.4: For breast cancer proteome data, use Limma package software to perform differential analysis between breast cancer samples and adjacent normal tissue samples, and screen out proteins that are differentially expressed in breast cancer tissues. These differentially expressed proteins are related to the occurrence and development of breast cancer. Step 2: Use machine learning methods to identify new types of breast cancer and then screen out protein markers: Step 2.1: Determine new breast cancer classification based on consensus clustering algorithm: Nuclear receptors and epigenetic factors that interact with estrogen receptors were screened out from the differentially expressed proteins as candidates for classification and markers. The expression data of these differentially expressed proteins in breast cancer samples were used to perform unsupervised clustering of breast cancer using a consensus clustering algorithm. The PAM algorithm of center point partitioning is used for unsupervised clustering. By selecting representative samples as center points, the samples in the data set are divided into the center points closest to them, and the optimal center point configuration is found so that the total distance between all samples and their corresponding center points is minimized. Step 2.2: Design an integrated classifier model PEC based on proteomic data, take the protein expression level in breast cancer samples as features, take breast cancer samples with classification labels as target variables, and fuse multiple separate classifiers to predict and screen protein markers. The specific design method of the integrated classifier model PEC is as follows: The grid search method is used to determine the best parameters of each classifier by exhaustively enumerating all possible combinations in the parameter space, thereby optimizing the performance of individual classifiers. In this process, a soft voting strategy is used to perform a weighted average of the prediction results of each classifier, where the weight depends on the confidence of each classifier. The individual classifiers are combined into a voting classifier, and the prediction results of multiple classifiers are combined to determine the final classification decision.

2. The method for screening breast cancer protein markers based on the estrogen receptor-epigenetic factor interaction network according to claim 1, characterized in that: The specific method for peptide library matching in step 1.1 is: Step 1.1.1: Sample preparation and protein extraction; Step 1.1.2: Separate proteins by polyacrylamide gel electrophoresis, separate the proteins according to their isoelectric points, and further separate them according to their molecular weight to obtain a protein map; Step 1.1.3: Digest the protein and use trypsin to hydrolyze the protein into peptides. Then use a mass spectrometer to analyze the digested protein and collect peptide mass spectrometry data. Step 1.1.4: Match the peptide mass spectrum measured in the above experiment with the theoretical peptide mass spectrum in the database. The peptide mass spectrum matching strategy uses the peptide matching index to determine the protein identification result by comparing the mass spectrum peak in the peptide mass spectrum with the theoretical mass spectrum peak in the database; Step 1.1.5: Detect protein expression levels and protein interactions through protein immunoblotting experiments, verify the proteins identified by mass spectrometry analysis, and use biological experiments to verify the functions and effects of the identified proteins in biological processes.

3. The method for screening breast cancer protein markers based on the estrogen receptor-epigenetic factor interaction network according to claim 1, characterized in that: The specific calculation steps of the PAM algorithm used in step 2.1 are: Step 2.1.1: Initialization: There are n samples in total, and k samples are randomly selected as center points; Step 2.1.2: Calculate distance and assign samples: Calculate the distance between each sample and the center point, and assign each sample to the category of the center point closest to it; Step 2.1.3: Update the center point: For all other points in the i-th class except the center point, calculate the sum of the distances between each sample and other samples in order, that is, the value of the criterion function, and the calculation formula is: ； Where: E represents the sum of squares of errors, that is, the sum of squares of distances from all points to their cluster centers. represents the number of cluster centers, that is, the number of cluster categories, p represents a point in the data set, that is, a data sample, o i Represents the i-th category C i The center point, dist(p, o i ) represents the data point p to the cluster center o i distance; Step 2.1.4: Iterative update: Repeat the process of assigning samples and updating center points until convergence, that is, the center points no longer change or the maximum number of iterations has been reached; Step 2.1.5: Output results: Get the final number of cluster centers .

4. The method for screening breast cancer protein markers based on the estrogen receptor-epigenetic factor interaction network according to claim 1, characterized in that: The specific method for designing the integrated classifier model PEC for data processing in step 2.2 is: Step 2.2.1: Input the protein expression matrix in the sample, where each row represents a sample, each column represents the expression level of a protein, and the classification information column represents the category to which each sample belongs; Step 2.2.2: Separate the feature matrix X and label vector y from the input data, divide it into training set and test set, and use 70% of the data as training set and 30% of the data as test set; Step 2.2.3: Initialize the model: Use the grid search method, for each parameter combination, use 10-fold cross validation on the training set, and select the preset parameter combination as the parameters of the final model; Step 2.2.4: Combine the initialized models into a voting classifier, using a soft voting strategy; Step 2.2.5: Fit the voting classifier using the training set data; Step 2.2.6: Output protein marker contribution value: The voting classifier performs weighted voting on the prediction results of each initialization model to obtain the final prediction result; Traverse each model in the voting classifier, extract its feature importance, calculate the average of all model feature importances, and get the average contribution of each feature; To calculate the contribution of proteins in different groups, a voting classifier was used to predict the training set to obtain the probability that each sample belonged to each category, and these probabilities were used as the contribution of each feature in different groups.

Images