Prediction method and system for subtype classification of triple negative breast cancer

The prediction classification model constructed through graph convolutional neural network uses messenger RNA and long-chain non-coding RNA feature data to solve the problem of subtype recognition of triple-negative breast cancer, achieve more accurate and stable subtype prediction, and improve the correct Understanding and support for the treatment of triple-negative breast cancer characteristics.

CN119069000AActive Publication Date: 2024-12-03THE ACAD OF TIANJIN UNIV HEFEI
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202411193945.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-28
Publication Date
2024-12-03
Estimated Expiration
2044-08-28

AI Technical Summary

Technical Problem

The existing technology is difficult to effectively identify and classify the subtypes of triple-negative breast cancer, resulting in unstable and incomparable treatment results. The traditional method brings uncertainty to the low incidence and heterogeneity of triple-negative breast cancer. .

Method used

Graphic convolutional neural network (GCNN) is used to construct a predictive classification model, and graph data sets are constructed through messenger RNA and long-chain non-coding RNA biological feature data, feature vectors are obtained and monospace discretized, adjacent matrix is ​​established, and graph convolution is used. The network performs feature extraction and classification to achieve accurate prediction of triple-negative breast cancer subtypes.

Benefits of technology

Improve the identification accuracy and stability of the subtype of triple-negative breast cancer, provide a new way to judge biological characteristics and disease correlation, enhance the learning ability of triple-negative breast cancer, and help discover disease-related Characteristics of .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119069000A_ABST
    Figure CN119069000A_ABST
Patent Text Reader

Abstract

The invention discloses a triple negative breast cancer subtype classification prediction method and system. The method comprises the following steps: constructing a graph data set; constructing a graph convolutional neural network prediction classification model; taking the graph data set as a graph convolutional neural network prediction classification model for training until the precision meets the requirement, and obtaining an optimal graph convolutional neural network prediction classification model; and predicting triple negative breast cancer subtype classification by using the optimal graph convolutional neural network prediction classification model. According to the prediction method and system for subtype classification of the triple negative breast cancer, the classification accuracy is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of breast cancer identification, and in particular to a prediction method and system for triple-negative breast cancer subtype classification. Background Art

[0002] Triple-negative breast cancer (TNBC) is a specific type of breast cancer with a relatively poor prognosis. It accounts for about one-fifth of all breast cancers, and nearly 20% of triple-negative breast cancer patients have mutations in the BRCA1 / 2 gene. Triple-negative breast cancer is characterized by the lack of expression of estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2) in breast tissue. Among them, the value of progesterone receptor and estrogen receptor is less than or equal to 1%, which is considered negative, and the result of human epidermal growth factor receptor 2 (- / +) is negative. In addition, if the result is (++), FISH detection is required. If there is no amplification, it is negative, and if there is amplification, it is positive. Due to the lack of these receptors (ER, PR, HER2), traditional hormone therapy and targeted therapy are usually ineffective for triple-negative breast cancer. Moreover, due to the uniqueness of triple-negative breast cancer, it grows very rapidly compared to other breast cancers and may rise rapidly in a short period of time. Therefore, it is particularly important to accurately determine the type of triple-negative breast cancer and to carry out targeted treatment in a timely manner.

[0003] Several studies have shown that triple-negative breast cancer mostly occurs in young premenopausal women. It has a higher chance of metastasis to the viscera than to the bones, and the chance of brain metastasis is also high. It is a highly diverse group of cancers and one of the most difficult to cure. Its high mortality rate is largely due to the complexity of the cancer and the significant differences in its clinical outcomes. Subtyping is necessary to better identify molecular-based therapies. Based on cancer genome clustering analysis, six TNBC subtypes were identified, including two basal-like (BL1 and BL2), immunomodulator (IM), mesenchymal (M), mesenchymal stem-like (MSL), and luminal androgen receptor (LAR) subtypes. Cell lines representing various subtypes have different sensitivities to targeted therapies currently in laboratory and clinical studies. Identifying various TNBC subtypes and molecular drivers in corresponding cell line models can provide deep insights into the heterogeneity of the disease, align TNBC patients with appropriate targeted therapies, and provide a preclinical platform for the development of effective treatments.

[0004] As we step into the information age, the combination of machine learning and medicine has become an irreversible trend, and more and more machine learning methods are being used to detect cancer, such as decision trees and neural networks. With the rapid development of high-throughput sequencing technology, rich gene expression profile data provides a lot of very accurate information for decision-making and accurate diagnosis of cancer patients. However, most of the above-mentioned machine learning methods are supervised learning methods. However, in bioinformatics, it is very expensive to obtain sticky note data, and having data is often not enough, so the role of semi-supervision is particularly important. Experts have now developed a semi-supervised learning method for predicting cancer clinical outcomes based on graph convolutional networks (GCN Graph Convolutional Network). Experts have also proposed an innovative method to apply GCN to gene selection. For example, Ning Shiqi et al. (Semi-supervised learning method for cancer clinical outcome prediction based on graph convolutional network [J]. Intelligent Computers and Applications, 2018, 8(06): 44-48+53.) made some improvements to the GCN network structure in order to distinguish normal samples from cancer samples. As a result, more disease-related genes were selected. At the same time, some more potential related genes that have not been discovered before were explored and mined, such as CLEC3B. Before the improvement, it could only be used to solve classification problems, but after the improvement, it was used for feature selection problems, selecting some genes related to the disease; Dong Hua et al. (Triple-negative breast cancer prediction model based on machine learning [J]. Journal of Yunnan University (Natural Science Edition), 2017, 39(S1): 111-115.) used machine learning algorithms, combined with the gene expression data of triple-negative breast cancer and non-triple-negative breast cancer in the TCGA database, and used the decision tree algorithm and support vector machine feature elimination algorithm in machine learning to construct a classification model to distinguish triple-negative breast cancer from non-triple-negative breast cancer. Among them, 9 characteristic genes were extracted through the decision tree algorithm with an accuracy of 95.5%, and 6 characteristic genes were extracted through the support vector machine feature elimination algorithm with an accuracy of 97.8%, breaking through the inevitable shortcomings of traditional diagnostic methods and improving their accuracy through continuous optimization.

[0005] In addition, many researchers abroad have made important contributions to the exploration of triple-negative breast cancer. For example, in 2016, Brian et al. (Generation of an algorithm based on minimal gene sets toclinically subtype triple negative breast cancer patients. BMC Cancer 16, 143 (2016).) used the same expression dataset as the original TNBC algorithm to build a new classification model. Gene set enrichment was used, followed by a tightened centroid analysis to reduce features, and then elastic net regularized linear modeling was used to identify genes for the centroid model for classification of all subtypes. The model using a small gene set summarizes the TNBC subtypes identified by the original 2188-gene model, which can predict the ability of treatment response in the case of standard chemotherapy; in 2018, Santonja et al. (Triple negative breast cancer subtypes and pathologic complete response rate to neoadjuvant chemotherapy. Oncotarget. 2018; 9(41): 26406-26416.) identified the Lehmann subtype by gene expression profiles in paraffin pre-treatment tumor biopsies of 125 TNBC patients treated with neoadjuvant anthracyclines and / or taxanes + / - carboplatin. At the same time, the clinicopathological characteristics of the Lehmann subtype and its relationship with pathological complete remission (PCR) of different treatments were explored. In the study, only tumors with the LAR phenotype showed a non-basal-like intrinsic subtype (HER2-rich luminal subtype). At the same time, it was confirmed that TNBC tumors are highly genetically diverse and, from a molecular perspective, they are very heterogeneous, with LAR being the least proliferative tumor subtype, while the BL1 subtype appears to be particularly sensitive to chemotherapy regimens including platinum-based drugs.

[0006] Although many studies at home and abroad have used a variety of methods to classify the molecular subtypes of triple-negative breast cancer, due to the low incidence and heterogeneity of triple-negative breast cancer, the limited number of samples may lead to instability and unreliability of analysis results. In addition, the criteria for subtype classification may differ in different studies, resulting in incomparable results. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a method for predicting the subtype classification of triple-negative breast cancer.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0009] A prediction method for triple-negative breast cancer subtype classification, comprising:

[0010] S10, constructing a graph dataset;

[0011] S20, building a graph convolutional neural network prediction and classification model;

[0012] S30, training the graph data set as a graph convolutional neural network prediction classification model until the accuracy meets the requirements, and obtaining the best graph convolutional neural network prediction classification model;

[0013] S40, Application of the best graph convolutional neural network predictive classification model to predict triple-negative breast cancer subtype classification;

[0014] Among them, S20 builds a graph convolutional neural network prediction classification model in the following way:

[0015] S21, taking the adjacency matrix in the graph data set as the input node of the neural network, constructing the weight matrix. In each layer of the neural network, the input of the node is linearly transformed through the weight matrix to obtain the output features of the node;

[0016] S22, normalizing the adjacency matrix to obtain a normalized adjacency matrix;

[0017] S23, construct a graph convolutional network, and use the weight matrix and normalized adjacency matrix as the input of the graph convolutional network. The graph convolutional network uses the adjacency matrix to propagate the features of the nodes, so that each node can obtain the feature information of its adjacent nodes, and process the feature propagation of each node by sharing the weight matrix; and perform feature extraction on the graph data, calculate the weighted sum of the features of each node in the graph data and all its neighbors, output the new features of each node without changing the adjacency structure of the graph.

[0018] In one embodiment of the present invention, constructing a graph dataset includes:

[0019] S11, obtaining messenger RNA biological characteristic data and long non-coding RNA biological characteristic data; and these two biological characteristic data contain 6 TNBC molecular subtypes;

[0020] S12, screening the messenger RNA biological feature data and the long non-coding RNA biological feature data, obtaining lnc.FAM83H.AS1, gene HSPA14 and gene MGARP, and using them as feature vectors;

[0021] S13, discretizing the data value of each eigenvector into equal width to obtain a multi-interval eigenvector;

[0022] S14, mapping the characteristic values ​​of the case to the intervals corresponding to the multi-interval characteristic vectors, performing edge connections in each interval according to the characteristic values ​​of the case, establishing edge connections between cases in the same interval to indicate their similarity or correlation in the characteristic space, traversing all intervals to establish edge connections, and obtaining an adjacency matrix;

[0023] S15, the adjacency matrix corresponding to the three eigenvectors forms a graph dataset.

[0024] In one embodiment of the present invention, the interval division of the multi-interval feature vector is carried out in the following manner:

[0025] Bin i =[min+i×width, min+(i+1)×width];

[0026] In the formula, Bin i It represents the range of the i-th interval, min represents the minimum value of the interval of the feature vector before division, and width represents the width of each interval.

[0027] In one embodiment of the present invention, the output feature of the node is obtained by the following formula:

[0028] Z = X·W + b;

[0029] In the formula, X is the input data, which represents the nodes of the current layer; W is the weight matrix, which represents the connection weights between nodes; Z is the output of the node, and b is the bias.

[0030] In one embodiment of the present invention, the output Z of the node is used as the input of the next layer of nodes after passing through the activation function, thereby completing the forward propagation calculation; wherein, the activation function of the output Z of the node should be obtained by the following formula:

[0031]

[0032] Where σ(Z) is the activation function of the node’s output Z, and exp is the natural exponential.

[0033] In one embodiment of the present invention, the normalized adjacency matrix is ​​obtained by the following formula:

[0034]

[0035] In the formula, is the normalized adjacency matrix, D is the degree matrix of the adjacency matrix A, and A is the original adjacency matrix.

[0036] In one embodiment of the present invention, a new feature of each node is output without changing the adjacency structure of the graph, which is obtained by the following formula:

[0037]

[0038] In the formula, H (l+1) represents the output features of the l+1th layer, H (l) represents the output feature of the lth layer, σ is a nonlinear activation function, is the normalized adjacency matrix, W (l) is the weight matrix of the lth layer, is the normalized adjacency matrix The degree matrix of .

[0039] The present invention also provides a prediction system for triple-negative breast cancer subtype classification, which uses the above-mentioned prediction method for triple-negative breast cancer subtype classification, including:

[0040] Data module, used to build graph datasets;

[0041] Build a prediction model module to build a graph convolutional neural network prediction classification model;

[0042] The training module is used to train the graph dataset as a graph convolutional neural network prediction and classification model until the accuracy meets the requirements and obtain the best graph convolutional neural network prediction and classification model;

[0043] A prediction module for applying the best graph convolutional neural network prediction classification model to predict triple-negative breast cancer subtype classification;

[0044] Among them, the graph convolutional neural network prediction classification model is constructed in the following way:

[0045] S21, taking the adjacency matrix in the graph data set as the input node of the neural network, constructing the weight matrix. In each layer of the neural network, the input of the node is linearly transformed through the weight matrix to obtain the output features of the node;

[0046] S22, normalizing the adjacency matrix to obtain a normalized adjacency matrix;

[0047] S23, construct a graph convolutional network, and use the weight matrix and normalized adjacency matrix as the input of the graph convolutional network. The graph convolutional network uses the adjacency matrix to propagate the features of the nodes, so that each node can obtain the feature information of its adjacent nodes, and process the feature propagation of each node by sharing the weight matrix; and perform feature extraction on the graph data, calculate the weighted sum of the features of each node in the graph data and all its neighbors, output the new features of each node without changing the adjacency structure of the graph.

[0048] In one embodiment of the present invention, constructing a graph dataset includes:

[0049] Obtain messenger RNA biological characteristic data and long non-coding RNA biological characteristic data; and these two biological characteristic data contain 6 TNBC molecular subtypes;

[0050] From the messenger RNA biological feature data and the long non-coding RNA biological feature data, lnc.FAM83H.AS1, gene HSPA14 and gene MGARP were obtained and used as feature vectors;

[0051] Discretize the data value of each eigenvector with equal width to obtain multi-interval eigenvectors;

[0052] Map the characteristic values ​​of the cases to the intervals corresponding to the multi-interval characteristic vectors. In each interval, edge connections are made according to the characteristic values ​​of the cases. Edge connections are established between cases in the same interval to indicate their similarity or association in the characteristic space. All intervals are traversed to establish edge connections and obtain an adjacency matrix.

[0053] The adjacency matrix corresponding to the three eigenvectors forms the graph dataset.

[0054] In one embodiment of the present invention, a new feature of each node is output without changing the adjacency structure of the graph, which is obtained by the following formula:

[0055]

[0056] In the formula, H (l+1) represents the output features of the l+1th layer, H (l) represents the output feature of the lth layer, σ is a nonlinear activation function, is the normalized adjacency matrix, W (l) is the weight matrix of the lth layer, is the normalized adjacency matrix The degree matrix of .

[0057] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention uses a graph convolutional neural network to determine the subtype classification of triple-negative breast cancer through molecular biological characteristics, proposes a method for constructing an adjacency graph of cases, and classifies and verifies graph data constructed with different biological characteristics, providing a new method for determining the correlation between biological characteristics and diseases, and providing guidance for future research directions.

[0058] This paper evaluates the efficiency and stability of the graph convolutional neural network model by training different numbers of cases, providing an effective solution for the research of related work.

[0059] Graph Convolutional Neural Network (GCNN) introduces more complex neural network structures based on graph convolutional networks, such as multi-layer perceptrons with activation functions, to enhance the expression and learning capabilities of features. It can also learn effective feature representations from complex molecular data, including gene expression patterns, protein interactions and other information, which helps to discover features related to triple-negative breast cancer. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of a prediction method for triple-negative breast cancer subtype classification according to an embodiment of the present invention.

[0061] Figure 2 Schematic diagram of a graph convolutional neural network according to an embodiment of the present invention.

[0062] Figure 3 Schematic diagram of the ROC curve of an embodiment of the present invention.

[0063] Figure 4 This is a block diagram of a prediction system for triple-negative breast cancer subtype classification according to an embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to facilitate those skilled in the art to understand the technical solution of the present invention, the technical solution of the present invention is further described in conjunction with the accompanying drawings of the specification.

[0065] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0066] See also Figure 1 As shown, the present invention discloses a prediction method for triple-negative breast cancer subtype classification, comprising:

[0067] S10, construct a graph dataset.

[0068] In one embodiment of the present invention, constructing a graph dataset includes:

[0069] S11, obtain messenger RNA biological characteristic data and long non-coding RNA biological characteristic data; and these two biological characteristic data contain 6 TNBC molecular subtypes.

[0070] In this embodiment, the biological feature data comes from: The Cancer Genome Atlas (TCGA The CancerGenome Atlas), a project jointly initiated by the National Human Genome Research Institute (NHGRI) and the National Cancer Institute (NCI). The TCGA project provides a wide range of cancer data resources, including genomic data, epigenomic data, transcriptome data, etc. from patient samples, which are of great reference significance for researchers to conduct research and discover therapies in the field of cancer. The data used in this embodiment come from confirmed cases in the database, with 305 cases, including two types of biological features of messenger RNA (mRNA) and long non-coding RNA (LncRNA) data, and also includes 6 TNBC molecular subtypes, and the number of features of the selected data set is 1391.

[0071] S12, screening from messenger RNA biological feature data and long non-coding RNA biological feature data, obtaining lnc.FAM83H.AS1, gene HSPA14 and gene MGARP, and using them as feature vectors.

[0072] In this embodiment, constructing a graph convolutional neural network prediction classification model requires constructing a case relationship graph. Considering the number and sensitivity of classified cases, this embodiment uses three feature vectors of lnc.FAM83H.AS1, gene HSPA14 and gene MGARP to construct an adjacency matrix.

[0073] AM83H.AS1 ​​is a long noncoding RNA (lncRNA) that interacts with other molecules such as miRNA and proteins and participates in the regulation of cancer-related signaling pathways. The gene HSPA14 is a heat shock protein that has molecular chaperone and helicase activities in cells and participates in protein folding, assembly and disassembly. The gene MGARP is a transcription factor that participates in biological processes such as signal transduction in the cell nucleus and gene expression regulation.

[0074] S13, discretizing the data value of each eigenvector into equal-width, and obtaining a multi-interval eigenvector.

[0075] The data values ​​of the three feature vectors extracted in step S12 are discretized with equal width and divided into 11 regions on average. Equal width discretization is a method of dividing data into several intervals according to the numerical range. Assume that the maximum value of the original data interval is max, the minimum value is min, and it is divided into k intervals, and the width of each interval is width. The equal width discretization is expressed as:

[0076] Bin i =[min+i×width, min+(i+1)×width];

[0077] Where 0≤i <k,Bin i It represents the range of the i-th interval, min represents the minimum value of the interval of the feature vector before division, and width represents the width of each interval.

[0078] S14, mapping the characteristic values ​​of the cases to the intervals corresponding to the multi-interval characteristic vectors, performing edge connections in each interval according to the characteristic values ​​of the cases, establishing edge connections between cases in the same interval to indicate their similarity or association in the characteristic space, traversing all intervals to establish edge connections, and obtaining an adjacency matrix. Each adjacency matrix represents a case association network in a region.

[0079] In this embodiment, edge connections are established between cases in the same interval. For example, if the feature values ​​of case A and case B are similar or the distance is less than a certain threshold, an edge is established between them to indicate the similarity between them.

[0080] S15, the adjacency matrix corresponding to the three eigenvectors forms a graph dataset.

[0081] S20, build a graph convolutional neural network prediction and classification model.

[0082] See also Figure 1 and Figure 2 As shown, in one embodiment of the present invention, a graph convolutional neural network prediction classification model is constructed in the following manner:

[0083] S21, uses the adjacency matrix in the graph data set as the input node of the neural network, constructs a weight matrix, and in each layer of the neural network, the input of the node is linearly transformed through the weight matrix to obtain the output features of the node.

[0084] In this embodiment, the output feature of the node is obtained by the following formula:

[0085] Z = X·W + b;

[0086] In the formula, X is the input data, which represents the nodes of the current layer; W is the weight matrix, which represents the connection weights between nodes; Z is the output of the node, which is obtained by linear transformation of the weight matrix and the input data; b is the bias, which is used to adjust the output of the model.

[0087] In the propagation process of the neural network, the input data X of each node is matrix multiplied with the weight matrix W, and the bias term b is added to obtain the linear transformation output Z of the node. After the output Z passes through the Sigmoid activation function, it is used as the input of the next layer of nodes to complete the forward propagation calculation. The formula is as follows:

[0088]

[0089] Where σ(Z) is the activation function of the node’s output Z, and exp is the natural exponential. This process maps the input value Z to a continuous range between 0 and 1.

[0090] S22, normalizing the adjacency matrix to obtain a normalized adjacency matrix.

[0091] In this embodiment, the adjacency matrix A is used to represent the relationship between nodes, and it is symmetrically normalized to reduce the influence between different node degrees, making the network structure more comparable. The formula is as follows:

[0092]

[0093] In the formula, is the normalized adjacency matrix, D is the degree matrix of the adjacency matrix A, and A is the original adjacency matrix. This is done by dividing the adjacency matrix A by the square root of the node's degree.

[0094] In the present invention, each graph data consists of a weight matrix W and an adjacency matrix To represent, as the input data of the model.

[0095] S23, construct a graph convolutional network, and use the weight matrix and normalized adjacency matrix as the input of the graph convolutional network. The graph convolutional network uses the adjacency matrix to propagate the features of the nodes, so that each node can obtain the feature information of its adjacent nodes, and process the feature propagation of each node by sharing the weight matrix; and perform feature extraction on the graph data, calculate the weighted sum of the features of each node in the graph data and all its neighbors, output the new features of each node without changing the adjacency structure of the graph.

[0096] In one embodiment of the present invention, the local first-order approximation of the Chebyshev polynomial is used to connect the weight matrix and the link matrix, and the calculation process of the graph convolution kernel is simplified, thereby reducing the computational complexity of calculating the Laplacian matrix in the graph convolution. The formula for outputting new features of each node without changing the adjacency structure of the graph is as follows:

[0097]

[0098] In the formula, H (l+1) represents the output features of the l+1th layer, H (l) represents the output feature of the lth layer, σ is a nonlinear activation function, is the normalized adjacency matrix, W (l) is the weight matrix of the lth layer, is the normalized adjacency matrix The degree matrix of .

[0099] S30, training the graph dataset as a graph convolutional neural network prediction classification model until the accuracy meets the requirements, and obtaining the best graph convolutional neural network prediction classification model.

[0100] In one embodiment of the present invention, each layer of the model shares a parameter change matrix (weight matrix W and adjacency matrix A), and the dimension of this matrix is ​​independent of the number of vertices in the input graph. Therefore, this parameter change matrix is ​​shared by all vertices in the same layer, and can be understood as a parameter transformation for sharing the features of different vertices. In addition, the model uses a graph convolutional network with two hidden layers, and the embedding dimensions of each layer are 32 and 16 respectively. The model transforms and updates the input features through two graph convolutional layers, mapping the features to a 32-dimensional space, and then mapping them to a 16-dimensional space. Such a two-layer architecture can help the model better learn the representation and features of the data, improve the nonlinear modeling capabilities of classification tasks, and at the same time ensure the efficiency and accuracy of learning.

[0101] S40, Application of the optimal graph convolutional neural network predictive classification model to predict triple-negative breast cancer subtype classification.

[0102] See also Figures 1 to 3 As shown, in one embodiment of the present invention, the model performance is evaluated through the following three performance results (a), (b), and (c).

[0103] (a) Impact of different data segmentation on accuracy

[0104] The graph dataset is divided into training set and test set, and is split in five ways: 1:2, 1:3, 2:1, 3:1 and 1:1. After 200 epochs, the efficiency and accuracy of the model are verified. The results are shown in Table 1:

[0105] Table 1 The impact of segmentation of different data sets on accuracy

[0106] Split ratio Accuracy after 200 epochs 1∶1 75.1% 1∶2 71.4% 1∶3 64.8% 2∶1 89.2% 3∶1 89.3%

[0107] It can be seen from Table 1 that with the gradual increase of training cases, the accuracy of the model is also gradually improving. After reaching 200 cases, the accuracy of the model is basically stable. This shows that the graph convolutional neural network prediction classification model adopted in the present invention can converge quickly and has high efficiency.

[0108] (b) Effect of different training iterations on accuracy

[0109] From the results of (a), we can see that the data is trained and tested with a split ratio of 3:1, and the accuracy is improved by less than 0.01. The effect of different numbers of iterations on the training accuracy is shown in Table 2:

[0110] Table 2 Effect of different iteration times on accuracy

[0111] epoch Training Accuracy 1 40.7% 10 72.0% 20 85.3% 50 92.2% 100 93.3% 200 98.6%

[0112] It can be seen from Table 2 that the graph convolutional neural network prediction classification model can quickly optimize parameters and approach a local optimal solution in the early stage of training. As the iteration proceeds, after 100 iterations, the model convergence speed becomes relatively slow and the optimization speed is relatively slow. Therefore, it is a relatively reasonable choice to use different cases for multiple training to obtain a better model in the present invention.

[0113] (c) The impact of different features on accuracy

[0114] The experiment uses three features, lnc.FAM83H.AS1, gene HSPA14 and gene MGARP, to construct an adjacency matrix to test the influence of different features on the disease. Table 3 lists the influence of different feature vectors on training accuracy:

[0115] Table 3 Effect of different feature vectors on accuracy

[0116] feature Training Accuracy lnc.FAM83H.AS1 95.8% Gene HSPA14 92.1% MGARP 92.3%

[0117] As can be seen from Table 3, different feature vector methods have different effects on classification accuracy. The lnc.FAM83H.AS1 ​​feature shows a greater disease relevance, which means that this feature vector should be paid more attention than other feature vectors.

[0118] Determine the evaluation index of the model: The ROC curve is used as the index of model evaluation. Its main function is to reflect the comprehensive index of sensitivity and specificity continuous variables. The ROC curve uses the false positive rate (FPR) as the horizontal axis and the true positive rate (TPR, also known as the recall rate) as the vertical axis, showing the classification performance of the model under different thresholds.

[0119] In the ROC curve, the horizontal axis is FPR, and the calculation formula is:

[0120]

[0121] FP stands for False Positive, TN stands for True Negative; the vertical axis is TPR, and the calculation formula is:

[0122]

[0123] TP stands for True Positive and FN stands for False Negative.

[0124] The area under the ROC curve (AUC Area Under the Curve) is called the AUC value. The closer the AUC value is to 1, the better the model performance is, and the better it can distinguish positive and negative examples. When the AUC value is 0.5, it means that the graph convolutional neural network prediction classification model is equivalent to random guessing. The present invention uses accurate cases / (accurate cases + missed cases) to represent the true positive rate, and misjudged cases / (misjudged cases + missed cases) to approximately represent the false positive rate. The ROC curve is shown in the attached figure. Figure 3 As shown in the figure, it can be seen that the AUC of the model reaches between 0.8 and 0.9, which means that the graph convolutional neural network prediction classification model has a high classification accuracy.

[0125] See also Figures 1 to 4 As shown, the present invention also provides a prediction system for triple-negative breast cancer subtype classification, which uses the above-mentioned prediction method for triple-negative breast cancer subtype classification, including:

[0126] Data module, used to build graph datasets;

[0127] Build a prediction model module to build a graph convolutional neural network prediction classification model;

[0128] The training module is used to train the graph dataset as a graph convolutional neural network prediction and classification model until the accuracy meets the requirements and obtain the best graph convolutional neural network prediction and classification model;

[0129] A prediction module for applying the best graph convolutional neural network predictive classification model to predict triple-negative breast cancer subtype classification.

[0130] Among them, the graph convolutional neural network prediction classification model is constructed in the following way:

[0131] S21, taking the adjacency matrix in the graph data set as the input node of the neural network, constructing the weight matrix. In each layer of the neural network, the input of the node is linearly transformed through the weight matrix to obtain the output features of the node;

[0132] S22, normalizing the adjacency matrix to obtain a normalized adjacency matrix;

[0133] S23, construct a graph convolutional network, and use the weight matrix and normalized adjacency matrix as the input of the graph convolutional network. The graph convolutional network uses the adjacency matrix to propagate the features of the nodes, so that each node can obtain the feature information of its adjacent nodes, and process the feature propagation of each node by sharing the weight matrix; and perform feature extraction on the graph data, calculate the weighted sum of the features of each node in the graph data and all its neighbors, output the new features of each node without changing the adjacency structure of the graph.

[0134] In this embodiment, constructing a graph dataset includes:

[0135] Obtain messenger RNA biological characteristic data and long non-coding RNA biological characteristic data; and these two biological characteristic data contain 6 TNBC molecular subtypes;

[0136] From the messenger RNA biological feature data and the long non-coding RNA biological feature data, lnc.FAM83H.AS1, gene HSPA14 and gene MGARP were obtained and used as feature vectors;

[0137] Discretize the data value of each eigenvector with equal width to obtain multi-interval eigenvectors;

[0138] Map the characteristic values ​​of the cases to the intervals corresponding to the multi-interval characteristic vectors. In each interval, edge connections are made according to the characteristic values ​​of the cases. Edge connections are established between cases in the same interval to indicate their similarity or association in the characteristic space. All intervals are traversed to establish edge connections and obtain an adjacency matrix.

[0139] The adjacency matrix corresponding to the three eigenvectors forms the graph dataset.

[0140] In this embodiment, the new features of each node are output without changing the adjacency structure of the graph, which is obtained by the following formula:

[0141]

[0142] In the formula, H (l+1) represents the output features of the l+1th layer, H (l) represents the output feature of the lth layer, σ is a nonlinear activation function, is the normalized adjacency matrix, W (l) is the weight matrix of the lth layer, is the normalized adjacency matrix The degree matrix of .

[0143] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting from any point of view, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

[0144] The above-described embodiments merely represent implementation methods of the invention. The protection scope of the present invention is not limited to the above-described embodiments. For those skilled in the art, several modifications and improvements may be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A prediction method for triple-negative breast cancer subtype classification, characterized in that: include: S10, constructing a graph dataset; S20, building a graph convolutional neural network prediction and classification model; S30, training the graph data set as a graph convolutional neural network prediction classification model until the accuracy meets the requirements, and obtaining the best graph convolutional neural network prediction classification model; S40, Application of the best graph convolutional neural network predictive classification model to predict triple-negative breast cancer subtype classification; Among them, S20 builds a graph convolutional neural network prediction classification model in the following way: S21, taking the adjacency matrix in the graph data set as the input node of the neural network, constructing the weight matrix. In each layer of the neural network, the input of the node is linearly transformed through the weight matrix to obtain the output features of the node; S22, normalizing the adjacency matrix to obtain a normalized adjacency matrix; S23, construct a graph convolutional network, and use the weight matrix and normalized adjacency matrix as the input of the graph convolutional network. The graph convolutional network uses the adjacency matrix to propagate the features of the nodes, so that each node can obtain the feature information of its adjacent nodes, and process the feature propagation of each node by sharing the weight matrix; and perform feature extraction on the graph data, calculate the weighted sum of the features of each node in the graph data and all its neighbors, output the new features of each node and do not change the adjacency structure of the graph.

2. The method for predicting triple-negative breast cancer subtype classification according to claim 1, characterized in that: Construct a graph dataset, including: S11, obtaining messenger RNA biological characteristic data and long non-coding RNA biological characteristic data; and these two biological characteristic data contain 6 TNBC molecular subtypes; S12, screening the messenger RNA biological feature data and the long non-coding RNA biological feature data, obtaining lnc.FAM83H.AS1, gene HSPA14 and gene MGARP, and using them as feature vectors; S13, discretizing the data value of each eigenvector into equal width to obtain a multi-interval eigenvector; S14, mapping the characteristic values ​​of the case to the intervals corresponding to the multi-interval characteristic vectors, performing edge connections in each interval according to the characteristic values ​​of the case, establishing edge connections between cases in the same interval to indicate their similarity or correlation in the characteristic space, traversing all intervals to establish edge connections, and obtaining an adjacency matrix; S15, the adjacency matrix corresponding to the three eigenvectors forms a graph dataset.

3. The method for predicting triple-negative breast cancer subtype classification according to claim 2, characterized in that: The interval partitioning of multi-interval feature vectors is done in the following way: Bees i =[min+i×width,min+(i+1)×width]; In the formula, Bin i It represents the range of the i-th interval, min represents the minimum value of the interval of the feature vector before division, and width represents the width of each interval.

4. The method for predicting triple-negative breast cancer subtype classification according to claim 1, characterized in that: Get the output features of the node using the following formula: Z = X·W + b; In the formula, X is the input data, which represents the nodes of the current layer; W is the weight matrix, which represents the connection weights between nodes; Z is the output of the node and b is the bias.

5. The method for predicting triple-negative breast cancer subtype classification according to claim 4, characterized in that: The output Z of the node passes through the activation function and is used as the input of the next layer of nodes to complete the forward propagation calculation; the activation function of the node output Z should be obtained by the following formula: Where σ(Z) is the activation function of the node’s output Z, and exp is the natural exponential.

6. The method for predicting triple-negative breast cancer subtype classification according to claim 1, characterized in that: The normalized adjacency matrix is ​​obtained by the following formula: In the formula, is the normalized adjacency matrix, D is the degree matrix of the adjacency matrix A, and A is the original adjacency matrix.

7. The method for predicting triple-negative breast cancer subtype classification according to claim 1, characterized in that: Output the new features of each node without changing the adjacency structure of the graph, obtained by the following formula: In the formula, H (l+1) represents the output features of the l+1th layer, H (l) represents the output feature of the lth layer, σ is a nonlinear activation function, is the normalized adjacency matrix, W (l) is the weight matrix of the lth layer, is the normalized adjacency matrix The degree matrix of .

8. A prediction system for triple-negative breast cancer subtype classification, using the prediction method for triple-negative breast cancer subtype classification according to any one of claims 1 to 7, characterized in that: include: Data module, used to build graph datasets; Build a prediction model module to build a graph convolutional neural network prediction classification model; The training module is used to train the graph dataset as a graph convolutional neural network prediction and classification model until the accuracy meets the requirements and obtain the best graph convolutional neural network prediction and classification model; A prediction module for applying the best graph convolutional neural network prediction classification model to predict triple-negative breast cancer subtype classification; Among them, the graph convolutional neural network prediction classification model is constructed in the following way: S21, taking the adjacency matrix in the graph data set as the input node of the neural network, constructing the weight matrix. In each layer of the neural network, the input of the node is linearly transformed through the weight matrix to obtain the output features of the node; S22, normalizing the adjacency matrix to obtain a normalized adjacency matrix; S23, construct a graph convolutional network, and use the weight matrix and normalized adjacency matrix as the input of the graph convolutional network. The graph convolutional network uses the adjacency matrix to propagate the features of the nodes, so that each node can obtain the feature information of its adjacent nodes, and process the feature propagation of each node by sharing the weight matrix; and perform feature extraction on the graph data, calculate the weighted sum of the features of each node in the graph data and all its neighbors, output the new features of each node and do not change the adjacency structure of the graph.

9. The prediction system for triple-negative breast cancer subtype classification according to claim 8, characterized in that: Construct a graph dataset, including: Obtain messenger RNA biological characteristic data and long non-coding RNA biological characteristic data; and these two biological characteristic data contain 6 TNBC molecular subtypes; From the messenger RNA biological feature data and the long non-coding RNA biological feature data, lnc.FAM83H.AS1, gene HSPA14 and gene MGARP were obtained and used as feature vectors; Discretize the data value of each eigenvector with equal width to obtain multi-interval eigenvectors; Map the characteristic values ​​of the cases to the intervals corresponding to the multi-interval characteristic vectors. In each interval, edge connections are made according to the characteristic values ​​of the cases. Edge connections are established between cases in the same interval to indicate their similarity or association in the characteristic space. All intervals are traversed to establish edge connections and obtain an adjacency matrix. The adjacency matrix corresponding to the three eigenvectors forms the graph dataset.

10. The prediction system for triple-negative breast cancer subtype classification according to claim 9, characterized in that: Output the new features of each node without changing the adjacency structure of the graph, obtained by the following formula: In the formula, H (l+1) represents the output features of the l+1th layer, H (l) represents the output feature of the lth layer, σ is a nonlinear activation function, is the normalized adjacency matrix, W (l) is the weight matrix of the lth layer, is the normalized adjacency matrix The degree matrix of .

Citation Information

Patent Citations

  • Underwater target feature extraction method based on convolutional neural network (CNN)

    CN107194404A

  • Breast cancer subtype classification method based on convolutional neural network

    CN113283477A

  • Method for expressing quantitative character and CNV association based on gene interaction network clustering and group sparse learning

    CN114038502A

  • Method for predicting personalized cancer driver gene by fusing gene features and graph convolution

    CN116741269A

  • Breast cancer subtype classification method and system based on graph convolutional neural network

    CN116959572A