A breast cancer typing prediction method and system based on LncRNA and immune cell characteristics

Abstract

The application provides a breast cancer typing prediction method and system based on LncRNA and immune cell characteristics, and the method comprises the following steps: obtaining a first RNA sequencing data set and a second RNA sequencing data set; inputting the first RNA sequencing data set into a preset LncRNA clustering model, so that the LncRNA clustering model clusters each RNA sequencing data in the first RNA sequencing data set into different LncRNA subtypes; inputting the first RNA sequencing data set into a preset immune cell feature clustering model, so that the immune cell feature clustering model clusters each RNA sequencing data in the first RNA sequencing data set into different immune subtypes; combining the various LncRNA subtypes and the various immune subtypes to obtain several different breast cancer types; inputting the pathological section image of a patient to be predicted into a preset typing prediction model, so that the typing prediction model generates a corresponding breast cancer typing prediction result, and the accuracy of the efficacy prediction and prognosis prediction of immunotherapy is improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for predicting breast cancer subtyping based on LncRNA and immune cell characteristics. Background Technology

[0002] Breast cancer is a common malignant tumor, and its clinical classification is crucial for developing effective treatment plans and predicting patient prognosis. The rapid development of immunotherapy has brought new possibilities to breast cancer treatment, especially the application of immune checkpoint inhibitors (ICIs). Immunotherapy activates the patient's own immune system to fight tumor cells, thereby inducing a durable anti-tumor response. Currently, based on different biomarker expression levels, breast cancer can be divided into several subtypes, and these subtypes may exhibit different responses to immunotherapy. Below are some common clinical classifications of breast cancer and their corresponding classification methods or models:

[0003] (1) Hormone receptor status (ER, PR, HER2): Breast cancer is often classified according to the expression of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2). This classification is not only helpful in determining treatment plans, but can also be related to the efficacy of immunotherapy.

[0004] (2) Immune cell infiltration: The effectiveness of immunotherapy is closely related to the infiltration of immune cells around the tumor. Some typing methods focus on the type and number of immune cells, such as lymphocytes and plasma cells, to help predict a patient's response to immunotherapy.

[0005] (3) PD-L1 expression: Programmed cell death ligand 1 (PD-L1) is an immune checkpoint molecule, and its overexpression may affect the efficacy of immunotherapy. Some typing methods focus on assessing the expression level of PD-L1 on the surface of tumor cells or immune cells as a biomarker for predicting immunotherapy response.

[0006] (4) Neoantigen load: Neoantigens are antigens expressed on the surface of tumor cells due to mutations. Some subtyping models focus on tumor neoantigen load because a high neoantigen load may be associated with a stronger immune response and a better prognosis.

[0007] However, current biomarkers do not achieve satisfactory accuracy. PD-L1 expression is a stratification factor or patient inclusion criterion in clinical trials and an indication for the use of ICIs in certain cancer types in treatment guidelines. However, clinical trials have also found efficacy of ICIs in patients without PD-L1 expression. Furthermore, only a small percentage of patients respond to these therapies. Therefore, current classification methods are not precise enough for predicting patient prognosis. Meanwhile, most existing breast cancer classification diagnoses are based on immunohistochemical (IHC) results. IHC has advantages such as high specificity, high sensitivity, simplicity, speed, relatively low cost, and good reproducibility; however, it is not comprehensive enough, and results may differ between different laboratories. Therefore, a classification prediction model combined with artificial intelligence technology is needed to predict breast cancer subtypes, assisting medical professionals in breast cancer classification diagnosis and treatment planning, and to overcome the limitations of immunohistochemical methods to some extent. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention provides a method and system for predicting breast cancer subtyping based on LncRNA and immune cell characteristics. Since the obtained breast cancer subtyping prediction results indicate the patient's suitability for immunotherapy, this invention improves the accuracy of predicting the efficacy and prognosis of immunotherapy for patients, providing data support for the formulation of tumor treatment plans.

[0009] In a first aspect, the present invention provides a method for predicting breast cancer subtyping based on LncRNA and immune cell characteristics, comprising:

[0010] Obtain first RNA sequencing datasets from several historical breast cancer patients and second RNA sequencing datasets from several patients who received immunotherapy;

[0011] The first RNA sequencing dataset is input into a preset LncRNA clustering model so that the LncRNA clustering model clusters each RNA sequencing data in the first RNA sequencing dataset into different LncRNA subtypes. The LncRNA clustering model is constructed based on the feature set of the second RNA sequencing dataset.

[0012] The first RNA sequencing dataset is input into a preset immune cell feature clustering model so that the immune cell feature clustering model clusters each RNA sequencing data in the first RNA sequencing dataset into different immune subtypes. The immune cell feature clustering model is constructed based on each immune cell in the second RNA sequencing dataset and the corresponding ssGSEA score.

[0013] By combining the various LncRNA subtypes and the various immune subtypes, several different breast cancer subtypes are obtained;

[0014] The pathological slide image of the patient to be predicted is input into a preset classification prediction model so that the classification prediction model generates a corresponding breast cancer classification prediction result. The breast cancer classification prediction result is one of the several different breast cancer classifications. The classification prediction model is constructed based on the pathological slide data of historical breast cancer patients and a pre-trained convolutional neural network model.

[0015] This invention provides a method for predicting breast cancer subtyping based on lncRNA and immune cell characteristics. First, an lncRNA clustering model and an immune cell characteristic clustering model are constructed based on a second RNA sequencing dataset to identify lncRNA subtypes and immune subtypes in the first RNA sequencing dataset. Then, by combining each lncRNA subtype and each immune subtype, several breast cancer subtypes are constructed. The breast cancer subtypes constructed in this embodiment fully consider the influence of long non-coding RNAs and immune cells on the efficacy of tumor immunotherapy, and reconstruct a breast cancer subtype that is more sensitive to immunotherapy. This breast cancer subtype has better sensitivity and accuracy in predicting the efficacy of immunotherapy than traditional indicators. Furthermore, in conjunction with the breast cancer subtypes, this embodiment also provides a subtype prediction model based on pathological images. This avoids the gene sequencing required for traditional tumor subtyping, and subtyping prediction can be performed using only clinical pathological slides, improving the convenience and accuracy of breast cancer subtyping prediction, lowering the clinical application threshold of the breast cancer subtypes, and providing decision-making basis for medical professionals to formulate cancer treatment plans.

[0016] In one possible implementation, constructing the LncRNA clustering model based on the feature set of the second RNA sequencing dataset includes:

[0017] The corresponding LncRNA expression dataset was obtained by transcription from the second RNA sequencing dataset;

[0018] Based on the magnitude of the influence factors of each feature in the LncRNA expression dataset on immunotherapy, several LncRNA features are extracted from the LncRNA expression dataset and combined to form an LncRNA feature set;

[0019] A clustering model of LncRNAs was constructed based on the LncRNA feature set.

[0020] This invention provides a method for constructing a LncRNA clustering model. Since the patients in the second RNA sequencing dataset are those receiving immunotherapy, features related to immunotherapy can be extracted from the second RNA sequencing dataset, and these features are then used to construct a clustering model. Specifically, the second RNA sequencing dataset is first transcribed into an LncRNA expression dataset. Several LncRNAs affecting efficacy differences are then selected from the LncRNA expression dataset; these are the LncRNA features, combined to form an LncRNA feature set. Finally, an LncRNA clustering model is constructed based on this feature set. This embodiment constructs an LncRNA clustering model using the second RNA sequencing dataset, improving the accuracy of subsequent clustering. Furthermore, the constructed LncRNA clustering model is an unsupervised machine learning model, requiring no human intervention or adjustment. It can automatically cluster the input dataset, generating multiple different LncRNA subtypes, thus improving the efficiency of LncRNA subtype generation.

[0021] In one possible implementation, the step of constructing the immune cell feature clustering model based on each immune cell in the second RNA sequencing dataset and its corresponding ssGSEA score includes:

[0022] The second RNA sequencing dataset is input into a preset bioinformatics analysis software so that the bioinformatics analysis software can determine several types of immune cells and their corresponding ssGSEA scores through ssGSEA analysis.

[0023] An immune cell feature clustering model was constructed based on the various immune cells and their corresponding ssGSEA scores.

[0024] This invention provides a method for constructing an immune cell feature clustering model, which is also based on the second RNA sequencing dataset of patients who have received immunotherapy. First, ssGSEA analysis is performed on the second RNA sequencing dataset using pre-set bioinformatics analysis software. This analysis identifies several types of immune cells and their corresponding ssGSEA scores. The ssGSEA scores of immune cells also reveal their impact on immunotherapy. Therefore, an immune cell feature clustering model is constructed based on these immune cells and their corresponding ssGSEA scores. This embodiment, based on the second RNA sequencing dataset, constructs an immune cell feature clustering model from the perspective of immune cells and combines it with the ssGSEA scores of each immune cell, improving the accuracy and rationality of subsequent clustering.

[0025] Furthermore, by combining the various LncRNA subtypes and the various immune subtypes, several different breast cancer subtypes are obtained, including:

[0026] Each LncRNA subtype and each immune subtype are combined in pairs to generate a first number of LncRNA-immune subtypes;

[0027] The LncRNA-immune subtypes were enriched and scored using GSEA metabolic pathway analysis to obtain the corresponding pathway enrichment scores.

[0028] Based on the enrichment scores of each pathway, the metabolic pathway enrichment characteristics of each LncRNA-immune subtype are determined. LncRNA-immune subtypes with the same metabolic pathway enrichment characteristics are merged to obtain a second number of breast cancer subtypes, wherein the second number is less than or equal to the first number.

[0029] In this embodiment of the invention, after obtaining each LncRNA subtype and each immune subtype, they are further merged into a two-dimensional index to generate a first number of LncRNA-immune subtypes. Among the various LncRNA-immune subtypes, there may be LncRNA-immune subtypes with similar therapeutic effects on immunotherapy. To improve the accuracy and efficiency of subsequent classification and prediction, LncRNA-immune subtypes with similar therapeutic effects on immunotherapy need to be merged into the same type. Therefore, this embodiment uses metabolic pathway enrichment characteristics as the evaluation index for each LncRNA-immune subtype, merging LncRNA-immune subtypes with the same metabolic pathway enrichment characteristics into the same breast cancer subtype, ultimately obtaining a second number of breast cancer subtypes. The breast cancer subtypes obtained in this embodiment significantly distinguish different immunotherapy efficacy and prognostic survival time, and reveal the metabolic connotation of each breast cancer subtype, providing data support for the formulation of subsequent tumor treatment plans.

[0030] In one possible implementation, the subtyping prediction model generates a corresponding breast cancer subtyping prediction result based on the pathological slide images of the patient to be predicted, including:

[0031] The pathological slide image is segmented based on the tissue regions in the pathological slide image to obtain several image blocks;

[0032] A convolutional neural network model with pre-trained parameters extracts several feature vectors from each of the image patches;

[0033] Based on the feature vectors, the image patches are sorted using a multi-instance learning algorithm based on an attention mechanism, and a corresponding attention score is assigned to each image patch.

[0034] Based on the attention scores, the image information of each image block is aggregated to generate aggregated information;

[0035] Based on the aggregated information, the convolutional neural network model is used to predict breast cancer subtypes in the pathological slice images through a multi-branch network architecture and multi-task objectives, generating breast cancer subtype prediction results for the patients to be predicted.

[0036] This invention provides a method for predicting breast cancer subtyping based on pathological slide images, using digital high-resolution histological slides as input to the subtyping prediction model. For each pathological slide image, the tissue content is automatically segmented, dividing it into thousands to tens of thousands of small image patches. Then, a pre-trained convolutional neural network model extracts corresponding feature vectors from each image patch. Based on the feature vectors, the image patches are sorted, and the image information of each image patch is aggregated according to their relative importance, i.e., attention scores, to obtain aggregated information for the entire pathological slide image. Finally, breast cancer subtyping prediction is performed based on the aggregated information, generating the predicted breast cancer subtyping result for the patient to be predicted. This achieves breast cancer subtyping diagnosis based on digital pathological slide features, avoiding the gene sequencing required for traditional tumor subtyping, and enabling subtyping prediction solely based on clinical pathological slides, thus improving the convenience and accuracy of breast cancer subtyping prediction.

[0037] Secondly, correspondingly, the present invention provides a breast cancer subtyping prediction system based on LncRNA and immune cell characteristics, including an acquisition module, an LncRNA subtype generation module, an immune subtype generation module, a breast cancer subtyping generation module, and a prediction module.

[0038] The acquisition module is used to acquire first RNA sequencing datasets from several historical breast cancer patients and second RNA sequencing datasets from several patients who received immunotherapy.

[0039] The LncRNA subtype generation module is used to input the first RNA sequencing dataset into a preset LncRNA clustering model, so that the LncRNA clustering model clusters each RNA sequencing data in the first RNA sequencing dataset into different LncRNA subtypes, wherein the LncRNA clustering model is constructed based on the feature set of the second RNA sequencing dataset;

[0040] The immune subtype generation module is used to input the first RNA sequencing dataset into a preset immune cell feature clustering model, so that the immune cell feature clustering model clusters each RNA sequencing data in the first RNA sequencing dataset into different immune subtypes. The immune cell feature clustering model is constructed based on each immune cell in the second RNA sequencing dataset and the corresponding ssGSEA score.

[0041] The breast cancer typing generation module is used to combine the various LncRNA subtypes and the various immune subtypes to obtain several different breast cancer typings.

[0042] The prediction module is used to input the pathological slide image of the patient to be predicted into a preset subtyping prediction model, so that the subtyping prediction model generates a corresponding breast cancer subtyping prediction result. The breast cancer subtyping prediction result is one of the several different breast cancer subtypings. The subtyping prediction model is constructed based on the pathological slide data of historical breast cancer patients and a pre-trained convolutional neural network model.

[0043] In one possible implementation, constructing the LncRNA clustering model based on the feature set of the second RNA sequencing dataset includes:

[0044] The corresponding LncRNA expression dataset was obtained by transcription from the second RNA sequencing dataset;

[0045] Based on the magnitude of the influence factors of each feature in the LncRNA expression dataset on immunotherapy, several LncRNA features are extracted from the LncRNA expression dataset and combined to form an LncRNA feature set;

[0046] A clustering model of LncRNAs was constructed based on the LncRNA feature set.

[0047] In one possible implementation, the step of constructing the immune cell feature clustering model based on each immune cell in the second RNA sequencing dataset and its corresponding ssGSEA score includes:

[0048] The second RNA sequencing dataset is input into a preset bioinformatics analysis software so that the bioinformatics analysis software can determine several types of immune cells and their corresponding ssGSEA scores through ssGSEA analysis.

[0049] An immune cell feature clustering model was constructed based on the various immune cells and their corresponding ssGSEA scores.

[0050] Furthermore, the breast cancer subtyping generation module combines the various LncRNA subtypes and the various immune subtypes to obtain several different breast cancer subtypes, including:

[0051] Each LncRNA subtype and each immune subtype are combined in pairs to generate a first number of LncRNA-immune subtypes;

[0052] The LncRNA-immune subtypes were enriched and scored using GSEA metabolic pathway analysis to obtain the corresponding pathway enrichment scores.

[0053] Based on the enrichment scores of each pathway, the metabolic pathway enrichment characteristics of each LncRNA-immune subtype are determined. LncRNA-immune subtypes with the same metabolic pathway enrichment characteristics are merged to obtain a second number of breast cancer subtypes, wherein the second number is less than or equal to the first number.

[0054] In one possible implementation, the prediction module includes an image segmentation unit, a feature extraction unit, a sorting unit, an information aggregation unit, and a prediction unit.

[0055] The image segmentation unit is used to segment the pathological slide image according to the tissue region in the pathological slide image to obtain several image blocks;

[0056] The feature extraction unit is used to extract several feature vectors from each image patch using a convolutional neural network model with pre-trained parameters;

[0057] The sorting unit is used to sort the image patches according to the feature vector using a multi-instance learning algorithm based on an attention mechanism, and to assign a corresponding attention score to each image patch.

[0058] The information aggregation unit is used to aggregate the image information of each image block according to the attention score, and generate aggregated information;

[0059] The prediction unit is used to predict breast cancer subtypes of the pathological slide images based on the aggregated information, through a multi-branch network architecture and multi-task objectives, using the convolutional neural network model, and to generate breast cancer subtype prediction results for the patients to be predicted. Attached Figure Description

[0060] Figure 1 : A schematic flowchart of an embodiment of a breast cancer subtyping prediction method based on LncRNA and immune cell characteristics provided by the present invention.

[0061] Figure 2 This is a schematic diagram of LncRNA subtype clustering results, representing an embodiment of a breast cancer subtyping prediction method based on LncRNA and immune cell characteristics provided by the present invention.

[0062] Figure 3 This is a schematic diagram of the prognostic survival curves of different LncRNA subtypes in an embodiment of a breast cancer subtype prediction method based on LncRNA and immune cell characteristics provided by the present invention.

[0063] Figure 4This is a schematic diagram of the immune subtype clustering results of an embodiment of a breast cancer subtype prediction method based on LncRNA and immune cell characteristics provided by the present invention.

[0064] Figure 5 This is a schematic diagram of the prognostic survival curves for different immune subtypes in an embodiment of a breast cancer subtype prediction method based on LncRNA and immune cell characteristics provided by the present invention.

[0065] Figure 6 : A schematic diagram of the metabolic pathway enrichment features of four breast cancer subtypes in an embodiment of a breast cancer subtype prediction method based on LncRNA and immune cell characteristics provided by the present invention.

[0066] Figure 7 This is a schematic diagram illustrating the process of constructing a breast cancer subtyping prediction method based on LncRNA and immune cell characteristics, as provided by the present invention.

[0067] Figure 8 This is a schematic diagram comparing the overall survival of patients with four different breast cancer subtypes treated with immunotherapy, as part of an embodiment of a breast cancer subtype prediction method based on LncRNA and immune cell characteristics provided by the present invention.

[0068] Figure 9 This is a schematic diagram illustrating the breast cancer subtyping prediction process of one embodiment of a breast cancer subtyping prediction method based on LncRNA and immune cell characteristics provided by the present invention.

[0069] Figure 10 This is a schematic diagram illustrating the construction process of a breast cancer typing prediction model, which is an embodiment of the breast cancer typing prediction method based on LncRNA and immune cell characteristics provided by the present invention.

[0070] Figure 11 : A schematic diagram of an embodiment of a breast cancer subtyping prediction system based on LncRNA and immune cell characteristics provided by the present invention.

[0071] Figure 12 : A schematic diagram of the prediction module of an embodiment of a breast cancer subtyping prediction system based on LncRNA and immune cell characteristics provided by the present invention. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] It should be noted that the step numbers in this document are only for the convenience of explaining the specific embodiments and are not intended to limit the order in which the steps are performed. In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0074] Throughout this specification, lncRNA refers to long non-coding RNA (lncRNA), which is a non-coding RNA longer than 200 nucleotides. Studies have shown that lncRNAs play crucial roles in numerous life processes, including dose compensation effects, epigenetic regulation, cell cycle regulation, and cell differentiation regulation. Abnormal expression or function of lncRNAs is closely related to the occurrence of human diseases, including several serious diseases that severely threaten human health, such as cancer and degenerative neurological diseases. These abnormalities manifest as abnormalities in the sequence and spatial structure of lncRNAs, abnormal expression levels, and abnormal interactions with binding proteins.

[0075] Example 1:

[0076] like Figure 1 As shown, Example 1 provides a breast cancer subtyping prediction method based on LncRNA and immune cell characteristics, including steps S1-S5:

[0077] Step S1: Obtain first RNA sequencing datasets from several historical breast cancer patients and second RNA sequencing datasets from several patients who received immunotherapy;

[0078] Step S2: Input the first RNA sequencing dataset into a preset LncRNA clustering model so that the LncRNA clustering model clusters each RNA sequencing data in the first RNA sequencing dataset into different LncRNA subtypes. The LncRNA clustering model is constructed based on the feature set of the second RNA sequencing dataset.

[0079] Step S3: Input the first RNA sequencing dataset into a preset immune cell feature clustering model so that the immune cell feature clustering model clusters each RNA sequencing data in the first RNA sequencing dataset into different immune subtypes. The immune cell feature clustering model is constructed based on each immune cell in the second RNA sequencing dataset and the corresponding ssGSEA score.

[0080] Step S4: Combining the various LncRNA subtypes and the various immune subtypes, several different breast cancer subtypes are obtained;

[0081] Step S5: Input the pathological slide image of the patient to be predicted into the preset classification prediction model so that the classification prediction model generates the corresponding breast cancer classification prediction result. The breast cancer classification prediction result is one of the several different breast cancer classifications. The classification prediction model is constructed based on the pathological slide data of historical breast cancer patients and a pre-trained convolutional neural network model.

[0082] This invention provides a method for predicting breast cancer subtyping based on lncRNA and immune cell characteristics. First, an lncRNA clustering model and an immune cell characteristic clustering model are constructed based on a second RNA sequencing dataset to identify lncRNA subtypes and immune subtypes in the first RNA sequencing dataset. Then, by combining each lncRNA subtype and each immune subtype, several breast cancer subtypes are constructed. The breast cancer subtypes constructed in this embodiment fully consider the influence of long non-coding RNAs and immune cells on the efficacy of tumor immunotherapy, and reconstruct a breast cancer subtype that is more sensitive to immunotherapy. This breast cancer subtype has better sensitivity and accuracy in predicting the efficacy of immunotherapy than traditional indicators. Furthermore, in conjunction with the breast cancer subtypes, this embodiment also provides a subtype prediction model based on pathological images. This avoids the gene sequencing required for traditional tumor subtyping, and subtyping prediction can be performed using only clinical pathological slides, improving the convenience and accuracy of breast cancer subtyping prediction, lowering the clinical application threshold of the breast cancer subtypes, and providing decision-making basis for medical professionals to formulate cancer treatment plans.

[0083] In a preferred embodiment, the first RNA sequencing dataset of the plurality of historical breast cancer patients mentioned in step S1 comes from the RNA sequencing data of 925 breast cancer patients from The Cancer Genome Atlas (TCGA), and the second RNA sequencing dataset of the plurality of patients who received immunotherapy comes from Sun Yat-sen Memorial Hospital (SYSMH), specifically from the RNA sequencing data of 83 breast cancer patients (another 7 patients received chemotherapy combined with immunotherapy) who were treated at Sun Yat-sen Memorial Hospital between September 2019 and February 2022.

[0084] In one possible implementation, step S2, which involves constructing the LncRNA clustering model based on the feature set of the second RNA sequencing dataset, includes:

[0085] The corresponding LncRNA expression dataset was obtained by transcription from the second RNA sequencing dataset;

[0086] Based on the magnitude of the influence factors of each feature in the LncRNA expression dataset on immunotherapy, several LncRNA features are extracted from the LncRNA expression dataset and combined to form an LncRNA feature set;

[0087] A clustering model of LncRNAs was constructed based on the LncRNA feature set.

[0088] In a preferred embodiment, 198 LncRNAs affecting efficacy differences were screened from transcriptome sequencing LncRNA expression data of 7 breast cancer patients (responsive vs. non-responsive: 3 vs. 4) at Sun Yat-sen Memorial Hospital of Sun Yat-sen University who received chemotherapy combined with immunotherapy. A deep learning-based LncRNA clustering model was then constructed based on these 198 LncRNAs. Further, unsupervised subtype clustering was performed on 925 breast cancer samples from the first RNA sequencing dataset using the LncRNA clustering model to construct LncRNA-based breast cancer immune subtypes. To ensure classification stability, 1000 iterations and an 80% resampling rate were performed, and the cumulative distribution function curve was used to determine the number of subtypes. The results showed 369, 334, and 222 patients in three different subtypes (LncRNA cluster 1, LncRNA-cluster 2, and LncRNA-cluster 3), respectively. The clustering results are shown below. Figure 2 As shown.

[0089] Figure 3 The prognostic survival curves for patients with three subtypes of immunotherapy-related LncRNAs were presented. Survival comparisons showed significant differences among the three clusters (log-rank test, P<0.001). Patients in LncRNA-cluster 1 had the best overall survival, while patients in LncRNA-cluster 3 had the worst prognosis.

[0090] This invention provides a method for constructing a LncRNA clustering model. Since the patients in the second RNA sequencing dataset are those receiving immunotherapy, features related to immunotherapy can be extracted from the second RNA sequencing dataset, and these features are then used to construct a clustering model. Specifically, the second RNA sequencing dataset is first transcribed into an LncRNA expression dataset. Several LncRNAs affecting efficacy differences are then selected from the LncRNA expression dataset; these are the LncRNA features, combined to form an LncRNA feature set. Finally, an LncRNA clustering model is constructed based on this feature set. This embodiment constructs an LncRNA clustering model using the second RNA sequencing dataset, improving the accuracy of subsequent clustering. Furthermore, the constructed LncRNA clustering model is an unsupervised machine learning model, requiring no human intervention or adjustment. It can automatically cluster the input dataset, generating multiple different LncRNA subtypes, thus improving the efficiency of LncRNA subtype generation.

[0091] In one possible implementation, step S3, which involves constructing the immune cell feature clustering model based on each immune cell in the second RNA sequencing dataset and its corresponding ssGSEA score, includes:

[0092] The second RNA sequencing dataset is input into a preset bioinformatics analysis software so that the bioinformatics analysis software can determine several types of immune cells and their corresponding ssGSEA scores through ssGSEA analysis.

[0093] An immune cell feature clustering model was constructed based on the various immune cells and their corresponding ssGSEA scores.

[0094] In a preferred embodiment, using the second RNA sequencing dataset, the infiltration levels of different immune cell populations were determined using ssGSEA in the Gene Set Variation Analysis section of the RBioConductor software package. A total of 28 immune cell types were identified through ssGSEA analysis. Then, an immune cell feature clustering model was constructed based on the ssGSEA scores of each immune cell type. Unsupervised subtype clustering was performed on 925 breast cancer samples from the first RNA sequencing dataset using this immune cell feature clustering model to construct breast cancer immune subtypes based on immune cell features. To ensure the stability of the classification, 1000 iterations and an 80% resampling rate were performed, and the cumulative distribution function curve was used to determine the number of subtypes. The results are as follows: Figure 4 As shown, the tumor samples are mainly divided into two subtypes.

[0095] There were 525 patients with the two different subtypes (Immune cell cluster 1 and Immune cell cluster 2) and 400 patients with the other. Figure 5 The prognostic survival curves are for patients with two immune subtypes. From... Figure 5 It was found that patients with immune cell subtype 1 had better survival than those with immune cell subtype 2 (log-rank test, P<0.001). This suggests that immune cell subtype 1 and immune cell subtype 2 may be at their highest and lowest immune states, respectively.

[0096] This invention provides a method for constructing an immune cell feature clustering model, which is also based on the second RNA sequencing dataset of patients who have received immunotherapy. First, ssGSEA analysis is performed on the second RNA sequencing dataset using pre-set bioinformatics analysis software. This analysis identifies several types of immune cells and their corresponding ssGSEA scores. The ssGSEA scores of immune cells also reveal their impact on immunotherapy. Therefore, an immune cell feature clustering model is constructed based on these immune cells and their corresponding ssGSEA scores. This embodiment, based on the second RNA sequencing dataset, constructs an immune cell feature clustering model from the perspective of immune cells and combines it with the ssGSEA scores of each immune cell, improving the accuracy and rationality of subsequent clustering.

[0097] Furthermore, in step S4, combining the various LncRNA subtypes and the various immune subtypes to obtain several different breast cancer subtypes includes:

[0098] Each LncRNA subtype and each immune subtype are combined in pairs to generate a first number of LncRNA-immune subtypes;

[0099] The LncRNA-immune subtypes were enriched and scored using GSEA metabolic pathway analysis to obtain the corresponding pathway enrichment scores.

[0100] Based on the enrichment scores of each pathway, the metabolic pathway enrichment characteristics of each LncRNA-immune subtype are determined. LncRNA-immune subtypes with the same metabolic pathway enrichment characteristics are merged to obtain a second number of breast cancer subtypes, wherein the second number is less than or equal to the first number.

[0101] In a preferred embodiment, the three LncRNA subtypes obtained in step S2 and the two immune subtypes obtained in step S3 are combined to obtain six combinations. By performing GSEA metabolic pathway analysis on these six combinations, the enrichment characteristics of the metabolic pathways were determined, and based on these characteristics, four breast cancer subtypes were finally obtained: high fatty acid metabolism type, high amino acid metabolism type, high glucose metabolism type, and high folic acid metabolism type. Among them, 'LncRNA-cluster 1 & Immune cell cluster 1' was identified as high-fat fatty acid metabolizer (High-FA), 'LncRNA-cluster 2 & Immune cell cluster 1' or 'LncRNA-cluster 1 & Immune cell cluster 2' was identified as high-amino acid metabolizer (High-AA), 'LncRNA-cluster 2 & Immune cell cluster 2' or 'LncRNA-cluster 3 & Immune cell cluster 1' was identified as high-glucose metabolizer (High-Glu), and 'LncRNA-cluster 3 & Immune cell cluster 2' was identified as high-folate metabolizer (High-Folate). The metabolic pathway enrichment characteristics of the four breast cancer subtypes are as follows: Figure 6 As shown. Figure 7 This is a schematic diagram illustrating the process of constructing a new breast cancer subtype based on the first RNA sequencing dataset and the second RNA sequencing dataset in this embodiment.

[0102] Furthermore, there were statistically significant differences in overall survival (OS) among the four groups after immunotherapy. The high-fat-metabolism group showed the best OS benefit from immunotherapy (log-rank test, P = 0.00063). More importantly, higher expression of immunotherapy-related lncRNAs was significantly associated with longer OS, and the number of tumor-infiltrating immune cells was also significantly associated with patient prognosis. A comparison of overall survival after immunotherapy in patients with the four breast cancer subtypes is shown in the figure below. Figure 8 As shown.

[0103] In this embodiment of the invention, after obtaining each LncRNA subtype and each immune subtype, they are further merged into a two-dimensional index to generate a first number of LncRNA-immune subtypes. Among the various LncRNA-immune subtypes, there may be LncRNA-immune subtypes with similar therapeutic effects on immunotherapy. To improve the accuracy and efficiency of subsequent classification and prediction, LncRNA-immune subtypes with similar therapeutic effects on immunotherapy need to be merged into the same type. Therefore, this embodiment uses metabolic pathway enrichment characteristics as the evaluation index for each LncRNA-immune subtype, merging LncRNA-immune subtypes with the same metabolic pathway enrichment characteristics into the same breast cancer subtype, ultimately obtaining a second number of breast cancer subtypes. The breast cancer subtypes obtained in this embodiment significantly distinguish different immunotherapy efficacy and prognostic survival time, and reveal the metabolic connotation of each breast cancer subtype, providing data support for the formulation of subsequent tumor treatment plans.

[0104] In one possible implementation, in step S5, the typing prediction model generates a corresponding breast cancer typing prediction result based on the pathological slide images of the patient to be predicted, such as... Figure 9 As shown, steps S501-S505 are included:

[0105] Step S501: Perform image segmentation on the pathological slide image according to the tissue region in the pathological slide image to obtain several image blocks;

[0106] Step S502: Extract several feature vectors from each image patch using a convolutional neural network model with pre-trained parameters;

[0107] Step S503: Based on the feature vector, sort the image patches using a multi-instance learning algorithm based on an attention mechanism, and assign a corresponding attention score to each image patch;

[0108] Step S504: Aggregate the image information of each image block according to the attention score to generate aggregated information;

[0109] Step S505: Based on the aggregated information, using the convolutional neural network model through a multi-branch network architecture and multi-task objectives, perform breast cancer subtyping prediction on the pathological slice image to generate breast cancer subtyping prediction results for the patient to be predicted.

[0110] This invention provides a method for predicting breast cancer subtyping based on pathological slide images, using digital high-resolution histological slides as input to the subtyping prediction model. For each pathological slide image, the tissue content is automatically segmented, dividing it into thousands to tens of thousands of small image patches. Then, a pre-trained convolutional neural network model extracts corresponding feature vectors from each image patch. Based on the feature vectors, the image patches are sorted, and the image information of each image patch is aggregated according to their relative importance, i.e., attention scores, to obtain aggregated information for the entire pathological slide image. Finally, breast cancer subtyping prediction is performed based on the aggregated information, generating the predicted breast cancer subtyping result for the patient to be predicted. This achieves breast cancer subtyping diagnosis based on digital pathological slide features, avoiding the gene sequencing required for traditional tumor subtyping, and enabling subtyping prediction solely based on clinical pathological slides, thus improving the convenience and accuracy of breast cancer subtyping prediction.

[0111] In a preferred embodiment, the subtyping prediction model is constructed based on pathomic data of breast cancer patients from the TCGA and SYSMH databases, with patient data presented as digitized high-resolution histological slides as the primary input. First, for each full-slice image, an artificial intelligence algorithm automatically segments the tissue content, dividing it into thousands to tens of thousands of small image patches. These images are processed by a Convolutional Neural Network (CNN) with fixed pre-trained parameters. This network acts as an encoder, extracting compact, descriptive feature vectors from each image patch. Second, through a multi-instance learning algorithm based on an attention mechanism, the CNN learns to rank all tissue regions in the slide using the feature vectors and aggregates their information across the entire image according to their relative importance, assigning greater weight to regions considered to have high diagnostic relevance. Third, by using a multi-branch network architecture and multi-task objectives, the CNN can predict different metabolic-immune subtypes. Furthermore, the attention score assigned to each region by the network can be used to interpret the model's predictions. Specific steps are as follows:

[0112] The specific steps are as follows:

[0113] ① Data collection and preparation:

[0114] Patient data collection: Pathogenomics data of breast cancer patients were collected from the TCGA and SYSMH databases, with digital high-resolution histological sections as the primary input.

[0115] Image preprocessing: Each whole slice image is automatically segmented to generate thousands to tens of thousands of small image blocks as processing units.

[0116] ② Convolutional Neural Network (CNN) processing:

[0117] Feature extraction: A CNN with pre-trained parameters is used as an encoder to extract compact and descriptive feature vectors from each image patch.

[0118] Attention-based ranking: Using an attention-based multi-instance learning algorithm, CNN learns to rank the organizational regions in a slide and aggregate information based on relative importance.

[0119] ③ Multi-task objective prediction:

[0120] Breast cancer subtype prediction: CNN predicts different breast cancer subtypes through a multi-branch network architecture and multi-task objectives.

[0121] Attention score interpretation: The attention score assigned by the network to each region is used to interpret the model's prediction results.

[0122] ④Result Output and Interpretation:

[0123] Diagnostic result generation: Based on the model's prediction results, generate pathological classification diagnostic results for breast cancer patients.

[0124] Attention Score Interpretation: The attention score output by the model can be used to interpret the importance assessment of each region, thereby providing interpretability and transparency of the diagnosis. The construction process of the genotyping prediction model in this embodiment is as follows: Figure 10 As shown.

[0125] Compared with existing technologies, this model has the following advantages:

[0126] (1) Integration of multi-omics data: This model fully integrates multi-omics data, including long non-coding RNA (LncRNA) and immune cell characteristics, and constructs a clinical subtyping model for breast cancer through multi-faceted information, making the prediction more comprehensive.

[0127] (2) Application of deep learning technology: By adopting deep learning technology, especially convolutional neural networks (CNN) and models with attention mechanisms, this model achieves efficient analysis of pathological images and provides a new intelligent means for the clinical classification of breast cancer.

[0128] (3) Superior performance metrics: The model achieved excellent performance on both the training and validation sets by using metrics such as the receiver operating characteristic (ROC) curve and its area under the curve (AUC). The micro-average AUC is as high as 0.96, and the macro-average ROC is excellent.

[0129] (4) Integration of classification and prognosis: This model can not only perform clinical classification of breast cancer, but also predict prognosis, providing doctors with more comprehensive information and helping to develop more individualized treatment plans.

[0130] (5) Lowering the threshold for clinical application: By using a classification and diagnosis model based on pathological images, this model avoids dependence on gene sequencing and can perform classification and diagnosis based solely on clinical pathological slides, thus lowering the threshold for clinical application of the new classification.

[0131] (6) Precise prediction of immunotherapy efficacy: This model has better sensitivity and accuracy than traditional indicators, providing accurate prediction of the efficacy and prognosis of breast cancer immunotherapy, and providing patients with more effective treatment options.

[0132] Example 2:

[0133] like Figure 11 As shown, correspondingly, Embodiment 2 provides a breast cancer subtyping prediction system based on LncRNA and immune cell characteristics, including an acquisition module 10, an LncRNA subtype generation module 20, an immune subtype generation module 30, a breast cancer subtyping generation module 40, and a prediction module 50.

[0134] The acquisition module 10 is used to acquire first RNA sequencing datasets of several historical breast cancer patients and second RNA sequencing datasets of several patients who received immunotherapy.

[0135] The LncRNA subtype generation module 20 is used to input the first RNA sequencing dataset into a preset LncRNA clustering model, so that the LncRNA clustering model clusters each RNA sequencing data in the first RNA sequencing dataset into different LncRNA subtypes, wherein the LncRNA clustering model is constructed based on the feature set of the second RNA sequencing dataset;

[0136] The immune subtype generation module 30 is used to input the first RNA sequencing dataset into a preset immune cell feature clustering model, so that the immune cell feature clustering model clusters each RNA sequencing data in the first RNA sequencing dataset into different immune subtypes. The immune cell feature clustering model is constructed based on each immune cell in the second RNA sequencing dataset and the corresponding ssGSEA score.

[0137] The breast cancer typing generation module 40 is used to combine the various LncRNA subtypes and the various immune subtypes to obtain several different breast cancer typings.

[0138] The prediction module 50 is used to input the pathological slide image of the patient to be predicted into a preset subtyping prediction model, so that the subtyping prediction model generates a corresponding breast cancer subtyping prediction result, wherein the breast cancer subtyping prediction result is one of the several different breast cancer subtypings, and the subtyping prediction model is constructed based on the pathological slide data of historical breast cancer patients and a pre-trained convolutional neural network model.

[0139] In one possible implementation, constructing the LncRNA clustering model based on the feature set of the second RNA sequencing dataset includes:

[0140] The corresponding LncRNA expression dataset was obtained by transcription from the second RNA sequencing dataset;

[0141] Based on the magnitude of the influence factors of each feature in the LncRNA expression dataset on immunotherapy, several LncRNA features are extracted from the LncRNA expression dataset and combined to form an LncRNA feature set;

[0142] A clustering model of LncRNAs was constructed based on the LncRNA feature set.

[0143] In one possible implementation, the step of constructing the immune cell feature clustering model based on each immune cell in the second RNA sequencing dataset and its corresponding ssGSEA score includes:

[0144] The second RNA sequencing dataset is input into a preset bioinformatics analysis software so that the bioinformatics analysis software can determine several types of immune cells and their corresponding ssGSEA scores through ssGSEA analysis.

[0145] An immune cell feature clustering model was constructed based on the various immune cells and their corresponding ssGSEA scores.

[0146] Furthermore, the breast cancer subtyping generation module 40 combines the various LncRNA subtypes and the various immune subtypes to obtain several different breast cancer subtypes, including:

[0147] Each LncRNA subtype and each immune subtype are combined in pairs to generate a first number of LncRNA-immune subtypes;

[0148] The LncRNA-immune subtypes were enriched and scored using GSEA metabolic pathway analysis to obtain the corresponding pathway enrichment scores.

[0149] Based on the enrichment scores of each pathway, the metabolic pathway enrichment characteristics of each LncRNA-immune subtype are determined. LncRNA-immune subtypes with the same metabolic pathway enrichment characteristics are merged to obtain a second number of breast cancer subtypes, wherein the second number is less than or equal to the first number.

[0150] In one possible implementation, such as Figure 12 As shown, the prediction module 50 includes an image segmentation unit 501, a feature extraction unit 502, a sorting unit 503, an information aggregation unit 504, and a prediction unit 505.

[0151] The image segmentation unit 501 is used to segment the pathological slide image according to the tissue region in the pathological slide image to obtain a number of image blocks.

[0152] The feature extraction unit 502 is used to extract several feature vectors from each of the image blocks using a convolutional neural network model with pre-trained parameters;

[0153] The sorting unit 503 is used to sort the image blocks according to the feature vector using a multi-instance learning algorithm based on an attention mechanism, and to assign a corresponding attention score to each image block.

[0154] The information aggregation unit 504 is used to aggregate the image information of each image block according to the attention score to generate aggregated information;

[0155] The prediction unit 505 is used to predict breast cancer subtyping of the pathological slice image based on the aggregated information, through a multi-branch network architecture and multi-task objectives, using the convolutional neural network model, and to generate breast cancer subtyping prediction results for the patient to be predicted.

[0156] This invention provides a breast cancer subtyping prediction system based on lncRNA and immune cell characteristics. First, an lncRNA clustering model and an immune cell characteristic clustering model are constructed based on a second RNA sequencing dataset to identify lncRNA subtypes and immune subtypes in the first RNA sequencing dataset. Then, by combining each lncRNA subtype and each immune subtype, several breast cancer subtypes are constructed. The breast cancer subtypes constructed in this embodiment fully consider the influence of long non-coding RNAs and immune cells on the efficacy of tumor immunotherapy, and reconstruct a breast cancer subtype that is more sensitive to immunotherapy. This breast cancer subtype has better sensitivity and accuracy in predicting the efficacy of immunotherapy than traditional indicators. Furthermore, in conjunction with the breast cancer subtypes, this embodiment also provides a subtype prediction model based on pathological images. This avoids the gene sequencing required for traditional tumor subtyping, and subtyping prediction can be performed using only clinical pathological slides, improving the convenience and accuracy of breast cancer subtyping prediction, lowering the clinical application threshold of the breast cancer subtypes, and providing decision-making basis for medical professionals to formulate cancer treatment plans.

[0157] For a more detailed explanation of the working principle and procedures of this embodiment, please refer to the relevant description in Embodiment 1.

[0158] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A breast cancer typing prediction method based on LncRNA and immune cell features, characterized in that, The method comprises the following steps: obtaining a first RNA sequencing data set of a plurality of historical breast cancer patients and a second RNA sequencing data set of a plurality of patients receiving immunotherapy; inputting the first RNA sequencing data set into a preset LncRNA clustering model, so that the LncRNA clustering model clusters each RNA sequencing data in the first RNA sequencing data set into different LncRNA subtypes, wherein the LncRNA clustering model is obtained according to a feature set of the second RNA sequencing data set; inputting the first RNA sequencing data set into a preset immune cell feature clustering model, so that the immune cell feature clustering model clusters each RNA sequencing data in the first RNA sequencing data set into different immune subtypes, wherein the immune cell feature clustering model is obtained according to each immune cell and the corresponding ssGSEA score of the second RNA sequencing data set; obtaining a plurality of different breast cancer subtypes by combining the LncRNA subtypes and the immune subtypes; inputting a pathological section image of a patient to be predicted into a preset subtype prediction model, so that the subtype prediction model generates a corresponding breast cancer subtype prediction result, wherein the breast cancer subtype prediction result is one of the plurality of different breast cancer subtypes, and the subtype prediction model is obtained based on the pathological section data of the historical breast cancer patients and a pre-trained convolutional neural network model.

2. The breast cancer typing prediction method based on LncRNA and immune cell characteristics according to claim 1, characterized in that, The LncRNA clustering model is obtained according to the feature set of the second RNA sequencing data set, comprising: transcribing the second RNA sequencing data set to obtain a corresponding LncRNA expression data set; extracting a plurality of LncRNA features from the LncRNA expression data set according to the influence factor size of each feature in the LncRNA expression data set on immunotherapy, and combining to form an LncRNA feature set; constructing an LncRNA clustering model according to the LncRNA feature set.

3. The breast cancer typing prediction method based on LncRNA and immune cell characteristics according to claim 1, characterized in that, The immune cell feature clustering model is obtained according to each immune cell and the corresponding ssGSEA score of the second RNA sequencing data set, comprising: inputting the second RNA sequencing data set into a preset bioinformatics analysis software, so that the bioinformatics analysis software determines a plurality of immune cells and the corresponding ssGSEA scores of each immune cell through ssGSEA analysis; constructing an immune cell feature clustering model according to the immune cells and the corresponding ssGSEA scores.

4. The breast cancer typing prediction method based on LncRNA and immune cell characteristics according to claim 1, characterized in that, The plurality of different breast cancer subtypes are obtained by combining the LncRNA subtypes and the immune subtypes, comprising: combining the LncRNA subtypes and the immune subtypes in pairs to generate a first number of LncRNA-immune subtypes; obtaining a corresponding pathway enrichment score by performing pathway enrichment scoring on each LncRNA-immune subtype through GSEA metabolic pathway analysis; According to the enrichment score of each pathway, the metabolic pathway enrichment characteristics of each LncRNA-immune subtype are determined, and LncRNA-immune subtypes with the same metabolic pathway enrichment characteristics are combined to obtain a second number of breast cancer subtypes, wherein the second number is less than or equal to the first number.

5. The breast cancer typing prediction method based on LncRNA and immune cell characteristics according to claim 1, characterized in that, The typing prediction model generates a corresponding breast cancer typing prediction result according to the pathological section image of the patient to be predicted, which comprises: According to the tissue region in the pathological section image, the pathological section image is segmented to obtain a plurality of image blocks; Using a convolutional neural network model with pre-trained parameters, a plurality of feature vectors are extracted from the image blocks; According to the feature vectors, a multi-instance learning algorithm based on an attention mechanism is used to sort the image blocks, and each image block is assigned a corresponding attention score; According to the attention score, the image information of the image blocks is aggregated to generate aggregated information; According to the aggregated information, the convolutional neural network model is used to predict the breast cancer typing of the pathological section image through a multi-branch network architecture and a multi-task target, and the breast cancer typing prediction result of the patient to be predicted is generated. 6.A breast cancer typing prediction system based on LncRNA and immune cell features, characterized in that, It comprises an acquisition module, an LncRNA subtype generation module, an immune subtype generation module, a breast cancer typing generation module, and a prediction module; The acquisition module is used to acquire a first RNA sequencing data set of a plurality of historical breast cancer patients and a second RNA sequencing data set of a plurality of patients receiving immunotherapy; The LncRNA subtype generation module is used to input the first RNA sequencing data set into a pre-set LncRNA clustering model, so that the LncRNA clustering model clusters each RNA sequencing data in the first RNA sequencing data set into different LncRNA subtypes, wherein the LncRNA clustering model is constructed according to a feature set of the second RNA sequencing data set; The immune subtype generation module is used to input the first RNA sequencing data set into a pre-set immune cell feature clustering model, so that the immune cell feature clustering model clusters each RNA sequencing data in the first RNA sequencing data set into different immune subtypes, wherein the immune cell feature clustering model is constructed according to each immune cell and the corresponding ssGSEA score of the second RNA sequencing data set; The breast cancer typing generation module is used to combine the LncRNA subtypes and the immune subtypes to obtain a plurality of different breast cancer subtypes; The prediction module is used to input the pathological section image of the patient to be predicted into a pre-set typing prediction model, so that the typing prediction model generates a corresponding breast cancer typing prediction result, wherein the breast cancer typing prediction result is one of the plurality of different breast cancer subtypes, and the typing prediction model is constructed based on the pathological section data of the historical breast cancer patients and the pre-trained convolutional neural network model.

7. The breast cancer typing prediction system based on LncRNA and immune cell characteristics of claim 6, wherein, The LncRNA clustering model is constructed according to the feature set of the second RNA sequencing data set, which comprises: transcriptionally obtaining a corresponding LncRNA expression dataset from the second RNA sequencing dataset; extracting a plurality of LncRNA features from the LncRNA expression dataset according to the influence factor size of each feature in the LncRNA expression dataset on immunotherapy, and combining to form an LncRNA feature set; constructing an LncRNA clustering model according to the LncRNA feature set.

8. The breast cancer typing prediction system based on LncRNA and immune cell characteristics of claim 6, wherein, The immune cell feature clustering model is constructed according to each immune cell in the second RNA sequencing dataset and the corresponding ssGSEA score, including: inputting the second RNA sequencing dataset into a preset bioinformatics analysis software, so that the bioinformatics analysis software determines a plurality of immune cells and the corresponding ssGSEA scores of various immune cells through ssGSEA analysis; constructing an immune cell feature clustering model according to the various immune cells and the corresponding ssGSEA scores.

9. The breast cancer typing prediction system based on LncRNA and immune cell characteristics of claim 6, wherein, The breast cancer typing generation module combines the various LncRNA subtypes and the various immune subtypes to obtain a plurality of different breast cancer types, including: combining the various LncRNA subtypes and the various immune subtypes in pairs to generate a first number of LncRNA-immune subtypes; performing pathway enrichment scoring on each LncRNA-immune subtype through GSEA metabolic pathway analysis to obtain a corresponding pathway enrichment score; determining the metabolic pathway enrichment characteristics of each LncRNA-immune subtype according to each pathway enrichment score, and merging LncRNA-immune subtypes with the same metabolic pathway enrichment characteristics to obtain a second number of breast cancer types, wherein the second number is less than or equal to the first number.

10. The breast cancer typing prediction system based on LncRNA and immune cell characteristics of claim 6, wherein, The prediction module includes an image segmentation unit, a feature extraction unit, a sorting unit, an information aggregation unit, and a prediction unit; The image segmentation unit is used to perform image segmentation on the pathological section image according to the tissue region in the pathological section image to obtain a plurality of image blocks. The feature extraction unit is used to extract a plurality of feature vectors from each image block using a pre-trained convolutional neural network model; The sorting unit is used to sort each image block using a multi-instance learning algorithm based on an attention mechanism according to the feature vector, and assign each image block a corresponding attention score; The information aggregation unit is used to aggregate image information of each image block according to the attention score to generate aggregated information; The prediction unit is used to perform breast cancer typing prediction on the pathological section image using the convolutional neural network model through a multi-branch network architecture and a multi-task target according to the aggregated information, to generate a breast cancer typing prediction result for the patient to be predicted.

Images