Knowledge mining method and system for tumor field

Abstract

The invention relates to the technical field of data mining and analysis, in particular to a knowledge mining method and system for the tumor field. The method comprises the following steps: acquiring a tumor multi-source heterogeneous data set, and performing data standardization processing to obtain a standardized tumor multi-source heterogeneous data set; performing multi-level feature extraction on the standardized tumor multi-source heterogeneous data set to obtain a tumor fusion feature set; performing hierarchical knowledge graph construction based on the tumor fusion feature set to obtain a tumor hierarchical knowledge graph; and constructing a graph neural network model by using the tumor hierarchical knowledge graph, and carrying out cancer cell dynamic prediction to obtain tumor dynamic knowledge reasoning result data. According to the method, cancer cell prediction is carried out by extracting the multi-level features, constructing the hierarchical knowledge graph and applying the graph neural network, and personalized medical research optimization is provided, so that the accuracy of cancer cell prediction and the personalization of the medical research direction are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining and analysis, and in particular to a knowledge mining method and system for the tumor field. Background Art

[0002] Most existing tumor knowledge mining methods rely on a single data source or shallow analysis methods in terms of feature extraction. For example, some methods only perform mutation analysis based on genetic data, while ignoring the comprehensive role of other key data (such as imaging features and clinical test indicators). This single data dependency is not only difficult to reflect the multidimensional characteristics of tumors, but also difficult to capture complex biological mechanisms. As a tool for efficiently organizing and representing complex relationships, knowledge graphs have been applied to a certain extent in the medical field. However, the construction of knowledge graphs in the field of tumors is still in its early stages, lacking hierarchical structural design, and failing to fully utilize the multi-level characteristics of data to reflect the complexity and logic of knowledge in the field of tumors. Current tumor knowledge mining methods have limited feature extraction capabilities, imperfect knowledge graphs, and weak dynamic reasoning capabilities. These defects not only restrict the effectiveness of existing technologies in practical applications. Summary of the invention

[0003] Based on this, it is necessary for the present invention to provide a knowledge mining method and system for the tumor field to solve at least one of the above technical problems.

[0004] To achieve the above object, a knowledge mining method for tumor field comprises the following steps: Step S1: obtaining a tumor multi-source heterogeneous data set, and performing data standardization processing to obtain a standardized tumor multi-source heterogeneous data set; Step S2: performing multi-level feature extraction on the standardized tumor multi-source heterogeneous data set to obtain a tumor fusion feature set; Step S3: constructing a hierarchical knowledge graph based on the tumor fusion feature set to obtain a tumor hierarchical knowledge graph; Step S4: constructing a graph neural network model using the tumor hierarchical knowledge graph, and performing dynamic prediction of cancer cells to obtain tumor dynamic knowledge reasoning result data; Step S5: Perform personalized analysis of patients based on the tumor dynamic knowledge reasoning result data, predict the probability of cancer cell inhibition, and generate a personalized analysis report.

[0005] Through standardized processing and multi-level feature extraction, the present invention can fully integrate and analyze multi-source heterogeneous data of tumors, reveal different biomarkers and clinical manifestations of tumors, and provide a scientific basis for personalized medical research. The dynamic prediction capabilities constructed by using the hierarchical knowledge graph of tumors and the graph neural network model can accurately predict the dynamic changes of cancer cells and the prognostic response of patients. Based on the dynamic knowledge reasoning result data of tumor patients, the probability of cancer cell inhibition is predicted, the direction of personalized medical research is formulated, and the prognostic risk of patients is reduced. Through the dynamic prediction of cancer cells and the assessment of the risk of cancer cell recurrence, the possibility of tumor recurrence can be discovered in advance and the risk of recurrence can be reduced. Through the fusion of multi-source data, combined with graph neural network and knowledge graph technology, intelligent analysis and comprehensive interpretation of tumor data are realized, providing strong support for multidisciplinary collaboration in the field of tumors.

[0006] Preferably, the present invention further provides a knowledge mining system for the tumor field, which is used to execute the above-mentioned knowledge mining method for the tumor field. The knowledge mining system for the tumor field comprises: A standardized data preprocessing module is used to obtain a tumor multi-source heterogeneous data set and perform data standardization processing to obtain a standardized tumor multi-source heterogeneous data set; A multi-level feature extraction module is used to perform multi-level feature extraction on standardized tumor multi-source heterogeneous data sets to obtain a tumor fusion feature set; A hierarchical knowledge graph construction module is used to construct a hierarchical knowledge graph based on the tumor fusion feature set to obtain a tumor hierarchical knowledge graph; The graph neural network dynamic prediction module is used to build a graph neural network model using the tumor hierarchical knowledge graph and to dynamically predict cancer cells to obtain tumor dynamic knowledge reasoning result data; The personalized analysis and prediction module is used to conduct personalized analysis of patients based on tumor dynamic knowledge reasoning result data, predict the probability of cancer cell inhibition, and generate personalized analysis reports.

[0007] The present invention can convert multi-source heterogeneous tumor data sets into unified standard data sets through a standardized data preprocessing module, providing a clear and accurate data basis for subsequent analysis. Through a multi-level feature extraction module, it is possible to deeply mine the key features in the data to form a tumor fusion feature set, thereby more comprehensively analyzing the multi-dimensional characteristics of the tumor. The hierarchical knowledge graph construction module effectively integrates knowledge in the field of tumors, provides structured knowledge support for cancer cell prediction and personalized medical research, and helps to form a more scientific basis for decision-making. The graph neural network dynamic prediction module and the personalized analysis and prediction module can achieve accurate dynamic prediction of cancer cells through the application of hierarchical knowledge graphs of tumors, thereby providing personalized medical research directions and cancer cell inhibition probability predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings: Figure 1 A schematic diagram of the steps of a knowledge mining method for the tumor field according to the present invention; Figure 2 for Figure 1 Detailed step flow diagram of step S1; Figure 3 for Figure 1 Detailed step flow chart of step S2 in FIG. DETAILED DESCRIPTION

[0009] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0010] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0011] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0012] To achieve this, please refer to Figures 1 to 3 The present invention provides a knowledge mining method for tumor field, the method comprising the following steps: Step S1: obtaining a tumor multi-source heterogeneous data set, and performing data standardization processing to obtain a standardized tumor multi-source heterogeneous data set; Step S2: performing multi-level feature extraction on the standardized tumor multi-source heterogeneous data set to obtain a tumor fusion feature set; Step S3: constructing a hierarchical knowledge graph based on the tumor fusion feature set to obtain a tumor hierarchical knowledge graph; Step S4: constructing a graph neural network model using the tumor hierarchical knowledge graph, and performing dynamic prediction of cancer cells to obtain tumor dynamic knowledge reasoning result data; Step S5: Perform personalized analysis of patients based on the tumor dynamic knowledge reasoning result data, predict the probability of cancer cell inhibition, and generate a personalized analysis report.

[0013] In the embodiment of the present invention, reference Figure 1 FIG. 1 is a schematic diagram of a step flow chart of a knowledge mining method for tumor field according to the present invention. In this example, the knowledge mining method for tumor field includes the following steps: Step S1: obtaining a tumor multi-source heterogeneous data set, and performing data standardization processing to obtain a standardized tumor multi-source heterogeneous data set; The embodiment of the present invention obtains a tumor multi-source heterogeneous data set, which includes tumor-related data from different data sources. These data sources have different formats and dimensions. In order to ensure the effectiveness and consistency of subsequent processing, the data needs to be standardized. Missing values ​​are filled, outliers are detected and removed for the original data to ensure the integrity and accuracy of the data. According to the different sources of the tumor multi-source heterogeneous data set, the data is uniformly converted into a structured format. The numerical data is normalized and mapped to a unified range. According to the specific needs of tumor research, key features are extracted, and the data dimension is reduced by dimensionality reduction methods to reduce redundant information. After the above-mentioned standardization process, a standardized tumor multi-source heterogeneous data set is finally obtained.

[0014] Step S2: performing multi-level feature extraction on the standardized tumor multi-source heterogeneous data set to obtain a tumor fusion feature set; The embodiment of the present invention divides the standardized tumor multi-source heterogeneous data set into different levels according to the data source and feature dimension, and extracts features from each layer respectively. For each layer of data, an adaptation method is used to extract features. The genome layer adopts a feature extraction method based on sequence analysis (such as k-mer feature extraction or mutation frequency statistics). The proteome layer adopts a feature extraction method based on a protein interaction network (such as network topology feature analysis). The imaging layer uses a deep learning model (such as a convolutional neural network, CNN) to extract image features. The clinical layer uses a statistical analysis method to extract key clinical indicators. Features extracted at different levels are integrated using a specific fusion method. Multi-level features are reduced in dimension and fused using a weighted average or principal component analysis (PCA) method. A deep learning framework (such as an autoencoder or a multimodal learning model) is used to semantically fuse cross-level features. In the fused tumor feature set, a feature selection algorithm (such as recursive feature elimination RFE or a selection method based on information gain) is further used to screen key features related to tumor research, eliminate redundant or irrelevant features, and improve the accuracy and reliability of the feature set. After the above-mentioned multi-level feature extraction and optimization processing, a tumor fusion feature set is finally obtained.

[0015] Step S3: constructing a hierarchical knowledge graph based on the tumor fusion feature set to obtain a tumor hierarchical knowledge graph; The embodiment of the present invention is based on the tumor fusion feature set, and classifies it according to the feature source and mechanism of action to generate feature sets of different dimensions. For each dimension feature after stratification, a corresponding knowledge unit is constructed and its internal association relationship is defined. The above knowledge units are hierarchically associated according to specific rules, and the structure is optimized using a graph construction algorithm to ensure that the weight distribution of nodes and edges is reasonable, and the graph content is dynamically updated according to the latest tumor medical research results to generate a hierarchical tumor knowledge graph.

[0016] Step S4: constructing a graph neural network model using the tumor hierarchical knowledge graph, and performing dynamic prediction of cancer cells to obtain tumor dynamic knowledge reasoning result data; The present invention is based on the tumor hierarchical knowledge graph, and defines the input and structure of the graph neural network. The nodes in the tumor hierarchical knowledge graph are represented as feature vectors, and the node features are derived from the tumor fusion feature set; the edges are represented as relationship matrices to describe the interactive relationship between the nodes. According to the characteristics of tumor data, a suitable graph neural network architecture is selected, such as a graph convolutional network, a graph attention network or a graph embedding model. The initial parameters of the model are set. Using the graph neural network, the node features are dynamically updated through multi-layer propagation rules. The features of each node are updated by aggregating the information of its neighboring nodes, and the aggregation function can select summation, mean or attention weighted summation. The activation function is used to perform nonlinear transformation on the aggregated features to enhance the feature expression ability. Through the superposition of multi-layer neural networks, the deep fusion and propagation of cross-level information are realized. Based on the dynamically updated node features, the behavior of cancer cells is predicted. The proliferation, metastasis or apoptosis state of cancer cells is predicted, and related dynamic feature data is generated. The trend of cancer cell state changes over time is inferred by using the time series modeling method combined with the output of the graph neural network. The prediction results are combined with the original information in the tumor hierarchical knowledge graph to generate tumor dynamic knowledge reasoning result data.

[0017] Step S5: Perform personalized analysis of patients based on the tumor dynamic knowledge reasoning result data, predict the probability of cancer cell inhibition, and generate a personalized analysis report.

[0018] The embodiment of the present invention performs multi-dimensional analysis based on the tumor dynamic knowledge reasoning result data and the individual characteristics of the patient. The patient characteristics are extracted and matched with the corresponding node characteristics in the tumor dynamic knowledge reasoning result data, and the correlation between the patient characteristics and the tumor characteristics is analyzed. Based on the patient's specific gene expression, imaging data and clinical indicators, the patient's disease progression trend and prognosis are evaluated. Using a similarity algorithm, the patient characteristics are compared with the existing tumor patient data in the database to identify potential similar cases. Combine the patient's personalized analysis results and the tumor dynamic knowledge reasoning result data to predict the probability of cancer cell inhibition. Combine the patient characteristics with the dynamic features in the tumor dynamic knowledge reasoning result data as the input of the prediction model. Construct a prediction model based on machine learning or deep learning, and train the model to predict the probability of cancer cells being effectively inhibited. The prediction results are verified according to historical case data to evaluate the accuracy and robustness of the prediction model. Generate a personalized analysis report for the patient based on the patient's personalized analysis results and the cancer cell inhibition probability prediction results.

[0019] Through standardized processing and multi-level feature extraction, the present invention can fully integrate and analyze multi-source heterogeneous data of tumors, reveal different biomarkers and clinical manifestations of tumors, and provide a scientific basis for personalized medical research. The dynamic prediction capabilities constructed by the hierarchical knowledge graph of tumors and the graph neural network model can accurately predict the dynamic changes of cancer cells and the prognostic response of patients. Based on the dynamic knowledge reasoning result data of tumor patients, the probability of cancer cell inhibition is predicted, and the direction of personalized medical research is formulated. Through the dynamic prediction of cancer cells and the risk assessment of cancer cell recurrence, the possibility of tumor recurrence can be discovered in advance and the risk of recurrence can be reduced. Through multi-source data fusion, combined with graph neural network and knowledge graph technology, intelligent analysis and comprehensive interpretation of tumor data are realized, providing strong support for multidisciplinary collaboration in the field of tumors.

[0020] Preferably, step S1 comprises the following steps: Step S11: collecting historical clinical test data through a medical test information system to obtain historical clinical test data; Step S12: collecting case medical image data through a medical image information system to obtain case medical image data; Step S13: collecting clinical symptom data through the electronic medical record system to obtain clinical symptom data of the tumor; Step S14: standardize the historical clinical test data, case medical imaging data and tumor clinical symptom data respectively, and merge the data to obtain a standardized tumor multi-source heterogeneous data set.

[0021] As an embodiment of the present invention, refer to Figure 2 As shown, Figure 1 Detailed step flow diagram of step S1 in the embodiment of the present invention, step S1 includes the following steps: Step S11: collecting historical clinical test data through a medical test information system to obtain historical clinical test data; The embodiment of the present invention connects to a medical laboratory information system to obtain historical clinical laboratory data related to tumors. The docking process includes exchanging data with the medical laboratory information system through a standardized data interface to ensure the standardization and consistency of data collection; sending a query request to the medical laboratory information system to specify the scope of the inspection items. According to the parameters of the data request, historical clinical laboratory data within the specified range is extracted from the medical laboratory information system, including but not limited to tumor marker data, blood test data, and gene test data. The collected historical clinical laboratory data is integrity checked to ensure the quality and availability of the data. The verified historical clinical laboratory data is stored in a unified database, and the data is marked to record the data source (medical laboratory information system) and time range (historical data collection time period). Finally, the historical clinical laboratory data is obtained.

[0022] Step S12: collecting case medical image data through a medical image information system to obtain case medical image data; The embodiment of the present invention connects to a medical imaging information system to obtain medical imaging data related to tumors. The docking process includes using a standardized communication protocol to dock with the medical imaging information system to ensure the stability and accuracy of image data transmission; sending a data acquisition request to the medical imaging information system according to the needs of tumor research, specifying the image type (such as CT, MRI, PET) and related case screening conditions. Extracting medical imaging data related to tumor cases from the medical imaging information system includes selecting a suitable imaging type according to the target tumor type (such as MRI for soft tissue tumors and PET for metabolic activity assessment); setting case screening conditions to ensure that the collected cases are representative and relevant; collecting multimodal imaging data (such as CT+PET) in a single case to provide multi-dimensional imaging information for subsequent analysis. The extracted case medical imaging data is preliminarily preprocessed to improve the availability and consistency of the data. The preprocessed case medical imaging data is stored in a unified data storage system, and the data source (medical imaging information system) and extraction conditions (such as case screening parameters, imaging type) are recorded. Finally, the case medical imaging data is obtained.

[0023] Step S13: collecting clinical symptom data through the electronic medical record system to obtain clinical symptom data of the tumor; The embodiment of the present invention connects to the electronic medical record system to collect clinical symptom data related to tumors. The docking process includes using a standardized data interface to ensure that the data interaction between the electronic medical record system and the acquisition module is stable and consistent; according to research needs, a query request is sent to the electronic medical record system to specify the acquisition scope. Extracting clinical symptom data related to tumors from the electronic medical record system includes extracting typical tumor symptoms recorded by patients in medical records; extracting tumor-related signs recorded by doctors in physical examinations; collecting symptoms reflected in laboratory tests and imaging examinations related to tumors; and extracting patient prognosis response descriptions. Perform a quality check on the extracted tumor clinical symptom data to ensure the accuracy and completeness of the data. The verified tumor clinical symptom data is stored in a unified data storage system, and the data source (electronic medical record system) and extraction conditions are recorded. Finally, the tumor clinical symptom data is obtained.

[0024] Step S14: standardize the historical clinical test data, case medical imaging data and tumor clinical symptom data respectively, and merge the data to obtain a standardized tumor multi-source heterogeneous data set.

[0025] The embodiment of the present invention standardizes historical clinical test data, case medical imaging data and tumor clinical symptom data respectively to ensure data format, range and semantic consistency. Unit conversion is performed on historical clinical test data from different sources, for example, blood glucose values ​​are unified into mmol / L; the test data is normalized using the normal reference range to adapt the data to the model input; and the missing test result data is processed by the mean filling or interpolation method. The case medical imaging data is converted into DICOM format to ensure file structure consistency; the image data is resampled and size normalized to unify the resolution and pixel spacing; contrast adjustment and denoising algorithms are applied to improve image quality to ensure clear image features. The symptom description terms in the tumor clinical symptom data are standardized, and the descriptions are mapped using a unified medical terminology library (such as SNOMEDCT); descriptive symptoms (such as "mild", "moderate" and "severe") are mapped into quantitative values; and the symptom data field names are ensured to be consistent with other data sets for easy merging. The standardized historical clinical test data, case medical imaging data and tumor clinical symptom data are merged to construct a unified standardized tumor multi-source heterogeneous data set.

[0026] The present invention can integrate multiple heterogeneous data sources through the comprehensive collection of historical clinical test data, medical imaging data and clinical symptom data, providing a rich information basis for subsequent analysis and prediction. Through standardized processing, the consistency of tumor data from different sources in format and dimension is ensured, the problems caused by data heterogeneity are eliminated, and the accuracy and reliability of subsequent analysis results are improved. Through precise data integration, the pathology, imaging and symptom characteristics of tumor patients can be fully understood, providing a scientific basis for the formulation of personalized medical research. Based on standardized multi-source heterogeneous data sets, it can effectively support early prediction and dynamic monitoring of cancer cells, and identify potential risks of tumors in advance through data fusion and in-depth analysis.

[0027] Preferably, step S2 comprises the following steps: Gene sequencing analysis is performed on standardized tumor multi-source heterogeneous data sets, and molecular-level feature extraction is performed to obtain a tumor molecular-level feature set; Conduct clinical tests and pathological image analysis on standardized tumor multi-source heterogeneous data sets, and extract cell-level features to obtain tumor cell-level feature sets; Perform medical image analysis on standardized tumor multi-source heterogeneous datasets and extract tissue-level features to obtain a tumor tissue-level feature set; Perform multimodal feature fusion on the tumor molecular level feature set, tumor cell level feature set and tumor tissue level feature set to obtain a preliminary fusion feature set; Correlation analysis and verification were performed on the preliminary fusion feature set to generate the tumor fusion feature set.

[0028] As an embodiment of the present invention, refer to Figure 3 As shown, Figure 1 Detailed step flow diagram of step S2 in the embodiment of the present invention, step S2 includes the following steps: S21: Perform gene sequencing analysis on standardized tumor multi-source heterogeneous data sets and extract molecular-level features to obtain a tumor molecular-level feature set; The embodiment of the present invention extracts gene sequencing data from a standardized tumor multi-source heterogeneous data set to ensure that the data is complete and consistent. Bioinformatics analysis tools are used to conduct in-depth analysis of tumor gene sequencing data to extract molecular-level features. Gene mutation information is identified; mutation frequency, functional impact prediction (such as SIFT and PolyPhen scores) and gene function categories (such as signal transduction genes, DNA repair genes) where the mutations are located are extracted. Transcriptome analysis is performed on sequencing data to calculate gene expression levels in tumor samples (such as FPKM or TPM values); differentially expressed genes are annotated, and their functional pathways related to tumor types are recorded. Mutated genes and differentially expressed genes are mapped to known biological pathways (such as KEGG, Reactome); key signal nodes in the pathway and their abnormal activity information are extracted. The extracted molecular-level features are preprocessed and standardized to ensure that the feature data has a unified scale and format. Finally, a tumor molecular-level feature set is obtained.

[0029] S22: Conduct clinical tests and pathological image analysis on standardized tumor multi-source heterogeneous data sets, and extract cell-level features to obtain tumor cell-level feature sets; The embodiment of the present invention extracts clinical test data and pathological image data from a standardized tumor multi-source heterogeneous data set to ensure that the data is complete, accurate and consistent. Image processing and analysis technology is used to extract features at the cell level of pathological images, and the features of tumor cells are analyzed in combination with clinical test data. Image processing technology (such as an automated image analysis method based on a convolutional neural network) is used to segment and classify cells in pathological slice images to identify different types of tumor cells. The morphological characteristics of cells are extracted from the image. The arrangement and tissue structure of the cell population are analyzed to determine the degree of infiltration and heterogeneity of the tumor. The biological characteristics of the cells are evaluated using biomarkers or gene expression data contained in the clinical test data. The extracted cell-level features are preprocessed and standardized to ensure the consistency and availability of the data. The extracted cell-level features are stored in a structured format to ensure that they can be effectively associated with feature sets at other levels. Finally, a tumor cell-level feature set is obtained.

[0030] S23: Perform medical image analysis on the standardized tumor multi-source heterogeneous dataset and extract tissue-level features to obtain a tumor tissue-level feature set; The embodiment of the present invention extracts medical imaging data from a standardized tumor multi-source heterogeneous data set to ensure the integrity and high quality of the data. The tissue-level features of the tumor are extracted using medical image analysis technology. The tumor region is identified and accurately located through a tumor region calibration algorithm (such as image segmentation technology) in the medical image data. The volume, shape, and boundary irregularity of the tumor are measured to evaluate the growth and expansion trend of the tumor. The tissue texture features of the tumor region are extracted to evaluate the heterogeneity of the tumor. The distribution of the tumor in the tissue is analyzed to determine whether there are tumor infiltration, metastasis, or other signs. For PET scans or functional MRI images, the metabolic activity or blood flow characteristics of the tumor region, such as the standard uptake value (SUV), are extracted to evaluate the metabolic characteristics and biological activity of the tumor. The extracted tissue-level features are standardized to ensure the consistency and comparability of the data. The processed tumor tissue-level feature set is stored in a structured database to ensure that the data is associated with feature sets at other levels (such as molecular and cellular levels). Finally, a tumor tissue-level feature set is obtained.

[0031] S24: performing multimodal feature fusion on the tumor molecular level feature set, the tumor cell level feature set and the tumor tissue level feature set to obtain a preliminary fusion feature set; The embodiment of the present invention selects a suitable multimodal data fusion method for feature sets at different levels. Feature sets at different levels are aligned and matched to ensure that the molecular, cellular and tissue-level features of each patient are correctly associated. For some missing features, they can be processed by interpolation or filling methods. The tumor molecular level feature set, tumor cell level feature set and tumor tissue level feature set are spliced ​​to form a unified preliminary fusion feature set. This step can be achieved by matrix splicing or vector splicing to ensure that all features are integrated into a unified representation. Use principal component analysis (PCA), t-SNE or other dimensionality reduction techniques to reduce the dimensionality of the preliminary fusion feature set to reduce the dimension of the feature space, improve computational efficiency, and retain key information of the data.

[0032] S25: Perform correlation analysis and verification on the preliminary fusion feature set to generate a tumor fusion feature set.

[0033] In order to verify the correlation between different features, the embodiment of the present invention selects appropriate statistical methods and machine learning techniques for analysis. Common correlation analysis methods include Pearson correlation coefficient, Spearman rank correlation coefficient, mutual information, partial least squares regression (PLSR) and principal component analysis (PCA). According to the type of feature set and the analysis target, a suitable correlation analysis method is selected. The correlation between different features is calculated using the selected correlation analysis method. The correlation analysis results are visualized to more intuitively show the correlation between different features. According to the results of the correlation analysis, further verification is performed to ensure the applicability of the features in the preliminary fusion feature set in the field of tumors. The stability and robustness of the correlation analysis are evaluated by cross-validation, especially when the data is complex or unbalanced. A statistical test method is used to verify whether the obtained correlation is statistically significant. Based on the results of the correlation analysis, important features are selected for further study, and redundant or low-correlation features are removed, thereby improving the effect and efficiency of subsequent models. Based on the correlation analysis and verification results, a tumor fusion feature set is finally generated.

[0034] The present invention extracts the molecular, cellular and tissue-level features of the tumor by respectively analyzing gene sequencing, clinical tests, pathological images and medical imaging data in detail. This multi-level feature extraction ensures that the biological characteristics of the tumor are fully captured from different angles, and improves the depth and breadth of data analysis. By fusing the features at the molecular, cellular and tissue levels, a preliminary fusion feature set is generated. This process can complement the advantages of different types of data, achieve maximum information integration, enhance the expressiveness of the feature set, and help to mine more comprehensive and accurate tumor features. By performing correlation analysis and verification on the preliminary fusion feature set, the rationality and effectiveness of the features are ensured. This analysis step can identify and confirm the intrinsic relationship between the features, improve the accuracy of the data set, exclude redundant and irrelevant features, and finally generate a tumor fusion feature set. Through feature extraction, fusion and verification, high-quality, multi-dimensional feature data is provided for the subsequent establishment of a prediction model. This fusion of multimodal features not only helps to improve the accuracy of tumor knowledge mining, but also helps to improve the scientificity and personalization of tumor medical research.

[0035] Preferably, the gene sequencing analysis in step S2 comprises the following steps: Perform gene sequencing mutation feature analysis on standardized tumor multi-source heterogeneous data sets to obtain gene mutation feature matrix data; The embodiment of the present invention preprocesses the gene sequencing data in the standardized tumor multi-source heterogeneous data set to ensure the integrity and accuracy of the data. The processed gene sequencing data is subjected to mutation feature analysis. The sequencing data is aligned to the reference genome, and a common alignment tool (such as BWA or Bowtie) is used to generate an alignment file (such as a BAM file). A mutation detection tool (such as GATK or MuTect) is used to identify common mutation types, including single nucleotide variations (SNVs), insertions and deletions (InDels), and structural variations (SVs). The detected mutations are functionally annotated, and the functional impact of the mutations and their potential role in tumors are analyzed using databases (such as COSMIC and dbSNP). According to the alignment and mutation identification results, gene mutation features are extracted and gene mutation feature matrix data is constructed. The row represents the sample ID; the column represents the gene (or mutation site); and the matrix value is the state of a gene or mutation in the sample. Based on the number of samples and the sparsity of gene mutation features, redundant or rare mutation features are eliminated to reduce feature noise. To ensure the accuracy of the gene mutation feature matrix, the extracted features are verified. Ensure that each sample contains complete gene mutation information without missing data. Statistics of gene mutation distribution characteristics, such as high-frequency mutation genes and the number of mutations among samples. By comparing with known tumor mutation feature libraries (such as TCGA and ICGC), the biological relevance of mutation features is confirmed. Finally, gene mutation feature matrix data is obtained.

[0036] Perform transcriptome expression profiling analysis on standardized tumor multi-source heterogeneous data sets to obtain gene expression profile feature matrix data; The embodiment of the present invention preprocesses the transcriptome data in the standardized tumor multi-source heterogeneous data set to ensure the integrity and consistency of the data. Perform expression profile analysis on the preprocessed transcriptome data. Align the transcriptome sequencing data to the reference genome or reference transcriptome using an alignment tool (such as HISAT2 or STAR). Generate an alignment file (such as a BAM file) to record the alignment of each transcript. Use an expression calculation tool (such as StringTie or featureCounts) to quantitatively analyze the alignment results. Generate a gene expression profile matrix in units of transcripts per million transcripts or transcripts per kilobase of transcripts. Use a differential analysis tool (such as DESeq2 or edgeR) to compare gene expression levels between different sample groups and identify significantly different genes. According to the transcriptome expression profile data, extract expression features and construct gene expression profile feature matrix data. The row represents the sample ID; the column represents the gene; the matrix value is the expression amount of the gene in the sample (such as TPM or FPKM value). Screen characteristic genes with significant biological significance or tumor-related, and eliminate noise or insignificant gene expression features. To ensure the accuracy and credibility of the gene expression profile feature matrix data, the extracted expression features are verified. Finally, the gene expression profile feature matrix data is obtained.

[0037] Perform epigenetic feature analysis on standardized tumor multi-source heterogeneous data sets to obtain epigenetic feature matrix data; The embodiment of the present invention preprocesses the epigenetic data in the standardized tumor multi-source heterogeneous data set to ensure the integrity and consistency of the data. Perform feature analysis according to the epigenetic data type. Align methylation sequencing data (such as WGBS or RRBS data) to the reference genome, using alignment tools (such as Bismark or BSMAP). Generate methylation level data for CpG sites, and record the coverage depth and methylation ratio of methylation sites. Calculate the methylation level of the promoter region and functional elements (such as enhancers) of the target gene. Construct a methylation feature matrix for the sample, featuring CpG sites. Align ATAC-seq data to generate chromatin openness feature data. Extract the signal intensity of the chromatin open region to identify active promoter and enhancer regions. Construct a chromatin openness feature matrix, featuring genomic regions. Feature extraction is performed on histone modification data (such as ChIP-seq data) to determine the peak region of histone modification, such as H3K27ac and H3K4me3. Extract the modification signal intensity of specific regions of the genome. Different types of epigenetic group data are integrated to generate epigenetic feature matrix data. Rows represent sample IDs; columns represent epigenetic sites or gene regions; matrix values ​​are methylation levels, chromatin open signals, or histone modification signals. Redundant or non-significantly different features are eliminated, and key epigenetic features related to tumors are retained. To ensure the accuracy and biological significance of epigenetic feature matrix data, validation is performed. Finally, epigenetic feature matrix data is obtained.

[0038] The gene mutation feature matrix data, gene expression spectrum feature matrix data and epigenetic feature matrix data were integrated, and molecular feature quality control was performed to obtain the tumor molecular level feature set.

[0039] The embodiment of the present invention performs row alignment on gene mutation feature matrix data, gene expression spectrum feature matrix data and epigenetic feature matrix data based on sample ID to ensure that each row corresponds to the same patient sample. Feature columns are aligned based on genes or sites to eliminate duplicate or conflicting features. The aligned feature matrix is ​​spliced ​​in the column direction to form an integrated feature matrix. The spliced ​​feature matrix is ​​filled with null values, and the missing values ​​are filled with the mean or median to ensure data integrity. In order to eliminate the dimensional differences between different data sources, standardization is performed. In order to reduce redundant information and computational complexity, the integrated feature matrix is ​​subjected to dimensionality reduction processing. The integrated feature set is quality evaluated and optimized, the numerical distribution of each feature is checked, and abnormal features (such as extremely sparse or severely skewed features) are eliminated. Verify whether the integrated features conform to known tumor biological laws, such as the presence of key mutation sites or specifically expressed genes. Finally, a feature set at the molecular level of the tumor is obtained.

[0040] The present invention ensures comprehensive coverage of tumor data at the molecular level by performing detailed analysis of gene mutations, gene expression profiles and epigenetic features respectively. These analyses cover multiple important molecular mechanisms of tumors, providing a comprehensive molecular basis for in-depth understanding of the occurrence and development of tumors. By integrating gene mutation feature matrix data, gene expression feature matrix data and epigenetic feature matrix data, information fusion at multiple molecular levels is achieved. This integration method not only enhances the diversity and complementarity of data, but also provides more abundant feature information for subsequent in-depth analysis and prediction. The integrated data is screened and corrected through molecular feature quality control, effectively eliminating noise data and redundant information, and ensuring a high-quality and accurate molecular feature set. This quality control process improves the credibility of the data and the accuracy of the analysis results. Through analysis and integration, an accurate and multidimensional molecular feature set is provided for subsequent tumor knowledge mining.

[0041] Preferably, step S3 comprises the following steps: Step S31: Designing a hierarchical knowledge ontology based on the tumor fusion feature set, and designing inter-layer association rules to generate a tumor hierarchical ontology model; The embodiment of the present invention designs a hierarchical framework of a tumor stratification ontology model based on the multimodal features of a tumor fusion feature set. The tumor stratification ontology model includes a molecular layer, a cell layer, a tissue layer, and a clinical layer. The main entities and attributes of each layer are clearly defined to ensure consistency within and between layers. A standardized medical terminology library (such as NCBI, SNOMED, ​​or ICD) is used to define key terms for each layer to avoid semantic conflicts. Based on the tumor fusion feature set, the correlation between different levels is analyzed. The constraint relationship between inter-layer entities and attributes is defined. Based on the correlation analysis, the derivation rules of inter-layer entities are defined. According to the hierarchical knowledge ontology design and the inter-layer association rules, a preliminary tumor stratification ontology model is generated. It is constructed in the Web Ontology Language (OWL) or Resource Description Framework (RDF) format to support semantic reasoning and query. The entity attributes of each layer and the association relationship with other layers are defined in the ontology model. The ontology reasoning tool is used to verify the logical consistency of the hierarchical ontology model to ensure that there is no conflict or redundancy. Finally, a tumor stratification ontology model is constructed.

[0042] Step S32: extracting entity information from the tumor fusion feature set based on the tumor hierarchical ontology model to obtain a tumor multi-source entity data set; The embodiment of the present invention extracts multimodal data corresponding to the hierarchical ontology model from the tumor fusion feature set, including molecular layer feature data, cell layer feature data, tissue layer feature data and clinical layer feature data. The tumor hierarchical ontology model is loaded to ensure that the entity and attribute definitions are consistent with the data structure of the tumor fusion feature set. According to the hierarchical definition of the tumor hierarchical ontology model, the entity information of the corresponding layer is extracted from the tumor fusion feature set. Gene mutations, gene expression and epigenetic features are extracted and mapped to molecular layer ontology entities; cell morphology, functional features and corresponding identifiers are extracted and mapped to cell layer ontology entities; tissue structure, morphological features and their spatial distribution information are extracted and mapped to tissue layer ontology entities; patient information and clinical symptoms are extracted and mapped to clinical layer ontology entities. Detailed attribute values ​​are assigned to each entity. Check whether each entity contains all key attributes defined in the hierarchical ontology model. Ensure that the entity attributes are consistent with the corresponding hierarchical semantics. The extracted and verified entity information is stored as a standardized structured data set to obtain a tumor multi-source entity data set.

[0043] Step S33: extracting intra-layer relations from the tumor multi-source entity dataset, and identifying cross-layer entity relations based on inter-layer association rules to obtain a tumor multi-source entity relation dataset; The embodiment of the present invention designs intra-layer relationship rules based on the entity definition and attributes of each layer in the tumor hierarchical ontology model, which are used to identify the relationship between entities in each layer of the tumor multi-source entity data set. Specific rules include co-mutation relationships, gene expression regulation relationships, and epigenetic modification association relationships between genes; functional interaction relationships between cell types; spatial distribution relationships, pathological connections, and microenvironment interactions between tissue structures; and logical relationships between patient medical directions and follow-up information. According to the intra-layer relationship rules, the relationship between intra-layer entities is identified and extracted from the tumor multi-source entity data set. The entities in the tumor multi-source entity data set are mapped to the corresponding hierarchical ontology model definition. Using the designed intra-layer rules, the attribute data of each layer of entities are calculated or logically inferred to identify the relationship between entities. Annotation information is added to the identified relationship, including the relationship type, relationship strength, and data source. Based on the inter-layer association rules of the tumor hierarchical ontology model, cross-layer entity relationship recognition rules are designed. The rules include molecule-cell relationships, such as the effect of gene mutations on specific cell functions, or the regulatory effect of gene expression profiles on cell differentiation; cell-tissue relationships, such as the correspondence between cell types and tumor tissue morphological characteristics; tissue-clinical relationships, such as the association between tumor tissue characteristics and patient clinical symptoms and results; cross-layer mapping rules establish mapping relationships between entity attributes of each layer and entities of other layers. Using cross-layer rules, cross-layer entity information of tumor multi-source entity datasets is cross-analyzed. The information of cross-layer entities is analyzed for association to identify potential inter-layer relationships. The reliability of the identified cross-layer relationships is ensured through data consistency verification and statistical analysis. The identified cross-layer entity relationships are stored as standardized relationship datasets, and their relationship types and association rules are annotated. The data of intra-layer relationships and cross-layer entity relationships are integrated to generate a tumor multi-source entity relationship dataset.

[0044] Step S34: constructing a knowledge subgraph based on the tumor multi-source entity dataset and the tumor multi-source entity relationship dataset, and verifying the integrity and consistency of the knowledge subgraph to obtain a tumor knowledge subgraph set; The embodiment of the present invention is based on a tumor multi-source entity data set and a tumor multi-source entity relationship data set to construct a knowledge subgraph that can reflect the multi-source heterogeneous information of a tumor. The objectives include representing entities and their relationships at the molecular level, cellular level, and tissue level of the tumor; and reflecting the association information between cross-level entities. Entity information is extracted from the tumor multi-source entity data set to generate nodes, each of which represents an entity. Hierarchical labels and attribute information are added to the nodes according to the hierarchical knowledge ontology model, including the molecular level, cellular level, tissue level, and clinical level. Relationship information is extracted from the tumor multi-source entity relationship data set to generate edges, each of which represents the relationship between nodes. Relationship type labels and weights are added to the edges according to the relationship source and type (such as intra-layer relationship or cross-layer relationship). Attribute values ​​in the tumor multi-source entity data set and the tumor multi-source entity relationship data set are mapped to nodes and edges, including time attributes, spatial attributes, and functional attributes of specific entities. Nodes and edges are grouped according to hierarchical logic to form multiple knowledge subgraphs, each of which corresponds to a specific layer or inter-layer association. Verify whether the constructed knowledge subgraph contains complete entity and relationship information to ensure that the subgraph can accurately reflect the characteristics of tumor multi-source heterogeneous data. Check whether the nodes of the knowledge subgraph completely cover the entities in the tumor multi-source entity dataset. Ensure that all hierarchical entities are correctly mapped to nodes. Check whether the edges of the knowledge subgraph completely cover the relationships in the tumor multi-source entity relationship dataset. Ensure that all intra-layer relationships and cross-layer relationships are correctly mapped to edges. Check whether the nodes and edges contain all required attributes and corresponding attribute values. Verify whether the constructed knowledge subgraph conforms to the hierarchical relationships and association rules defined in the hierarchical knowledge ontology model to ensure the logical consistency of the subgraph structure. Check whether the hierarchical labels of nodes and edges conform to the definition of the hierarchical knowledge ontology model. Ensure that entities and relationships at different levels meet the logical constraints of the inter-layer association rules. Check whether the semantic information of nodes and edges is consistent with the data source of entities and relationships. Verify whether nodes and edges of the same type in the subgraph have consistent semantic descriptions. Check whether there are cyclic relationships or isolated nodes in the subgraph that do not conform to the hierarchical logic. The knowledge subgraph that passes the integrity verification and consistency verification is stored as the tumor knowledge subgraph set.

[0045] Step S35: inter-layer connection and graph index construction are performed on the tumor knowledge subgraph, and time series attributes are added to obtain an initial hierarchical knowledge graph; The embodiment of the present invention performs inter-layer connection operations on each subgraph in the tumor knowledge subgraph set to ensure that subgraphs at different levels can establish connectivity through cross-level relationships to form an overall graph structure. Based on the cross-level entity relationships in the tumor multi-source entity relationship data set, cross-level edges are added between knowledge subgraphs. The direction and weight of the edge are determined, and semantic labels are attached according to the relationship type. According to the inter-layer association rules in the tumor hierarchical ontology model, the nodes and edges in the knowledge subgraphs of each layer are integrated to avoid redundant connections. Ensure that the semantic information of the inter-layer connection conforms to the logical rules of the knowledge subgraph and retains the hierarchical structural features. Generate an efficient indexing mechanism for the initial hierarchical knowledge graph to support subsequent fast query and retrieval. Index the graph nodes according to the entity type. Including attribute indexes (such as timestamps, locations) and semantic indexes (such as entity categories). Index the edges according to the relationship type (such as intra-layer relationships, cross-level relationships). Add weight and semantic information indexes to facilitate fast queries based on relationship importance. Logically partition the graph by level or field (such as gene, imaging, pathology), and establish partition indexes to optimize distributed storage and computing. Attach time series attributes to nodes and edges in the initial hierarchical knowledge graph to support time series analysis and dynamic knowledge update. Attach a timestamp to each node to record the creation or modification time of the node entity data. Attach time series data for entities with dynamically changing characteristics (such as gene expression levels). Attach time attributes to edges to record the creation time and validity period of entity relationships. For cross-level relationships, support time-based contextual analysis. Use a unified time format and time interval unit to ensure the consistency of time series attributes in the entire graph. Generate the initial hierarchical knowledge graph through inter-layer connections, graph index construction, and addition of time series attributes.

[0046] Step S36: Construct a confidence score based on the initial hierarchical knowledge graph, and evaluate the knowledge reliability based on the confidence score to generate a tumor hierarchical knowledge graph.

[0047] The embodiment of the present invention constructs confidence scores for each node and edge based on the initial hierarchical knowledge graph to evaluate the reliability of knowledge and help screen out high-quality knowledge. A confidence value is assigned to each entity node according to the source, data quality and verification degree of the node. Entities with higher source credibility (such as clinical data from authoritative sources) are given higher scores. The update frequency and historical verification information of the node can also be used as a reference for confidence. Scoring is performed based on the relationship strength, stability and verification degree of cross-level associations of the edge. When the confidence of the cross-level relationship is low, more evidence is needed. The weight of the edge (such as the influence of gene mutations on tumors) should also be taken into account. Combined with the machine learning model, the confidence is comprehensively scored according to different features. Based on the confidence score, the overall reliability of the tumor hierarchical knowledge graph is evaluated to ensure that the extracted knowledge is consistent with the clinical and biological background. A minimum confidence threshold is set for nodes and edges. Nodes and edges below this threshold are considered unreliable and filtered or marked as pending verification. The entire knowledge graph is checked for consistency to ensure that entities and relationships at all levels remain consistent under multimodal data. According to the evaluation results, the unreliable parts are corrected or supplemented with new data to continuously improve the credibility of the knowledge graph. After confidence scoring and reliability evaluation, the final tumor hierarchical knowledge graph is generated.

[0048] The present invention constructs a hierarchical ontology model of tumors by designing a hierarchical knowledge ontology based on a tumor fusion feature set and designing inter-layer association rules. This process provides a clear structural framework for subsequent knowledge extraction and entity relationship identification, ensuring the systematic and hierarchical nature of tumor knowledge. The tumor hierarchical ontology model is used to extract entity information from the tumor fusion feature set to obtain a tumor multi-source entity data set. This step provides comprehensive and specific tumor data by accurately extracting entity information related to tumors, which is convenient for subsequent analysis and reasoning. By extracting intra-layer relationships from the tumor multi-source entity data set and identifying cross-level entity relationships according to inter-layer association rules, the entity relationships in the tumor knowledge graph are further enriched. This process not only deepens the correlation between multi-level entities of tumors, but also strengthens the integrity of tumor information. By constructing a tumor knowledge subgraph and performing integrity and consistency verification, the high quality and accuracy of the tumor knowledge subgraph are ensured. This step effectively eliminates erroneous or inconsistent information in the data and improves the credibility of the knowledge graph. The initial hierarchical knowledge graph is obtained by constructing inter-layer connections and graph indexes, and adding time series attributes. This step improves the operability and scalability of the knowledge graph, making it more adaptable to complex tumor research and clinical applications. Based on the confidence score construction and reliability evaluation, the high credibility of the tumor hierarchical knowledge graph is ensured. Through this evaluation, the most valuable knowledge can be screened out, providing a more reliable basis for clinical decision-making and research.

[0049] Preferably, step S36 includes the following steps: Step S361: Design a multi-dimensional confidence scoring standard based on the initial hierarchical knowledge graph to obtain confidence scoring standard data; In order to ensure the credibility of the tumor hierarchical knowledge graph, the embodiment of the present invention designs a multi-dimensional confidence scoring standard for quantifying the reliability of nodes, relationships and paths. Different weights are assigned according to the source of the node data, and the authoritative data source is given a higher confidence. The node is scored according to the degree of verification of the existing literature or research, and the verified data has a higher credibility. The strength of the relationship is evaluated by statistically analyzing the frequency and consistency of the relationship. Common and consistent relationships are given a higher score. When the confidence of the cross-level entity relationship is low, it is necessary to further improve the score through additional data or verification. The score is scored according to the logic of the path and the support of multi-source data, and the path connection should have biological or clinical rationality. The temporal relationship involved in the path needs to conform to the clinical or biological development law, and a high score is given to the temporal path that conforms to the actual situation. The scoring criteria of each dimension are quantified into a numerical range (such as an interval of 0 to 1), and the scoring criteria of each dimension are comprehensively considered by weighted average to form an overall confidence scoring standard data.

[0050] Step S362: Calculate the node confidence, relationship confidence and path confidence of the initial hierarchical knowledge graph according to the confidence scoring standard data to obtain graph confidence scoring data; The embodiment of the present invention performs the calculation of node confidence, relationship confidence and path confidence of the initial hierarchical knowledge graph according to the confidence scoring standard data. For each node, the confidence of the node is calculated based on the authority of its source data and the completeness and accuracy of the data. Check whether each node has been verified by multiple independent studies or databases, and improve the confidence of the node by the number of verifications and consistency. For the relationship between each pair of entities, its confidence is evaluated by counting its frequency of occurrence in multiple data sets and studies, as well as the consistency of the relationship in different contexts. Frequent and consistent relationships will receive a higher score. For cross-level entity relationships (such as connections from the molecular level to the cellular level), the corresponding confidence is assigned according to the degree of cross-level verification and the strength of the supporting data. The coherence of the path is evaluated based on the rationality and consistency of the nodes and their relationships involved in the path. Paths with reasonable logic and sufficient supporting data will obtain higher confidence. For paths involving temporal relationships, check whether they conform to the laws of biological or clinical development, and paths that conform to actual biological timing will obtain higher confidence scores. The calculated node confidence, relationship confidence, and path confidence will be integrated into the graph confidence score data.

[0051] Step S363: Use the graph confidence score data to perform knowledge reliability evaluation on the initial hierarchical knowledge graph to obtain graph reliability evaluation report data; The embodiment of the present invention performs a knowledge reliability assessment of the initial hierarchical knowledge graph based on the graph confidence score data. According to the confidence score of the node, each node is analyzed to see whether there is data missing, information redundancy or conflict. For nodes with lower confidence, priority review and verification are performed to determine whether further correction or deletion is required. The reliability of the relationship between the nodes in the graph is analyzed, including the consistency of the data source and the biological rationality of the relationship. Relationships with high confidence will be considered reliable, while relationships with low confidence require further verification. The logic and temporal rationality of the paths in the graph are assessed. For those high-confidence paths, their consistency in clinical or experimental data is further analyzed to assess whether they have sufficient knowledge reliability. Based on the confidence score data, different weights are assigned to each evaluation dimension. Combined with the evaluation results of nodes, relationships and paths, the overall reliability score of the entire initial hierarchical knowledge graph is calculated. The score reflects the accuracy, completeness and credibility of the knowledge graph. Through the above steps, the graph reliability assessment report data is generated.

[0052] Step S364: Use the graph reliability assessment report data to modify the initial hierarchical knowledge graph to obtain a tumor hierarchical knowledge graph.

[0053] The embodiment of the present invention amends the initial hierarchical knowledge graph based on the graph reliability assessment report data. Through the node reliability data in the assessment report, those nodes with low confidence are identified. For these nodes, the quality and consistency of their source data are checked, and the node information is amended based on the results of data update or re-examination. For nodes with missing data, relevant external data sources are found for supplementation; for redundant or erroneous nodes, they are deleted or replaced. According to the relationship confidence score in the graph, the low-confidence relationship is re-examined. Inconsistent or erroneous relationships are corrected to ensure the biological rationality and data consistency of the relationship. The definition of cross-level relationships is adjusted to ensure its cross-level accuracy. Low-confidence paths are identified through the path confidence score in the assessment report. Check the key nodes and relationships in the path, and correct those paths caused by incomplete or erroneous data. For paths involving time series attributes, ensure that they conform to the actual biological situation, and make necessary corrections or additions to the time series attributes. Based on the above corrections, the initial hierarchical knowledge graph is updated to generate a tumor hierarchical knowledge graph.

[0054] The present invention ensures the comprehensiveness and scientificity of the scoring system by designing a multi-dimensional confidence scoring standard based on the initial hierarchical knowledge graph. By setting independent scoring standards for different dimensions (such as nodes, relationships, and paths), this step can carefully evaluate each element in the tumor knowledge graph, laying a solid foundation for subsequent reliability analysis. According to the confidence scoring standard data, the node confidence, relationship confidence, and path confidence of the initial hierarchical knowledge graph are calculated to obtain the confidence scoring data of the graph. This step accurately evaluates each element in the graph through a comprehensive scoring system, ensures a detailed analysis of the quality of the knowledge graph, and thus improves the reliability of the data. The graph confidence scoring data is used to evaluate the knowledge reliability and generate the graph reliability evaluation report data. The evaluation process can clarify the weak links in the graph and help identify the parts with low credibility, thereby providing guidance for further optimization of the knowledge graph. The initial hierarchical knowledge graph is corrected based on the graph reliability evaluation report data to obtain a tumor hierarchical knowledge graph. Through this correction step, the accuracy and reliability of the tumor knowledge graph have been effectively improved, ensuring that the final knowledge graph can provide high-quality information support for research and clinical applications in the field of tumors.

[0055] Preferably, step S4 comprises the following steps: Step S41: constructing a heterogeneous graph neural network model using a tumor hierarchical knowledge graph to obtain a tumor graph neural network model; In the embodiment of the present invention, each entity in the tumor hierarchical knowledge graph is regarded as a node in the graph. The relationship between the nodes is modeled as an edge in the graph. For the heterogeneous node and edge types in the tumor knowledge graph, the technology of heterogeneous graph neural network is adopted to process different types of nodes and edges through different graph neural network layers. The feature information of adjacent nodes is aggregated through a multi-layer graph convolutional network (GCN) or other types of graph neural networks. Each layer of graph neural network can generate a higher level representation based on its node features and adjacency relationships. In the feature representation learning of different types of nodes, different types of node and relationship information are integrated to ensure that the graph neural network can process heterogeneous data and retain the complex semantic relationship between nodes. Based on the graph neural network model, node embedding vectors of the tumor hierarchical knowledge graph are generated through node feature aggregation and graph convolution operations. These embedding vectors can effectively represent each node in the graph and capture the complex relationship between nodes. Through the multi-level graph neural network structure, the embedding vector of each node is further refined so that in the subsequent reasoning and analysis process, the high-dimensional feature information of nodes and relationships can be used for more accurate reasoning. Finally, a tumor graph neural network model is obtained.

[0056] Step S42: Designing a cancer cell prediction task based on the tumor fusion feature set to obtain cancer cell dynamic prediction task data; The embodiment of the present invention clarifies the task objectives of dynamic prediction of cancer cells according to the needs of tumor research, including but not limited to time dimension prediction, space dimension prediction and clinical dimension prediction. Extract features related to the cancer cell prediction task from the tumor fusion feature set. Extract biomarker indicators from the patient's historical clinical test data as basic predictors; extract cancer cell morphological features obtained by image processing from case medical imaging data; extract symptom descriptions from tumor clinical symptom data as auxiliary features. Define prediction task labels according to task objectives; divide data groups by samples to ensure that the distribution of training data and test data is consistent; for data sets with unbalanced labels, use undersampling or oversampling methods to balance sample distribution. Format and preprocess the task data to ensure that the data adapts to the input of the tumor graph neural network model. Finally, the task data for dynamic prediction of cancer cells is obtained.

[0057] Step S43: Based on the cancer cell dynamic prediction task data, a tumor graph neural network model is used to perform cancer cell dynamic prediction to obtain tumor dynamic knowledge reasoning result data.

[0058] The embodiment of the present invention uses cancer cell dynamic prediction task data as input to ensure that the data format meets the requirements of the tumor graph neural network model. In the tumor graph neural network model, each node transmits and aggregates information through a graph structure. According to the goal of the cancer cell dynamic prediction task, the model will perform multi-level feature extraction and aggregation on the cancer cell features of the node to obtain a high-order representation of each node (tumor-related entity). For cancer cell dynamic prediction tasks involving time evolution, the model processes time dependencies through a temporal graph convolutional network (such as T-GCN), combines historical data with future predictions, and captures the dynamic change trend of cancer cells. The model also considers data from different modalities (such as clinical, imaging, and test data) for fusion, performs weighted aggregation on node features, and improves prediction accuracy. After completing information transmission and aggregation, the tumor graph neural network model will output cancer cell dynamic prediction results, including cancer cell growth prediction, metastasis risk assessment, and prognostic response prediction. The predicted results such as cancer cell dynamic changes and metastasis paths are mapped back to the original tumor map to generate structured reasoning result data, and obtain tumor dynamic knowledge reasoning result data.

[0059] The present invention utilizes the construction of a tumor hierarchical knowledge graph and a heterogeneous graph neural network model. The method can effectively fuse different types of tumor data and accurately predict the dynamic changes of cancer cells, which provides a more accurate prediction basis for tumor medical research. By designing cancer cell prediction tasks based on tumor fusion feature sets, it is possible to analyze the behavioral changes of cancer cells from multiple angles, identify potential cancer cell development patterns, and improve the accuracy of intervention. Using the tumor graph neural network model for dynamic prediction and knowledge reasoning, the development trend of the tumor can be evaluated in real time. In the process of cancer cell prediction, deep learning is combined with the knowledge graph in the field of tumors to provide scientific data support for researchers and clinicians in the field of tumors, and promote the application of personalized cancer medical research.

[0060] Preferably, step S5 comprises the following steps: Step S51: integrating multi-level features of patients based on the tumor dynamic knowledge reasoning result data to obtain a tumor patient feature data set; The embodiment of the present invention comprehensively establishes a multi-dimensional feature database of patients by integrating the reasoning results in the tumor dynamic knowledge reasoning result data and combining the basic information and clinical symptoms of the patients. According to the different characteristics of the tumor, the patient's characteristics are divided into multiple levels, including clinical level characteristics, molecular level characteristics, imaging level characteristics and prognostic response level characteristics. The various types of collected data are cleaned and standardized to remove invalid data, missing values ​​and outliers to ensure the accuracy and consistency of the data. Data from different sources are uniformly coded and standardized for further integration and analysis. The features of each level are integrated using a data fusion method. Based on the reasoning results in the tumor dynamic knowledge graph, the correlation analysis of the features is performed to identify the potential relationship between the features of each level, thereby constructing a comprehensive tumor patient feature data set. After integration, a tumor patient feature data set is obtained.

[0061] Step S52: designing a personalized molecular feature map based on the tumor patient feature data set, and performing specific marker screening to obtain a molecular feature map of the tumor patient; The embodiment of the present invention extracts the molecular feature information of the patient based on the tumor patient feature data set. Through statistical analysis and machine learning algorithms, the influence of different features is evaluated, and molecular features with high correlation with tumors are screened out. According to the individual characteristics and pathological status of tumor patients, a personalized molecular feature map is designed. By mapping the key molecular features of each patient to the map, a highly personalized molecular map is constructed. Each node represents a molecular feature, and the relationship between nodes represents their mutual influence or association. Combined with the pathological data, clinical response and changes in biological markers of tumor patients, the hierarchical structure of the molecular feature map is further refined. The nodes and edges of each level represent different types of biological information and patient characteristics to form a comprehensive patient molecular feature map. Using known tumor marker databases and related literature, combined with patient molecular feature maps, marker screening is performed. Through comparative analysis, it is determined which molecular markers have an important predictive effect on the patient's tumor type and prognosis response. Bioinformatics tools and experimental verification methods are used to verify the screened potential markers to ensure their reliability and effectiveness in specific tumor types. Finally, a molecular feature map of tumor patients is obtained.

[0062] Step S53: performing cancer cell inhibition analysis according to the tumor patient analysis characteristic graph to obtain cancer cell inhibition recommendation data for the tumor patient; The embodiment of the present invention clarifies the goal of cancer cell inhibition analysis based on the molecular feature map of tumor patients, including but not limited to molecular target selection, drug screening and optimization of medical research direction. Based on the key molecular markers identified in the molecular feature map of tumor patients, the association analysis of cancer cell inhibition targets is performed. Relevant gene or protein data is extracted from the molecular feature map of the patient, and markers related to cancer cell proliferation, metastasis and apoptosis are screened out. The target molecules are predicted by computational models, and verified in combination with literature and databases to evaluate the inhibition potential of the targets. Gene regulatory networks are constructed to analyze the gene expression patterns in tumor cells and find key nodes that regulate cancer cell proliferation. Drug screening and medical research direction recommendations are performed using the information in the molecular feature map. According to the molecular characteristics of the patient, combined with the drug library (including targeted drugs and immunotherapy drugs), drugs that match the characteristics of the patient's cancer cells are screened out. Based on the molecular feature map of the patient, personalized drug recommendations are provided using the drug sensitivity database and drug response prediction model. Medical research plans are designed for multiple molecular targets. The inhibition effect of cancer cells is obtained by simulating and evaluating the inhibition effect of different drug combinations by establishing a tumor growth model or using computational biology methods. Based on the molecular characteristics of the patient, the inhibition effect of different drugs or drug combinations is predicted. Evaluate drug sensitivity through in vitro experimental data, historical clinical data or computational models to obtain personalized drug response prediction results. Analyze the correlation between cancer cell inhibition effect and specific molecular markers. Generate cancer cell inhibition recommendation data for tumor patients based on cancer cell inhibition analysis results.

[0063] Step S54: performing a prognostic risk assessment on the cancer cell inhibition suggestion data of the tumor patient to obtain cancer cell recurrence risk assessment data of the tumor patient; The embodiment of the present invention integrates relevant clinical data, medical research information and personalized molecular feature data of patients according to cancer cell inhibition recommendation data of tumor patients as input for prognostic risk assessment. Based on the cancer cell inhibition recommendation data and the molecular features of tumor patients, a tumor recurrence risk assessment model is constructed. The patient's historical medical records, cancer cell inhibition schemes, and molecular feature map data are processed to extract features related to cancer cell recurrence. Through statistical analysis and machine learning methods, molecular markers and clinical features significantly associated with the risk of cancer cell recurrence are found. An appropriate algorithm is selected to train the prediction model. The stability and accuracy of the model are evaluated using cross-validation. The constructed prognostic risk assessment model is used to predict the risk of cancer cell recurrence in tumor patients. The patient's cancer cell inhibition recommendation data, molecular feature map and other relevant data are input into the evaluation model. The model calculates the probability of cancer cell recurrence and the estimated recurrence time based on the patient's individual data, and evaluates the prognostic risk. The prediction results obtained by the prognostic risk assessment model are used to perform recurrence risk classification and recurrence time prediction, and provide key factors related to cancer cell recurrence. Based on the above evaluation results, cancer cell recurrence risk assessment data for tumor patients is generated.

[0064] Step S55: Integrate the molecular feature maps of tumor patients, the cancer cell inhibition recommendation data of tumor patients, and the cancer cell recurrence risk assessment data of tumor patients to obtain a personalized analysis report.

[0065] The embodiment of the present invention pre-processes the molecular feature map of tumor patients, the cancer cell inhibition recommendation data of tumor patients, and the cancer cell recurrence risk assessment data of tumor patients, including removing duplicate data, filling missing values, and standardizing data formats. The multi-dimensional information such as the molecular characteristics, medical research recommendations, and recurrence risk of tumor patients is comprehensively analyzed. Based on the integrated data, a personalized analysis report is automatically generated. The generated personalized analysis report will be displayed in a combination of charts, images, and text descriptions.

[0066] The present invention integrates the tumor dynamic knowledge reasoning result data and performs multi-level feature integration, which can integrate various information of patients and construct a detailed tumor patient feature data set. Based on the patient's feature data, a personalized molecular feature map is designed and specific markers are screened to reduce unnecessary side effects. Through cancer cell inhibition analysis and prognostic risk assessment, the cancer cell inhibition status and recurrence risk of tumor patients can be accurately assessed, and response strategies can be formulated in advance to reduce the risk of cancer cell recurrence. The patient's molecular feature map, cancer cell inhibition recommendations and recurrence risk assessment results are integrated to generate a personalized analysis report to support the dynamic adjustment of personalized medical research.

[0067] Preferably, step S52 includes the following steps: Step S521: constructing a molecular interaction network based on the tumor patient feature data set, and performing molecule-microenvironment interaction mapping to obtain the tumor patient molecular interaction network data; The embodiment of the present invention collects comprehensive patient characteristics based on a tumor patient characteristic data set, including the patient's genomic information, transcriptome data, proteome data and clinical background data. The molecular interaction network of tumor patients is constructed using the gene and protein information in the tumor patient characteristic data set. The known interaction relationship between genes is extracted through literature or databases to construct a gene-gene interaction network. Based on the proteome data of tumor patients, a protein interaction network is constructed through protein-protein interaction (PPI) data. The constructed molecular interaction network is mapped with the relevant data of the tumor microenvironment to reveal the interaction relationship between tumor cells and their microenvironment. Finally, the molecular interaction network data of tumor patients is obtained.

[0068] Step S522: performing signal pathway abnormality analysis on the molecular interaction network data of tumor patients to obtain a pathway abnormality map of tumor patients; The embodiment of the present invention uses the molecular interaction network data of tumor patients to perform signal pathway analysis on the molecular interaction network of tumor patients through software tools (such as PathwayStudio, Reactome or KEGG). Determine the biological pathways that are significantly changed in the tumor patient data. According to the patient data, identify which signal pathway activities are abnormally regulated. Based on the above signal pathway analysis results, construct a pathway abnormality map of tumor patients.

[0069] Step S523: identifying key regulatory factors according to the abnormal pathway map of the tumor patient, and using the key regulatory factors to perform regulatory pattern analysis to obtain a regulatory pattern map of the tumor patient; The embodiment of the present invention uses an algorithm (such as machine learning or statistical methods) to identify key regulatory factors in each signal pathway based on the abnormal pathway map of tumor patients. Key regulatory factors are usually molecules that have a significant impact on the function of the pathway. By comparing the patient group data, the important role of these factors in tumor progression is identified. Using the identified key regulatory factors, a regulatory model is constructed. The regulatory model analysis aims to reveal how these factors interact and regulate the signal pathways of tumor patients. Finally, a regulatory model map of tumor patients is obtained.

[0070] Step S524: Using the tumor patient regulation pattern map to screen molecular markers, obtain a tumor patient molecular characteristic map.

[0071] The embodiment of the present invention performs molecular marker screening based on the tumor patient regulation pattern map. The goal of screening is to find molecular features related to tumor development, prognosis, drug response, etc. Identify molecules that are differentially expressed in tumor tissue and normal tissue. Select molecules that play a significant role in the regulatory network, such as key regulatory factors or regulatory molecules. Use network topology analysis methods to screen out molecules in the core position of the regulatory network. Through screening, a molecular feature map of tumor patients is obtained.

[0072] The present invention can reveal the complex interactions between tumor cells and the microenvironment by constructing a molecular interaction network based on a characteristic data set of tumor patients and performing molecular-microenvironment interaction mapping. This step helps to deeply understand the impact of the tumor microenvironment on tumor development and provides a theoretical basis for subsequent personalized medical research. By performing signal pathway abnormality analysis on molecular interaction network data, abnormal conditions in the signal transduction process of tumor patients can be identified. This analysis helps to reveal the key pathways of tumor occurrence and progression and provide direction for targeted intervention. By analyzing the pathway abnormality map of tumor patients, key regulatory factors are identified, and these factors are used to perform regulatory pattern analysis. Accurately identify important molecules in tumors and their regulatory mechanisms, and provide strong support for the design of personalized medical research strategies. Molecular markers are screened using the regulatory pattern map of tumor patients to obtain accurate molecular feature maps of tumor patients.

[0073] Preferably, the present invention further provides a knowledge mining system for the tumor field, which is used to execute the knowledge mining method for the tumor field as described above. The knowledge mining system for the tumor field comprises: A standardized data preprocessing module is used to obtain a tumor multi-source heterogeneous data set and perform data standardization processing to obtain a standardized tumor multi-source heterogeneous data set; A multi-level feature extraction module is used to perform multi-level feature extraction on standardized tumor multi-source heterogeneous data sets to obtain a tumor fusion feature set; A hierarchical knowledge graph construction module is used to construct a hierarchical knowledge graph based on the tumor fusion feature set to obtain a tumor hierarchical knowledge graph; The graph neural network dynamic prediction module is used to build a graph neural network model using the tumor hierarchical knowledge graph and to dynamically predict cancer cells to obtain tumor dynamic knowledge reasoning result data; The personalized analysis and prediction module is used to conduct personalized analysis of patients based on tumor dynamic knowledge reasoning result data, predict the probability of cancer cell inhibition, and generate personalized analysis reports.

[0074] The present invention can convert multi-source heterogeneous tumor data sets into unified standard data sets through a standardized data preprocessing module, providing a clear and accurate data basis for subsequent analysis. Through a multi-level feature extraction module, it is possible to deeply mine the key features in the data to form a tumor fusion feature set, thereby more comprehensively analyzing the multi-dimensional characteristics of the tumor. The hierarchical knowledge graph construction module effectively integrates knowledge in the field of tumors, provides structured knowledge support for cancer cell prediction and personalized medical research, and helps to form a more scientific basis for decision-making. The graph neural network dynamic prediction module and the personalized analysis and prediction module can achieve accurate dynamic prediction of cancer cells through the application of hierarchical knowledge graphs of tumors, thereby providing personalized medical research directions and cancer cell inhibition probability predictions.

[0075] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is not limited by the above description, so it is intended to include all changes within the meaning and scope of the equivalent elements of the application documents in the present invention.

[0076] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A knowledge mining method for the field of tumors, characterized in that: The following steps are involved: Step S1: obtaining a tumor multi-source heterogeneous data set, and performing data standardization processing to obtain a standardized tumor multi-source heterogeneous data set; Step S2: performing multi-level feature extraction on the standardized tumor multi-source heterogeneous data set to obtain a tumor fusion feature set; Step S3: constructing a hierarchical knowledge graph based on the tumor fusion feature set to obtain a tumor hierarchical knowledge graph; Step S4: constructing a graph neural network model using the tumor hierarchical knowledge graph, and performing dynamic prediction of cancer cells to obtain tumor dynamic knowledge reasoning result data; Step S5: Perform personalized analysis of patients based on the tumor dynamic knowledge reasoning result data, predict the probability of cancer cell inhibition, and generate a personalized analysis report.

2. The knowledge mining method for tumor field according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: collecting historical clinical test data through a medical test information system to obtain historical clinical test data; Step S12: collecting case medical image data through a medical image information system to obtain case medical image data; Step S13: collecting clinical symptom data through the electronic medical record system to obtain clinical symptom data of the tumor; Step S14: standardize the historical clinical test data, case medical imaging data and tumor clinical symptom data respectively, and merge the data to obtain a standardized tumor multi-source heterogeneous data set.

3. The knowledge mining method for tumor field according to claim 1, characterized in that: Step S2 includes the following steps: Gene sequencing analysis is performed on standardized tumor multi-source heterogeneous data sets, and molecular-level feature extraction is performed to obtain a tumor molecular-level feature set; Conduct clinical tests and pathological image analysis on standardized tumor multi-source heterogeneous data sets, and extract cell-level features to obtain tumor cell-level feature sets; Perform medical image analysis on standardized tumor multi-source heterogeneous datasets and extract tissue-level features to obtain a tumor tissue-level feature set; Perform multimodal feature fusion on the tumor molecular level feature set, tumor cell level feature set and tumor tissue level feature set to obtain a preliminary fusion feature set; Correlation analysis and verification were performed on the preliminary fusion feature set to generate the tumor fusion feature set.

4. The knowledge mining method for tumor field according to claim 3, characterized in that: The gene sequencing analysis in step S2 includes the following steps: Perform gene sequencing mutation feature analysis on standardized tumor multi-source heterogeneous data sets to obtain gene mutation feature matrix data; Perform transcriptome expression profiling analysis on standardized tumor multi-source heterogeneous data sets to obtain gene expression profile feature matrix data; Perform epigenetic feature analysis on standardized tumor multi-source heterogeneous data sets to obtain epigenetic feature matrix data; The gene mutation feature matrix data, gene expression spectrum feature matrix data and epigenetic feature matrix data were integrated, and molecular feature quality control was performed to obtain the tumor molecular level feature set.

5. The knowledge mining method for tumor field according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: Designing a hierarchical knowledge ontology based on the tumor fusion feature set, and designing inter-layer association rules to generate a tumor hierarchical ontology model; Step S32: extracting entity information from the tumor fusion feature set based on the tumor hierarchical ontology model to obtain a tumor multi-source entity data set; Step S33: extracting intra-layer relations from the tumor multi-source entity dataset, and identifying cross-layer entity relations based on inter-layer association rules to obtain a tumor multi-source entity relation dataset; Step S34: constructing a knowledge subgraph based on the tumor multi-source entity dataset and the tumor multi-source entity relationship dataset, and verifying the integrity and consistency of the knowledge subgraph to obtain a tumor knowledge subgraph set; Step S35: inter-layer connection and graph index construction are performed on the tumor knowledge subgraph, and time series attributes are added to obtain an initial hierarchical knowledge graph; Step S36: Construct a confidence score based on the initial hierarchical knowledge graph, and evaluate the knowledge reliability based on the confidence score to generate a tumor hierarchical knowledge graph.

6. The knowledge mining method for tumor field according to claim 5, characterized in that: Step S36 includes the following steps: Step S361: Design a multi-dimensional confidence scoring standard based on the initial hierarchical knowledge graph to obtain confidence scoring standard data; Step S362: Calculate the node confidence, relationship confidence and path confidence of the initial hierarchical knowledge graph according to the confidence scoring standard data to obtain graph confidence scoring data; Step S363: Use the graph confidence score data to perform knowledge reliability evaluation on the initial hierarchical knowledge graph to obtain graph reliability evaluation report data; Step S364: Use the graph reliability assessment report data to modify the initial hierarchical knowledge graph to obtain a tumor hierarchical knowledge graph.

7. The knowledge mining method for tumor field according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: constructing a heterogeneous graph neural network model using a tumor hierarchical knowledge graph to obtain a tumor graph neural network model; Step S42: Designing a cancer cell prediction task based on the tumor fusion feature set to obtain cancer cell dynamic prediction task data; Step S43: Based on the cancer cell dynamic prediction task data, a tumor graph neural network model is used to perform cancer cell dynamic prediction to obtain tumor dynamic knowledge reasoning result data.

8. The knowledge mining method for tumor field according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: integrating multi-level features of patients based on the tumor dynamic knowledge reasoning result data to obtain a tumor patient feature data set; Step S52: designing a personalized molecular feature map based on the tumor patient feature data set, and performing specific marker screening to obtain a molecular feature map of the tumor patient; Step S53: performing cancer cell inhibition analysis according to the molecular characteristic map of the tumor patient to obtain cancer cell inhibition recommendation data of the tumor patient; Step S54: performing a prognostic risk assessment on the cancer cell inhibition suggestion data of the tumor patient to obtain cancer cell recurrence risk assessment data of the tumor patient; Step S55: Integrate the molecular feature maps of tumor patients, the cancer cell inhibition recommendation data of tumor patients, and the cancer cell recurrence risk assessment data of tumor patients to obtain a personalized analysis report.

9. The knowledge mining method for tumor field according to claim 8, characterized in that: Step S52 includes the following steps: Step S521: constructing a molecular interaction network based on the tumor patient feature data set, and performing molecule-microenvironment interaction mapping to obtain the tumor patient molecular interaction network data; Step S522: performing signal pathway abnormality analysis on the molecular interaction network data of tumor patients to obtain a pathway abnormality map of tumor patients; Step S523: identifying key regulatory factors according to the abnormal pathway map of the tumor patient, and using the key regulatory factors to perform regulatory pattern analysis to obtain a regulatory pattern map of the tumor patient; Step S524: Using the tumor patient regulation pattern map to screen molecular markers, obtain a tumor patient molecular characteristic map.

10. A knowledge mining system for the field of tumors, characterized in that: Used to execute the knowledge mining method for the tumor field as claimed in claim 1, the knowledge mining system for the tumor field comprises: A standardized data preprocessing module is used to obtain a tumor multi-source heterogeneous data set and perform data standardization processing to obtain a standardized tumor multi-source heterogeneous data set; A multi-level feature extraction module is used to perform multi-level feature extraction on standardized tumor multi-source heterogeneous data sets to obtain a tumor fusion feature set; A hierarchical knowledge graph construction module is used to construct a hierarchical knowledge graph based on the tumor fusion feature set to obtain a tumor hierarchical knowledge graph; The graph neural network dynamic prediction module is used to build a graph neural network model using the tumor hierarchical knowledge graph and to dynamically predict cancer cells to obtain tumor dynamic knowledge reasoning result data; The personalized analysis and prediction module is used to conduct personalized analysis of patients based on tumor dynamic knowledge reasoning result data, predict the probability of cancer cell inhibition, and generate personalized analysis reports.

Images