AI-driven traditional Chinese medicine-glioma target point synergistic effect network prediction system

Abstract

The invention discloses an AI-driven traditional Chinese medicine-glioma target point synergistic effect network prediction system, which comprises a multi-mode biomedical data acquisition and preprocessing module, an AI-driven traditional Chinese medicine-glioma target point network prediction module and a target point prediction module, the data acquisition module is used for acquiring heterogeneous biomedical data associated with traditional Chinese medicine components and biological activity thereof, glioma related targets and molecular characteristics thereof, and traditional Chinese medicine-target interaction from a plurality of data sources; and the multi-level biological network construction module is electrically connected with the multi-modal biomedical data acquisition and preprocessing module. The invention relates to the technical field of bioinformatics. According to the AI-driven traditional Chinese medicine-glioma target point synergistic effect network prediction system, massive heterogeneous biomedical data are effectively integrated and standardized through the multi-mode biomedical data acquisition and preprocessing module, the defects that a data source is single and complex association is difficult to capture in a traditional method are overcome, and the multi-level biological network construction module has the advantages of high efficiency, high reliability and the like. A multi-dimensional and multi-scale biological network is constructed from the perspective of system biology.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics technology, specifically to an AI-driven network prediction system for the synergistic effect of traditional Chinese medicine on glioma targets. Background Technology

[0002] Gliomas, as the most common and aggressive type of primary intracranial tumors, pose a serious threat to human health due to their high heterogeneity, diffuse growth, and widespread resistance to conventional treatments (such as surgery, radiotherapy, and chemotherapy). Despite significant advancements in diagnostic and treatment methods in modern medicine, improving the prognosis of glioma patients remains a major challenge, urgently requiring the development of novel, highly effective treatment strategies with minimal side effects.

[0003] Referring to patent publication number "CN112365980B", a method and system for visualization of multi-target auxiliary diagnosis and prospective treatment evolution of brain tumors are disclosed, including: Module M1: acquiring paired multi-target multimodal MRI data of brain tumors before and after treatment and preprocessing the paired multi-target multimodal MRI data of brain tumors before and after treatment to obtain standardized paired multi-target multimodal MRI data of brain tumors before and after treatment, Ioriginal and Ilater; Module M2.

[0004] As described above, existing technologies are increasingly showing limitations in addressing complex diseases like gliomas, which are highly heterogeneous and drug-resistant, especially when it is necessary to accurately analyze the "multi-target synergistic network" of complex traditional Chinese medicine (TCM) systems. Specifically, although traditional network pharmacology methods can construct drug-target networks, they often rely on static database matching and statistical correlation analysis when processing massive and heterogeneous biomedical data. This makes it difficult to effectively capture the dynamic and nonlinear synergistic patterns between TCM components and between TCM components and disease targets. The reason for this is that the efficacy of TCM is not simply attributed to the effect of a single active ingredient on a single target, but rather to the overall effect of multiple components interacting and synergistically enhancing each other in a complex biological network. This deficiency makes it difficult to accurately assess the synergistic efficacy of potential targets in complex biological environments, even if they are identified. This limits the potential for a deeper understanding of the molecular mechanisms of TCM in combating gliomas and guiding clinical applications. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an AI-driven predictive system for the synergistic effects of traditional Chinese medicine (TCM) and glioma targets. This system solves the inherent limitations of existing technologies in predicting the deep, dynamically evolving, and non-obvious synergistic effects between complex TCM systems and glioma pathological networks using computational methods.

[0006] To achieve the above objectives, the present invention provides the following technical solution: An AI-driven network prediction system for the synergistic effect of traditional Chinese medicine and glioma targets includes: The multimodal biomedical data acquisition and preprocessing module is designed to acquire heterogeneous biomedical data related to Chinese medicine components and their bioactivity, glioma-related targets and their molecular characteristics, and Chinese medicine-target interactions from multiple data sources. A multi-level biological network construction module is used to construct multiple interconnected multi-level biological networks based on structured input data, which describe the complex relationships between traditional Chinese medicine components, glioma targets, and disease pathways. The synergistic deep learning model construction and training module is used to receive multi-layer biological networks and their initial node feature vectors, and to build and train a deep learning model based on a graph neural network architecture. The Synergistic Network Analysis and Interpretability Module is used for in-depth analysis of the prediction results of synergistic networks. The Personalization and Clinical Translation module is used to combine the analysis results of the Synergistic Network Analysis and Interpretability module with individual glioma patient-specific molecular characterization data.

[0007] Preferably, the multimodal biomedical data acquisition and preprocessing module is configured to acquire heterogeneous biomedical data from at least one of the following data sources: a database of traditional Chinese medicine components, a database of glioma targets, a database of drug-target interactions, and a database of disease pathways; The multimodal biomedical data acquisition and preprocessing module is further configured to remove missing and outlier values ​​through data cleaning algorithms, process high-dimensional data using dimensionality reduction techniques, ensure data quality and consistency, and ultimately integrate the data from different sources and in different formats into a unified set of multidimensional feature vectors and an association matrix as the structured input data.

[0008] Preferably, the multi-level biological network construction module is configured to construct at least one of the following multi-level biological networks: a Chinese medicine component similarity network, a glioma target interaction network, a Chinese medicine component-target association network, a disease pathway association network, and a Chinese medicine compound synergistic network; The multi-level biological network construction module realizes the dynamic construction and management of the network through the graph computing library, and generates an initial node feature vector for each node in the multi-level biological network. This vector combines the structured features and network topology features from the multimodal biomedical data acquisition and preprocessing module.

[0009] Preferably, the deep learning model in the synergistic deep learning model construction and training module includes a multimodal graph encoder. The multimodal graph encoder is configured to receive initial feature vectors of traditional Chinese medicine component nodes, initial feature vectors of glioma target nodes, and adjacency matrices of different network types, and includes: Multiple stacked graph attention network layers, each GAT layer can adaptively learn the attention weights between a node and its neighboring nodes, thereby capturing the local and global contextual information of nodes in different biological networks; A cross-modal feature fusion layer is set up on the node embeddings of the outputs of the multiple stacked GAT layers, with a fusion mechanism based on a Transformer encoder or a specially designed cross-attention module. A collaborative representation layer for traditional Chinese medicine compound formulas: For compound formulas involving multiple traditional Chinese medicine components, the layer aggregates the fusion vectors of all monomeric components in the compound formula through a pooling layer to generate a composite feature representation representing the interaction between the entire compound formula and the glioma target network. The aggregation process can further consider the relative concentration or dosage ratio of each component in the compound formula.

[0010] Preferably, the deep learning model in the synergistic deep learning model construction and training module further includes a multi-task prediction head, which is configured to receive the multimodal fusion vector or complex composite feature representation, and includes at least one of the following parallel prediction heads: synergistic effect strength prediction head, key target identification head, and mechanism of action pathway prediction head.

[0011] Preferably, the synergistic deep learning model construction and training module adopts at least one of the following training strategies: self-supervised pre-training, multi-task learning and transfer learning and fine-tuning; The training process employs the Adam optimizer, combined with a learning rate scheduling strategy, and uses a cross-validation strategy for evaluation. Early stopping is also employed to prevent overfitting.

[0012] Preferably, the synergistic network parsing and interpretability module is configured to perform in-depth parsing of the synergistic network prediction results using at least one of the following interpretation methods: gradient-based interpretation method, attention weight analysis, subgraph extraction and visualization, feature importance assessment and pathway enrichment analysis; The output of the synergistic network analysis and interpretability module includes, but is not limited to: a list of key synergistic Chinese medicine components and their contribution, a list of key glioma targets and their weights, a key component-target interaction map, a synergistic subnetwork topology diagram, a report on enriched signaling pathways, and a report on interpretive evidence.

[0013] Preferably, the personalization and clinical translation module is configured for: patient molecular feature integration, glioma subtype-specific prediction, synergistic risk-benefit assessment, and generation of clinical decision support reports.

[0014] Beneficial effects This invention provides an AI-driven network prediction system for the synergistic effect of traditional Chinese medicine on glioma targets. Compared with existing technologies, it has the following advantages: 1. This AI-driven network prediction system for the synergistic effect of traditional Chinese medicine and glioma targets effectively integrates and standardizes massive heterogeneous biomedical data through a multimodal biomedical data acquisition and preprocessing module. It overcomes the shortcomings of traditional methods, such as single data sources and difficulty in capturing complex relationships. The multi-level biological network construction module constructs multi-dimensional and multi-scale biological networks from the perspective of systems biology, providing structured input for deep learning models to capture network-level synergistic effects.

[0015] 2. This AI-driven network prediction system for the synergistic effect of traditional Chinese medicine (TCM) and glioma targets, by employing advanced graph attention networks and multi-task learning paradigms, can accurately model the deep, non-linear "1+1>2" synergistic effect between multiple components of TCM and multiple targets of glioma, rather than simply predicting a single effect, thus significantly improving the accuracy and generalization ability of the prediction.

[0016] 3. This AI-driven TCM-glioma target synergistic effect network prediction system, through synergistic effect network analysis and interpretability modules, uses a series of advanced interpretable artificial intelligence technologies to make the "black box" operation of the model transparent, revealing the molecular mechanism and key driving factors of synergistic effects, and providing a solid theoretical foundation for biological validation and clinical translation. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the workflow of the AI-driven network prediction system for synergistic effects of traditional Chinese medicine and glioma targets in this invention. Figure 2 This is a schematic diagram of the architecture of the synergistic deep learning model of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figures 1-2The AI-driven TCM-glioma target synergistic effect network prediction system includes: a multimodal biomedical data acquisition and preprocessing module, which is set up to acquire heterogeneous biomedical data related to TCM components and their bioactivity, glioma-related targets and their molecular characteristics, and TCM-target interactions from multiple data sources, and performs standardization, cleaning, feature extraction and unified formatting on the heterogeneous biomedical data to generate structured input data that can be used by subsequent modules; The multimodal biomedical data acquisition and preprocessing module is configured to acquire heterogeneous biomedical data from at least one of the following data sources: The Traditional Chinese Medicine (TCM) component database module acquires the chemical structure information of TCM monomer compounds and uses cheminformatics tools to convert them into fixed-length molecular fingerprints and other structure-based molecular descriptor vectors, while also acquiring their physicochemical properties and ADMET property data. The glioma target database module acquires identifiers, amino acid or nucleotide sequences, known three-dimensional structural data, gene expression profile data, and gene mutation data of proteins and genes related to glioma occurrence, development, and drug resistance. It also extracts feature vectors of protein sequences through a pre-trained protein sequence embedding model. The drug-target interaction database module acquires data on the binding affinity between known Chinese medicine components and protein targets, or the interaction relationships verified based on high-throughput screening experiments. The disease pathway database module acquires information on signaling pathways related to gliomas, including key node proteins or genes in the pathways and cross-correlation between pathways; and the clinical and patient data module acquires clinical diagnostic information, tumor subtypes, treatment responses, prognostic data, and patient-specific molecular omics data of glioma patients. Furthermore, the multimodal biomedical data acquisition and preprocessing module is further configured to remove missing and outlier values ​​through data cleaning algorithms, process high-dimensional data using dimensionality reduction techniques, ensure data quality and consistency, and ultimately integrate data from different sources and in different formats into a unified set of multidimensional feature vectors and an association matrix as structured input data.

[0020] The multi-level biological network construction module is used to construct multiple interconnected multi-level biological networks based on structured input data, which describe the complex relationships between Chinese medicine components, glioma targets and disease pathways, and generate an initial node feature vector for each node in the multi-level biological network. The multi-level biological network building module is configured to build at least one of the following multi-level biological networks: The Chinese medicine component similarity network uses Chinese medicine monomer compounds as nodes, and the edges between nodes represent the similarity of their chemical structure or pharmacological activity. The similarity is quantified by calculating the Tanimoto coefficient between molecular fingerprints or the distance metric based on the topological features of chemical structure, and weighted or binarized edges are constructed according to preset thresholds. The glioma target interaction network uses glioma-related proteins or genes as nodes. The edges between nodes represent protein-protein physical interactions, gene co-expression, shared functional pathways, or genetic interactions. The network is constructed by integrating PPI data from a protein interaction database and can adjust edge weights or extract subnetworks based on gene expression data of specific glioma subtypes. The traditional Chinese medicine component-target association network is a bipartite graph, in which one type of node represents traditional Chinese medicine components and the other type of node represents glioma targets. The edges between nodes represent the known or calculated interactions between traditional Chinese medicine components and targets. The strength of the interactions can be weighted based on binding affinity data or predicted probabilities. The disease pathway association network uses glioma-related signaling pathways as nodes, and the edges between nodes represent cross-associations, co-regulatory relationships, or shared key molecules between pathways; and the traditional Chinese medicine compound synergistic network uses multiple monomeric components in the compound as nodes, and the edges represent potential synergistic or antagonistic effects between components. The effects can be constructed based on literature mining or preliminary computational prediction results. The multi-level biological network construction module realizes the dynamic construction and management of the network through the graph computing library, and generates an initial node feature vector for each node in the multi-level biological network. This vector combines the structured features and network topology features from the multimodal biomedical data acquisition and preprocessing module.

[0021] The deep learning model construction and training module for synergistic effects is used to receive a multi-level biological network and its initial node feature vectors, and to build and train a deep learning model based on a graph neural network architecture to predict the deep synergistic effect network between Chinese medicine components and glioma targets, quantify its synergistic effect, and output the prediction results of the synergistic effect network. The deep learning model in the synergistic deep learning model construction and training module includes a multimodal graph encoder. The multimodal graph encoder is configured to receive initial feature vectors from traditional Chinese medicine component nodes, initial feature vectors from glioma target nodes, and adjacency matrices for different network types, and includes: Multiple stacked graph attention network layers, each GAT layer can adaptively learn the attention weights between a node and its neighboring nodes, thereby capturing the local and global contextual information of the node in different biological networks. The GAT layer performs the following operations on each node: after linear transformation of the node's feature vector, attention is calculated with the features of the neighboring nodes to generate attention coefficients. These coefficients are normalized by softmax and used to weight and aggregate the features of the neighboring nodes to generate an updated node embedding. The GAT layer adopts a multi-head attention mechanism and concatenates or averages the multiple independent attention aggregation results. A cross-modal feature fusion layer is set up on top of the node embeddings output by multiple stacked GAT layers. A fusion mechanism based on a Transformer encoder or a specially designed cross-attention module is set up. The mechanism can learn how to dynamically weigh node features from different biological networks and generate a multimodal fusion vector that reflects the potential synergistic relationship between the TCM component and glioma target pair. A TCM compound synergistic representation layer is also set up. For compound formulas involving multiple TCM components, the layer aggregates the fusion vectors of all individual components in the compound through a pooling layer to generate a composite feature representation that represents the interaction between the entire compound and the glioma target network. The aggregation process can further consider the relative concentration or dosage ratio of each component in the compound.

[0022] The deep learning model in the collaborative deep learning model construction and training module further includes a multi-task prediction head, which is configured to receive multimodal fusion vectors or complex composite feature representations and includes at least one of the following parallel prediction heads: The synergy strength prediction head is a fully connected neural network that outputs a continuous value representing the synergy strength score between traditional Chinese medicine and glioma target network. The score is based on regression prediction using existing synergy index or customized synergy efficacy index. The prediction head uses mean squared error or mean absolute error as the loss function. The key target identification head, a multi-label classifier, is used to predict which specific targets in the glioma pathological network will produce significant synergistic effects when traditional Chinese medicine acts on them. The prediction head is trained using a binary cross-entropy loss function. The mechanism of action pathway prediction head, a multi-label classifier, is used to predict which glioma-related signaling pathways are mainly affected by the synergistic effect. The prediction head is also trained using a binary cross-entropy loss function.

[0023] The synergistic deep learning model construction and training module employs at least one of the following training strategies: Self-supervised pre-training involves training the deep learning model on large-scale biological network data before formal training to learn general, biologically meaningful node embedding representations. Pre-training tasks include node attribute prediction, link prediction, or graph reconstruction. Multi-task learning optimizes the loss function of all prediction heads by weighted combination of multi-task prediction heads in an end-to-end manner. The model can learn the shared features and interdependencies between tasks, thereby improving the overall prediction performance and generalization ability. Transfer learning and fine-tuning: The model is pre-trained on a broader drug-disease interaction dataset and then fine-tuned on a glioma-specific traditional Chinese medicine-target synergistic effect dataset to adapt to the data distribution and prediction task of the specific domain. The training process uses the Adam optimizer, combined with a learning rate scheduling strategy, and is evaluated using a cross-validation strategy. Early stopping is also employed to prevent overfitting.

[0024] The Synergistic Network Analysis and Interpretability Module is used to perform in-depth analysis of the synergistic network prediction results, thereby revealing the intrinsic mechanism of synergistic effects between traditional Chinese medicine components and glioma targets, and providing interpretable and biologically meaningful insights, outputting an interpretable report. The Synergistic Network Analysis and Interpretability Module is set up to perform in-depth analysis of synergistic network prediction results using at least one of the following interpretation methods: The gradient-based interpretation method uses methods such as Integrated Gradients or Grad-CAM for GNNs to calculate the gradient of the synergistic effect prediction result relative to the input node features and network structure. By analyzing these gradient values, the traditional Chinese medicine components, glioma targets and their specific interaction edges that contribute the most to the synergistic effect prediction are identified. Attention weight analysis module directly extracts and analyzes the attention weights learned by the GAT layer and cross-modal feature fusion layer in the synergistic deep learning model construction and training module to reveal key synergistic pathways or patterns. Subgraph extraction and visualization: Based on the analysis results based on gradient or attention weights, the module extracts the smallest subnetworks that are highly correlated with the predicted synergistic effects from the vast biological network. The subnetworks contain the core Chinese medicine components, glioma targets and the interaction edges between them that make significant contributions, and are graphically displayed through network visualization tools. Feature importance assessment module uses SHAP or other perturbation-based methods to evaluate the contribution of each input feature to the synergistic effect prediction results, and the evaluation results are presented in the form of a ranking list or force graph; and pathway enrichment analysis module inputs the set of glioma targets identified by the key target identification head into the pathway enrichment analysis tool to determine the biological processes, signaling pathways or disease ontology in which these targets are significantly enriched in the context of high synergistic effect, and the analysis results provide a systematic explanation of the molecular mechanism of synergistic anti-glioma effect of traditional Chinese medicine; The outputs of the Synergistic Network Analysis and Interpretability Module include, but are not limited to: a list of key synergistic TCM components and their contribution, a list of key glioma targets and their weights, a key component-target interaction map, a synergistic subnetwork topology diagram, a report on enriched signaling pathways, and a report on interpretive evidence.

[0025] The Personalization and Clinical Translation module combines the analysis results of the Synergistic Network Analysis and Interpretability module with the specific molecular characteristic data of individual glioma patients to achieve personalized treatment strategy recommendations and clinical decision support, and generate personalized clinical decision support reports. The Personalization and Clinical Translation module is set up for: Patient molecular feature integration: The module receives and integrates molecular omics data specific to a single glioma patient. After preprocessing, the data is used as additional input features to fine-tune a pre-trained synergistic deep learning model, or it can be incorporated into the model inference process through a specific embedding layer to generate personalized prediction results for the patient. Glioma subtype-specific prediction: Based on the patient's glioma subtype diagnosis, the module calls a synergistic effect prediction sub-model or parameter set optimized for a specific subtype to improve the accuracy and specificity of the prediction, and identify the most effective combination of synergistic Chinese medicine components and targets for that subtype. The synergistic risk-benefit assessment module integrates toxicity data and potential off-target effect predictions of traditional Chinese medicine components. When recommending synergistic traditional Chinese medicine regimens, the module not only considers their synergistic efficacy against glioma, but also assesses their potential toxic side effects and drug interaction risks. The risk-benefit assessment is achieved through a multi-criteria decision analysis model. Its inputs include the predicted synergistic strength, the specificity of key targets, the known toxic dose of the components, and potential inhibitory or inducing effects of drug-metabolizing enzymes. The output is a comprehensive risk-benefit score. The clinical decision support report generation module generates detailed clinical decision support reports based on personalized prediction results and risk-benefit assessments. The report content includes: recommended AI-driven synergistic combinations of traditional Chinese medicine ingredients, predicted synergistic effect strength, key targets and signaling pathways, potential mechanism of action diagrams, personalized interpretation of patient molecular characteristics, early warning of potential drug interactions, and risk-benefit trade-offs for treatment plans. The report is output in a standardized and structured format.

[0026] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0027] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI-driven network prediction system for synergistic effects of traditional Chinese medicine and glioma targets, characterized by: include: The multimodal biomedical data acquisition and preprocessing module is designed to acquire heterogeneous biomedical data related to Chinese medicine components and their bioactivity, glioma-related targets and their molecular characteristics, and Chinese medicine-target interactions from multiple data sources. A multi-level biological network construction module is used to construct multiple interconnected multi-level biological networks based on structured input data, which describe the complex relationships between traditional Chinese medicine components, glioma targets, and disease pathways. The synergistic deep learning model construction and training module is used to receive multi-layer biological networks and their initial node feature vectors, and to build and train a deep learning model based on a graph neural network architecture. The Synergistic Network Analysis and Interpretability Module is used for in-depth analysis of the prediction results of synergistic networks. The Personalization and Clinical Translation module is used to combine the analysis results of the Synergistic Network Analysis and Interpretability module with individual glioma patient-specific molecular characterization data.

2. The AI-driven network prediction system for synergistic effects of traditional Chinese medicine and glioma targets according to claim 1, characterized in that: The multimodal biomedical data acquisition and preprocessing module is configured to acquire heterogeneous biomedical data from at least one of the following data sources: traditional Chinese medicine component database, glioma target database, drug-target interaction database, and disease pathway database; The multimodal biomedical data acquisition and preprocessing module is further configured to remove missing and outlier values ​​through data cleaning algorithms, process high-dimensional data using dimensionality reduction techniques, ensure data quality and consistency, and ultimately integrate the data from different sources and in different formats into a unified set of multidimensional feature vectors and an association matrix as the structured input data.

3. The AI-driven network prediction system for synergistic effects of traditional Chinese medicine and glioma targets according to claim 1, characterized in that: The multi-level biological network construction module is configured to construct at least one of the following multi-level biological networks: a Chinese medicine component similarity network, a glioma target interaction network, a Chinese medicine component-target association network, a disease pathway association network, and a Chinese medicine compound synergistic network. The multi-level biological network construction module realizes the dynamic construction and management of the network through the graph computing library, and generates an initial node feature vector for each node in the multi-level biological network. This vector combines the structured features and network topology features from the multimodal biomedical data acquisition and preprocessing module.

4. The AI-driven network prediction system for synergistic effects of traditional Chinese medicine and glioma targets according to claim 1, characterized in that: The deep learning model in the synergistic deep learning model construction and training module includes a multimodal graph encoder. This multimodal graph encoder is configured to receive initial feature vectors of traditional Chinese medicine component nodes, initial feature vectors of glioma target nodes, and adjacency matrices for different network types, and includes: Multiple stacked graph attention network layers, each GAT layer can adaptively learn the attention weights between a node and its neighboring nodes, thereby capturing the local and global contextual information of nodes in different biological networks; A cross-modal feature fusion layer is set up on the node embeddings of the outputs of the multiple stacked GAT layers, with a fusion mechanism based on a Transformer encoder or a specially designed cross-attention module. A collaborative representation layer for traditional Chinese medicine compound formulas: For compound formulas involving multiple traditional Chinese medicine components, the layer aggregates the fusion vectors of all monomeric components in the compound formula through a pooling layer to generate a composite feature representation representing the interaction between the entire compound formula and the glioma target network. The aggregation process can further consider the relative concentration or dosage ratio of each component in the compound formula.

5. The AI-driven network prediction system for synergistic effects of traditional Chinese medicine and glioma targets according to claim 4, characterized in that: The deep learning model in the synergistic deep learning model construction and training module further includes a multi-task prediction head, which is configured to receive the multimodal fusion vector or complex composite feature representation and includes at least one of the following parallel prediction heads: synergistic effect strength prediction head, key target identification head, and mechanism of action pathway prediction head.

6. The AI-driven network prediction system for synergistic effects of traditional Chinese medicine and glioma targets according to claim 1, characterized in that: The synergistic deep learning model construction and training module adopts at least one of the following training strategies: self-supervised pre-training, multi-task learning and transfer learning and fine-tuning; The training process employs the Adam optimizer, combined with a learning rate scheduling strategy, and uses a cross-validation strategy for evaluation. Early stopping is also employed to prevent overfitting.

7. The AI-driven network prediction system for synergistic effects of traditional Chinese medicine and glioma targets according to claim 1, characterized in that: The synergistic network parsing and interpretability module is configured to perform in-depth parsing of the synergistic network prediction results using at least one of the following interpretation methods: gradient-based interpretation method, attention weight analysis, subgraph extraction and visualization, feature importance assessment and pathway enrichment analysis; The output of the synergistic network analysis and interpretability module includes, but is not limited to: a list of key synergistic Chinese medicine components and their contribution, a list of key glioma targets and their weights, a key component-target interaction map, a synergistic subnetwork topology diagram, a report on enriched signaling pathways, and a report on interpretive evidence.

8. The AI-driven network prediction system for synergistic effects of traditional Chinese medicine and glioma targets according to claim 1, characterized in that: The personalization and clinical translation module is configured for: patient molecular feature integration, glioma subtype-specific prediction, synergistic risk-benefit assessment, and generation of clinical decision support reports.

Images