An intelligent analysis system and method of traditional Chinese medicine based on knowledge graph constraint reasoning and bidirectional network pharmacology

Abstract

The application discloses a traditional Chinese medicine intelligent analysis system and method based on knowledge graph constraint reasoning and bidirectional network pharmacology, first, multi-source traditional Chinese medicine databases are integrated, a seven-layer associated knowledge graph covering prescriptions, medicinal materials, compounds, protein targets, syndromes and diseases is constructed, and entity aliases, standard IDs and evidence traceability annotations are unified; secondly, a reliable AI syndrome differentiation engine is designed, which includes enhanced entity recognition and candidate evidence recall, knowledge graph constraint large model reasoning and database reverse verification, solves the general large model traditional Chinese medicine question and answer illusion and the conclusion without traceability defect; meanwhile, a bidirectional network pharmacology calculation core is constructed, forwardly, a whole-link analysis of prescription and drug molecular action mechanism is realized, and reversely, disease / target driven candidate prescription and drug mining is supported; a probability joint scoring quantization multi-component synergistic effect is innovatively proposed, and medicinal material screening is modeled as a 0 / 1 combination optimization mathematical problem with multiple constraints, and a scene intelligent optimization algorithm is adopted to output the optimal compatibility.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of information technology in traditional Chinese medicine and artificial intelligence. More specifically, this invention relates to an intelligent analysis system and method for traditional Chinese medicine based on knowledge graph-constrained reasoning and bidirectional network pharmacology. Background Technology

[0002] Traditional Chinese medicine (TCM) knowledge has accumulated over thousands of years of clinical experience, with complex relationships between its prescriptions, medicinal materials, components, targets, diseases, and syndromes. As TCM modernization research deepens, higher demands are placed on the structured organization, computational analysis, and evidence verification of this knowledge. However, the traditional TCM knowledge system is characterized by multi-source heterogeneity, semantic complexity, and multiple layers of relationships, making its modernization highly dependent on systematic data governance and computational methods.

[0003] The existing technology has the following core defects: 1. Databases are fragmented and lack complete knowledge links; existing mainstream TCM databases such as TCMID, TCMSP, SymMap, and BATMAN-TCM mainly rely on static retrieval and one-way network analysis, making it difficult to connect clinical symptoms, syndrome identification, prescription compatibility, and molecular mechanisms; researchers need to switch between multiple web pages, R language packages, and Python scripts, the data formats are not consistent, and the research process has extremely low reproducibility. 2. General-purpose large-scale models for answering questions about traditional Chinese medicine (TCM) have factual biases and their conclusions lack source tracing. GPT, Claude, and other domestic general-purpose large-scale models rely solely on statistical analysis of training corpora to generate TCM answers, without the constraints of a professional knowledge base. Recommended prescriptions and targets are prone to factual errors. The output results cannot be traced back to literature or database sources, and there is no distinction between levels of evidence, which fails to meet the rigor requirements of TCM research and medical institutions. 3. Network pharmacology only supports forward analysis, and traditional network analysis methods or tools lack the ability to mine reverse candidates; it mainly focuses on a one-way "medicinal material-component-target-pathway" approach to explain the mechanism of action of known prescriptions; there is no systematic calculation scheme to address the urgent need in new drug development to find effective Chinese medicines and prescriptions from the disease / target, and it cannot automatically backtrack candidate medicinal materials and prescriptions and quantify and rank them; 4. The synergistic effect of multiple components in traditional Chinese medicine relies solely on empirical judgment and lacks standardized quantitative models. The synergistic effect of multiple components in traditional Chinese medicine compound prescriptions is a core pharmacodynamic feature. Existing analytical tools only perform simple network topology statistics and lack mathematical models to quantify the comprehensive effect of different combinations of medicinal materials on disease targets. There are no automated compatibility optimization algorithms, and prescription selection relies entirely on researchers' experience. There is a lack of quantitative evaluation methods for the synergistic effect of multiple components, making it difficult to achieve the transformation from empirical prescription to data-driven prescription. 5. Existing software patents related to traditional Chinese medicine are mainly concentrated on single functional modules, such as traditional Chinese medicine database retrieval systems, network pharmacology visualization tools, or component prediction models based on machine learning; there is a lack of traditional Chinese medicine analysis systems that integrate knowledge graph-constrained reasoning, bidirectional network pharmacology calculations, and multi-component synergistic quantitative analysis. In summary, existing TCM research tools suffer from the following main deficiencies: (1) a lack of a complete knowledge chain connecting "symptoms-syndromes-prescriptions-medicinal materials-components-targets-pathways"; (2) a lack of systematic computational capabilities to discover candidate prescriptions from the perspective of diseases / targets; (3) a lack of quantitative evaluation methods for the synergistic effects of multiple components; (4) a lack of a reliable AI reasoning mechanism based on knowledge graph constraints, making it difficult to guarantee the accuracy and traceability of output results; and (5) a lack of engineering delivery capabilities for scientific research scenarios, resulting in high costs and poor reproducibility when switching between tools. Therefore, it is urgent to establish a TCM intelligent analysis system and method based on knowledge graph constraint reasoning and bidirectional network pharmacology that combines professional knowledge organization, reliable reasoning, and full-chain computational capabilities. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a traditional Chinese medicine intelligent analysis system and method based on knowledge graph-constrained reasoning and bidirectional network pharmacology. It employs a technical system combining retrieval-enhanced generation, knowledge graph constraints, and database reverse verification to confine large-scale model reasoning within a credible knowledge framework, significantly reducing the risk of model illusion. It supports bidirectional network pharmacology computation and multi-component synergistic quantitative analysis, balancing the dual needs of mechanism research and candidate discovery. The invention achieves the following technical objectives: 1. Construct a seven-layer standardized knowledge graph that connects "symptoms-syndromes-prescriptions-medicinal materials-compounds-targets-pathways-diseases", unify multi-source heterogeneous TCM data, and build a full-link computable data foundation; 2. Design an AI dialectical reasoning mechanism that combines enhanced retrieval generation with knowledge graph constraints to limit the output of the large model to a set of credible candidates, and ensure the accuracy and traceability of the output results through reverse verification of the database, eliminating the illusion of TCM question answering in the large model, and making all output results traceable and verifiable; 3. Construct a bidirectional network pharmacology computation engine that supports both positive "formula-medicinal material-component-target-pathway" mechanism interpretation and reverse "disease / target-component-medicinal material-formula" candidate discovery, taking into account both classical formula research and natural drug candidate screening. 4. By introducing probabilistic joint scoring and 0 / 1 combination optimization methods, the synergistic effects of multiple components of traditional Chinese medicine are quantitatively evaluated, realizing the transformation of prescription analysis from experience-based judgment to data-driven analysis; 5. Provides engineering delivery capabilities for scientific research scenarios, supporting web access, API calls and Docker private deployment, reducing the cost of switching between tools and ensuring the reproducibility of research processes.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A traditional Chinese medicine intelligent analysis system based on knowledge graph-constrained reasoning and bidirectional network pharmacology adopts a five-layer architecture, consisting of an application service layer, a computational analysis layer, a reasoning engine layer, a knowledge graph layer, and a multi-source data layer from top to bottom; wherein: The multi-source data layer is used to integrate publicly available TCM databases such as TCMID, TCMSP, SymMap, and BATMAN-TCM to complete entity normalization, conflict verification, evidence annotation, and output standardized raw datasets. The knowledge graph layer is used to construct a multi-layered association knowledge graph that connects symptoms, syndromes, prescriptions, medicinal materials, compounds, protein targets, pathways, and diseases. The graph relationships include prescription-medicinal materials, medicinal materials-compounds, compounds-targets, targets-pathways, syndromes-prescriptions, symptoms-syndromes, and targets-diseases. Each relationship is accompanied by source database, evidence level, and original literature tracing fields. The inference engine layer is used for trusted AI identification. It sequentially performs multimodal input entity recognition and synonym normalization, knowledge graph candidate evidence recall, graph-constrained large model structured reasoning, and underlying database reverse verification, and outputs the identification prescription recommendation results with confidence and traceability information. The computational analysis layer integrates a forward network pharmacology computation module, a reverse network pharmacology computation module, a multi-component synergistic quantification optimization module, and an interactive visualization module; among which, the multi-component synergistic quantification optimization module includes a probability joint scoring submodule and a 0 / 1 herbal combination optimization submodule; The application service layer provides three delivery forms: a web-based visual interactive interface, a RESTful API call interface, and a Docker offline private deployment image, and exposes all search, calculation, optimization, and export functions to the outside world; The system executes the following integrated analysis process: multimodal input → entity recognition and synonym normalization → knowledge graph candidate recall → knowledge graph constrained large model reasoning → database reverse verification → bidirectional network pharmacology calculation → multi-component synergistic quantification and 0 / 1 combination optimization → enrichment visualization and standardized report export.

[0006] In a preferred embodiment, the intelligent analysis system for traditional Chinese medicine includes a data governance and knowledge graph module, a retrieval-enhanced reasoning module, a bidirectional network pharmacology calculation module, a multi-component synergistic quantitative optimization module, and an enrichment analysis and visualization module; wherein... The data governance and knowledge graph module is used to integrate multi-source data and build structured knowledge graphs; The retrieval-enhanced reasoning module is used for entity recognition, candidate recall, knowledge graph-constrained reasoning, and database reverse verification. A bidirectional network pharmacology computation module is used for forward mechanism elucidation, reverse candidate discovery, and multi-component synergistic quantification. The multi-component synergistic quantification optimization module is used to quantify the efficacy probability of multiple components acting together on the target; it constructs a multi-objective weighted objective function and automatically outputs the optimal herbal combination scheme. The enrichment analysis and visualization module is used for pathway enrichment, network visualization, and report export.

[0007] A method for intelligent analysis of traditional Chinese medicine based on knowledge graph-constrained reasoning and bidirectional network pharmacology, executed using the aforementioned intelligent analysis system for traditional Chinese medicine based on knowledge graph-constrained reasoning and bidirectional network pharmacology, includes the following steps: S1. Multi-source data governance and multi-layer knowledge graph construction; S2, Enhanced entity recognition and candidate evidence recall; S3, structured dialectical reasoning with knowledge graph constraints on large-scale models; S4. Reverse verification of the underlying database; S5, Forward Network Pharmacological Computation; S6, Reverse Network Pharmacology Calculation; S7, Multi-component synergistic quantization and 0 / 1 combinatorial optimization; S8, offline enrichment analysis and interactive visualization.

[0008] In a preferred embodiment, multi-source data governance and multi-layered knowledge graph construction specifically include the following steps: (1) The data layer integrates publicly available TCM database resources and establishes a unified entity dictionary, relational schema and evidence field. The entity dictionary includes six core entities: prescriptions, medicinal materials, compounds, protein targets, diseases and syndromes. For each entity, a mapping relationship is established between Chinese and English names, pinyin, aliases and standard IDs. The standard IDs include PubChem ID, Gene Symbol, OMIM or ICD code. (2) By deduplication, synonym merging, conflict labeling, missing data checking, and source recording, the data version, original source, and processing process are preserved. For entities with the same name but different characteristics, they are distinguished and labeled according to the source database and the level of evidence; for relationships with inconsistent levels of evidence, the highest level is given priority, and conflict information is recorded for user reference. (3) Construct a multi-layer knowledge graph with "prescription-medicine-chemical-target" as the core, connecting disease, syndrome and pathway information. The graph association includes prescription-medicinal material, medicinal material-compound, compound-target, target-pathway, syndrome-prescription, symptom-syndrome and target-disease. Each relationship is accompanied by source database, evidence level and original literature tracing field. The association is divided into three levels of evidence according to experimental verification, literature report and database prediction.

[0009] In a preferred embodiment, the enhanced entity recognition and candidate evidence recall process specifically includes the following steps: (1) Receive any one or more inputs such as natural language symptoms, prescriptions, names of medicinal materials, compounds, target IDs, gene names, and disease names. First, perform word segmentation and entity recognition on the input text, and use a pre-built entity dictionary for precise matching and fuzzy matching. For symptom descriptions, use a large language model for semantic understanding and entity extraction to identify symptom entities, syndrome candidates, and implicit prescription intentions. (2) The identified entities are normalized by synonyms, and aliases, common names and historical names are uniformly mapped to standard names; for entities that cannot be matched precisely, Levenshtein edit distance and Jaccard similarity are used to expand the candidates and generate a candidate list of similar entities. (3) Based on entity type and relational paths in the knowledge graph, retrieve candidate prescriptions, medicinal materials, ingredients, targets, and their associated evidence from the knowledge graph; the retrieval process follows these rules: For symptom / syndrome input, candidate prescriptions are recalled along the "symptom-syndrome-prescription" path; For prescription input, an association network is developed along the path of "prescription-medicinal material-compound-target-pathway"; For disease / target inputs, candidate prescriptions are traced back along the path of "target-compound-medicinal material-prescription", and the data source, matching method and confidence score of each related prescription are recorded simultaneously.

[0010] In a preferred embodiment, the structured dialectical reasoning of a large model constrained by a knowledge graph specifically includes the following steps: (1) Using the recall candidate entity set, associated evidence and knowledge graph subgraph as context, construct structured prompt words. The prompt words clearly limit the model to reasoning and output only within the candidate set, and prohibit the model from generating conclusions beyond the candidate range based on prior knowledge in the training corpus. (2) Call a large model compatible with the OpenAI interface to output structured JSON format results; the JSON format results include a list of recommended prescriptions, reasons for recommendation, sources of evidence, confidence scores, and associated pathway information. The model output strictly follows JSON Schema constraints to ensure field integrity and type consistency; (3) Post-process the model output. The processing steps include JSON format verification, field integrity check, and entity name matching verification with the knowledge graph. For each recommended prescription output by the model, extract its corresponding medicinal material composition, active ingredients and target points, and prepare for reverse verification in step S4.

[0011] In a preferred embodiment, the underlying database performs reverse verification, specifically including the following steps: (1) For the recommended prescriptions, medicinal materials, components and targets output by the model in step 3, perform reverse query verification against the underlying database one by one; the verification content includes: whether the prescription exists in the database, whether the composition of the medicinal materials is consistent with the database record, whether the target prediction of the components is supported by literature, and whether the pathway enrichment results are consistent with the known mechanisms. (2) For items that fail verification, mark them as "pending confirmation" and prompt the user; for items that pass verification, enhance their confidence score and attach database source links and original literature citations; (3) Generate a verification report, which includes the recommended conclusions, data basis, evidence level, source traceability and uncertainty explanation; all results retain the source, version, parameters and operation log to ensure the reproducibility of the results.

[0012] In a preferred embodiment, the forward network pharmacology calculation specifically includes the following steps: (1) For the prescription or medicinal materials input by the user, construct a positive correlation network along the path of "prescription or medicinal material-component-target-pathway". First, obtain all the medicinal materials contained in the prescription, then obtain all the compounds corresponding to each medicinal material, then obtain the predicted or verified target corresponding to each compound, and finally obtain the biological pathways in which the target participates. (2) Nodes and edges in the network are scored; the weight of compound-target edge is calculated based on the number of literature reports, database prediction score and experimental verification level; the weight of target-pathway edge is calculated based on the enrichment significance p-value and pathway participation in the KEGG and GO databases. (3) Perform topological analysis on the network, calculate the degree centrality, betweenness centrality and PageRank value of each node, and identify key nodes, hub genes and key pathways in the network; at the same time, perform KEGG / GO / DO enrichment analysis through hypergeometric distribution test or Fisher's exact test, and output a list of significantly enriched pathways and their statistical parameters.

[0013] In a preferred embodiment, the reverse network pharmacology calculation specifically includes the following steps: (1) For the disease name, target ID or gene name input by the user, construct a reverse association network along the path of "disease / target-component-medicinal material-prescription"; first obtain the relevant target from the disease, or directly from the target, then obtain the compound that acts on the corresponding target, then obtain the medicinal material containing the corresponding compound, and finally obtain the prescription containing the corresponding medicinal material. (2) The candidate prescriptions in the reverse network are scored in multiple dimensions, including target coverage, pathway enrichment significance, network modularity and evidence level; a comprehensive score is generated by weighted summation, and the TOP-N candidate prescriptions are output after sorting; the target coverage is the proportion of the active ingredients contained in the candidate prescriptions that cover the input target. (3) Analyze the formulation rules of candidate prescriptions, extract high-frequency co-occurring medicinal material pairs and drug group patterns, compare and verify them with classical prescription theory, and provide an explanation for the rationality of candidate prescriptions.

[0014] In a preferred embodiment, multi-component synergistic quantization and 0 / 1 combination optimization specifically include the following steps: (1) Probability joint scoring; For multiple active ingredients in the candidate prescription, calculate the comprehensive probability of their effect on key targets. Suppose that prescription F contains n active ingredients C1, C2, ..., Cn, and the set of key targets is T = {t1, t2, ..., tm}. Then the probability of active ingredient Ci acting on target tj is p(Ci->tj), which is obtained by normalizing the predicted scores based on literature reports and databases. The combined effect probability of multiple components on target tj was calculated using a probabilistic joint scoring model: where i = 1, 2, ..., n; This formula is based on the principle of the complement of independent event probabilities. It assumes that the effects of each component on the target are independent of each other. Then the probability that at least one component acts on the target is 1 minus the probability that none of the components act on the target. (2) 0 / 1 combinatorial optimization; the selection of candidate medicinal materials is transformed into a 0 / 1 combinatorial optimization problem: given a set of candidate medicinal materials H={h1,h2,...,hk}, each medicinal material hi contains a set of active ingredients, select a subset of medicinal materials S that is included in H, so as to maximize the objective function; the objective function is defined as: f(S)=a×Coverage(S,T)+b×Synergy(S)+c×Evidence(S)-d×Cost(S); Where Coverage(S,T) represents the coverage of subset S to the set of key targets T, Synergy(S) represents the synergistic strength of components within subset S (based on probabilistic joint scoring), Evidence(S) represents the sum of evidence levels for the association between components and targets within subset S, and Cost(S) represents the cost of medicinal materials in subset S. a, b, c, and d are weighting coefficients, which can be adjusted by the user according to research needs; (3) The above 0 / 1 combinatorial optimization problem is solved by using an optimization algorithm based on maximum likelihood estimation. For a small candidate set of k<=20, exhaustive search is used to ensure global optimum. For a large candidate set of k>20, genetic algorithm or simulated annealing algorithm is used for approximate solution. The number of iterations and convergence threshold are set to ensure the quality of the solution. The optimized combination of medicinal materials, corresponding active ingredients, target sites, expected pathways and quantitative evaluation indicators are output.

[0015] In a preferred embodiment, offline enrichment analysis and interactive visualization specifically include the following steps: (1) It integrates an offline KEGG / GO / DO enrichment analysis engine, which does not rely on external network services, ensuring the stability and reproducibility of the computing environment; it supports two enrichment methods, ORA (Over-Representation Analysis) and GSEA (Gene Set Enrichment Analysis), and outputs pathway enrichment results, statistical parameters and visualization charts. The statistical parameters include p-value, FDR and NES. (2) Generate interactive network visualization diagrams, including multi-layer network diagrams of prescription-medicinal materials-compound-target-pathway, enrichment analysis bubble diagrams, target interaction network diagrams, and component-target heatmaps; users can zoom, drag, filter, and export through the Web interface; (3) Supports exporting analysis results in multiple formats, including PDF reports, CSV data tables, PNG images and JSON parameter files. All exported files contain analysis parameters, data sources and version information to ensure the reproducibility of the research process.

[0016] The technical effects and advantages of the intelligent analysis system and method for traditional Chinese medicine based on knowledge graph constrained reasoning and bidirectional network pharmacology of this invention are as follows: 1. Knowledge Graph Enhanced Trustworthy AI Diagnosis: The system establishes a "retrieval enhancement - graph constraint - database reverse verification" mechanism to recall candidate prescriptions from trustworthy data, and then a large model completes the dialectical reasoning. Compared with traditional general-purpose large models, this invention limits the model output to the scope of candidate knowledge, ensures factual accuracy through database reverse verification, and makes the recommendation basis searchable and traceable, significantly reducing the risk of model illusion. Compared with static database retrieval, this invention introduces the semantic understanding and reasoning capabilities of a large model, which can process natural language symptom descriptions and generate structured dialectical interpretations; 2. Bidirectional Network Pharmacological Computation: The system simultaneously supports forward "formula-medicinal material-component-target-pathway" mechanism elucidation and reverse "disease / target-component-medicinal material-formula" candidate discovery. Compared with existing unidirectional network analysis tools, the bidirectional computing capability of this invention takes into account both classical formula research and natural drug candidate screening, forming a complete link from clinical symptoms to molecular mechanisms, filling the gap in the field of reverse discovery in existing tools; 3. Multi-component synergistic quantification and intelligent combination optimization: The system characterizes the comprehensive effect of multiple components on key targets through probabilistic joint scoring, transforming candidate medicinal material screening into a 0 / 1 combination optimization problem and automatically generating interpretable priority solutions. Compared with traditional empirical formulation methods, this invention realizes the transformation of formulation analysis from experience-based judgment to data-driven approach, providing a quantitative analysis tool for the discovery of new Chinese medicine drugs and the optimization of existing prescriptions. 4. Low-threshold engineering delivery: The system supports input methods including Chinese and English names, pinyin, gene names, and multiple IDs. It integrates offline KEGG / GO / DO enrichment, interactive network visualization, Web / API calls, and Docker private deployment. Compared with R packages and Python scripts that require programming skills, this invention significantly lowers the barrier to entry for network pharmacology, balancing ease of use, reproducibility, and data security. 5. Integrated Innovation Across the Entire Chain: The system connects the entire chain of "symptoms-syndrome-prescription-medicinal materials-ingredients-targets-pathways," integrating data querying, AI inference, network computing, enrichment analysis, combinatorial optimization, and visualization on a unified platform. Compared to traditional workflows that require switching between multiple tools, this invention significantly reduces the costs of cross-database retrieval, format conversion, and multi-tool switching, improving research efficiency and reproducibility. Attached Figure Description

[0017] Figure 1 This is a system architecture diagram of a traditional Chinese medicine intelligent analysis system based on knowledge graph constrained reasoning and bidirectional network pharmacology according to the present invention. Figure 2 This is a schematic diagram of the knowledge graph structure of a traditional Chinese medicine intelligent analysis system based on knowledge graph constrained reasoning and bidirectional network pharmacology according to the present invention. Figure 3This is a flowchart of a method for intelligent analysis of traditional Chinese medicine based on knowledge graph-constrained reasoning and bidirectional network pharmacology according to the present invention. Figure 4 This is a flowchart of the 0 / 1 combination optimization process in the intelligent analysis method of traditional Chinese medicine based on knowledge graph constraint reasoning and bidirectional network pharmacology 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] like Figure 1-2 As shown, this invention provides a traditional Chinese medicine intelligent analysis system based on knowledge graph-constrained reasoning and bidirectional network pharmacology. It adopts a five-layer architecture, from top to bottom: application service layer, computational analysis layer, reasoning engine layer, knowledge graph layer, and multi-source data layer; wherein: (1) Multi-source data layer It integrates publicly available and authoritative TCM database resources, including but not limited to TCMID, TCMSP, SymMap, and BATMAN-TCM, as well as classic prescription literature and public target pathway databases such as KEGG, GO, DO, and GeneBank. Data governance was performed on the integrated resources to output a standardized raw dataset. The governance process included entity extraction, differentiation of homonyms and heteronyms, synonym unification, conflict association labeling, evidence level classification, standard ID mapping, and structured data storage. A total of 578,000 standardized records were generated, including 6,200+ prescriptions, 4,700+ medicinal materials, 13,100 compounds, and 14,200 protein targets. (2) Knowledge Graph Layer A seven-layer knowledge graph is constructed, with core entities including prescriptions, medicinal materials, compounds, protein targets, syndromes, diseases, and pathways. Pathways are the functional dimensions associated with targets and do not belong to the six types of entities in the core entity dictionary. The seven-layer graph = six types of entities + pathway association dimension. Relationship types: multiple medicinal materials contained in the prescription, multiple compounds contained in the medicinal materials, compounds acting on one or more targets, targets participating in one or more pathways, syndromes corresponding to one or more prescriptions, which syndrome category the symptoms belong to, targets associated with one or more diseases; Each relationship is further supported by a ternary attribute: source database, level of evidence (1. experimental verification; 2. literature report; 3. database prediction), and original literature tracing field. Built-in synonym unification mechanism: such as unifying Sichuan Army as Da Huang, unifying COX-2 as PTGS2, etc., and supporting multiple forms of retrieval including Chinese and English, Pinyin, and gene ID; (3) Inference Engine Layer This layer is the core of trusted AI dialectics, which sequentially performs multimodal input entity recognition and synonym normalization, knowledge graph candidate evidence recall, graph-constrained large model structured reasoning, and underlying database reverse verification, outputting dialectical prescription recommendation results with confidence and traceability information; 1) Multimodal input support: natural language symptoms, prescription names, medicinal materials, compounds, target genes, and disease names; extracting implicit intentions of symptoms and syndromes through large-scale semantic modeling; 2) Candidate Recall: Match the graph path according to the input entity type to limit the range of candidate entities and avoid the generation of large models without boundaries; 3) Graph-constrained reasoning: Construct a structured Prompt to force large models to output only within the recalled candidate set, and output unified JSON structured results; compatible with mainstream large models such as GPT-4, Claude, and Tongyi 1000 Questions; 4) Reverse verification: The model-recommended prescriptions or medicinal materials, targets, and pathways are back-checked against the underlying map to verify consistency, mark anomalies, and add all traceability evidence to improve the credibility of the results. (4) Calculation and analysis layer It integrates a bidirectional network pharmacology calculation module, a multi-component synergistic quantification optimization module, and an interactive visualization module; the bidirectional network pharmacology calculation module further includes a forward network pharmacology calculation module and a reverse network pharmacology calculation module; among them, 1) Forward Network Pharmacology Module Input formula / medicinal material, construct a "formula-medicinal material-compound-target-pathway" network layer by layer; calculate node topology indicators (degree centrality, betweenness centrality, PageRank) to identify core pharmacodynamic substances and targets; perform ORA and GSEA enrichment analysis offline, and output significant KEGG / GO / DO pathways; 2) Reverse Network Pharmacology Module Input a set of diseases / targets, and trace back to targets → compounds → medicinal materials → prescriptions; multi-dimensional comprehensive scoring, including: target coverage, pathway enrichment significance, network modularity, and evidence level; sort and output TOP-N candidate prescriptions; automatically discover high-frequency co-occurring drug pairs and analyze compatibility patterns; 3) Multi-component collaborative quantization optimization module It includes two sub-modules: probabilistic joint scoring and 0 / 1 combinatorial optimization. ① Probability Joint Scoring Submodule: Used to quantify the overall efficacy probability of multiple components acting on the target through the independent event complement formula; ②0 / 1 Combinatorial Optimization Submodule: Used to construct a multi-objective weighted objective function, with two optimization algorithms: exhaustive search and genetic / simulated annealing, and automatically outputs the optimal medicinal material combination scheme; 4) Enrichment and Visualization Module The offline enrichment engine does not rely on the external network; the analysis results generate multi-layer network diagrams, target interaction diagrams, enrichment bubble charts, and component target heatmaps; the view supports zooming, filtering, and drag-and-drop interaction; it also supports exporting analysis results in multiple formats such as PDF reports, CSV data tables, PNG images, and complete JSON parameter files, and retains all calculation parameters to ensure reproducibility; (5) Application Service Layer We offer three delivery options to fully cover the needs of research institutions, businesses, and medical organizations: 1) Web-based visual interactive interface: including the Streamlit front-end, 12 independent function panels, accessible directly without installing a browser; 2) RESTful API: Provides 23 standardized interfaces, OpenAPI automatically generates documentation, and supports batch automated analysis using Python / R / Java; 3) Docker private image: Encapsulates complete database, engine, and visualization services, and can be deployed offline on local / private cloud to protect the privacy of proprietary clinical and experimental data; All search, calculation, optimization, and export functions are made available to the public.

[0020] The intelligent analysis system for traditional Chinese medicine executes the following integrated analysis process: multimodal input → entity recognition and synonym normalization → knowledge graph candidate recall → knowledge graph constrained large model reasoning → database reverse verification → bidirectional network pharmacology calculation → multi-component synergistic quantification and 0 / 1 combination optimization → enrichment visualization and standardized report export.

[0021] A traditional Chinese medicine intelligent analysis system based on knowledge graph-constrained reasoning and bidirectional network pharmacology includes a data governance and knowledge graph module, a retrieval-enhanced reasoning module, a bidirectional network pharmacology calculation module, a multi-component synergistic quantitative optimization module, and an enrichment analysis and visualization module; among which, The data governance and knowledge graph module is used to integrate multi-source data and build structured knowledge graphs; The retrieval-enhanced reasoning module is used for entity recognition, candidate recall, knowledge graph-constrained reasoning, and database reverse verification. A bidirectional network pharmacology computation module is used for forward mechanism elucidation, reverse candidate discovery, and multi-component synergistic quantification. The multi-component synergistic quantification optimization module is used to quantify the efficacy probability of multiple components acting together on the target; it constructs a multi-objective weighted objective function and automatically outputs the optimal herbal combination scheme. The enrichment analysis and visualization module is used for pathway enrichment, network visualization, and report export.

[0022] like Figure 3-4 As shown, this invention also provides a method for intelligent analysis of traditional Chinese medicine based on knowledge graph-constrained reasoning and bidirectional network pharmacology, which is executed using the aforementioned intelligent analysis system for traditional Chinese medicine based on knowledge graph-constrained reasoning and bidirectional network pharmacology, and includes the following steps: S1. Multi-source data governance and multi-layered knowledge graph construction 1.1 Collect TCMID, TCMSP, SymMap, and BATMAN-TCM multi-source TCM datasets and extract six types of entities: prescriptions, medicinal materials, compounds, targets, diseases, and syndromes; 1.2 Entity Standardization: Establish a mapping table of aliases, pinyin, Chinese and English, and standard codes to distinguish between entities with the same name but different names, and record the original source of each data entry; classify the evidence of association relationships, retain all sources for conflicting data, and prioritize them according to experimental evidence; 1.3 A seven-layer knowledge graph was constructed, and all entity, association, and evidence attributes were entered to form a structured knowledge base of 578,000 entries; S2. Enhanced entity recognition and candidate evidence recall 2.1 Receive text input from users of any type, segment it into words, and then perform exact matching and fuzzy matching using entity dictionaries; 2.2 The identified entities are processed by synonym normalization, and common names, ancient names and abbreviations are uniformly converted into standard entities in the graph; for entities that cannot be matched precisely, candidates are expanded by combining Levenshtein edit distance and Jaccard similarity fuzzy matching to generate a candidate list of similar entities. 2.3 Recall all associated candidate entities and corresponding evidence along a preset path in the graph based on the input entity type: For symptom / syndrome input, candidate prescriptions are recalled along the path of "symptom → syndrome → prescription"; For prescription input, an association network is developed along the path of "prescription → medicinal materials → compound → target → pathway"; For disease / target input, candidate prescriptions are backtracked along the reverse path of "target → compound → medicinal material → prescription"; Record the source, matching method, and confidence score of each piece of related data; S3. Large-scale structured dialectical reasoning constrained by knowledge graphs 3.1 The candidate set, graph subgraph, and association evidence are encapsulated into a structured constraint Prompt, which explicitly restricts the model to using only entities within the recall scope for reasoning; 3.2 The large model is invoked to generate structured JSON format results. The structured JSON format results include a list of recommended prescriptions, reasons for recommendation, sources of evidence, confidence scores, and associated pathway information. 3.3 JSON format validation, filtering of abnormal output, and extraction of recommended prescriptions, medicinal materials, active ingredients, and targets for subsequent verification; S4, Underlying Database Reverse Verification 4.1 For all formulas, ingredients, targets, and pathways output by the model, back-check the underlying knowledge graph to verify whether the formula exists in the database, whether the composition of the medicinal materials is consistent with the database records, whether the target prediction of the ingredients is supported by literature, and whether the pathway enrichment results are consistent with the known mechanisms. 4.2 For projects that pass verification, increase the confidence level by adding database links and literature citations; for projects that fail verification, mark them as "pending confirmation" and indicate the risks. 4.3 Generate a complete verification report, which includes recommended conclusions, data basis, level of evidence, source tracing, and uncertainty explanation; all results retain source, version, parameters, and operation logs to ensure the reproducibility of the results; S5, Forward Network Pharmacological Computation 5.1 For the prescription or medicinal material input by the user, construct a positive correlation network along the path of "prescription or medicinal material-component-target-pathway"; 5.2 Evaluate the nodes and edges in the network; 5.3 Perform topological analysis on the network, calculate the degree centrality, betweenness centrality, and PageRank value of each node, and identify key nodes and key pathways in the network; at the same time, through KEGG / GO / DO enrichment analysis, output a list of significantly enriched pathways and their statistical parameters. S6, Reverse Network Pharmacology Computation 6.1 Reverse mapping from input disease / target to all active compounds, medicinal materials containing the compound, and complete prescription; 6.2 Analyze candidate prescriptions using multi-dimensional weighted scoring and output the top-N candidate prescriptions after sorting. 6.3 Statistically analyze the high-frequency co-occurrence drug pairs and drug group patterns of candidate prescriptions to complete the analysis of compatibility rules; S7, Multi-component synergistic quantization and 0 / 1 combinatorial optimization 7.1 Probability-based joint scoring, using the formula , where i=1,2,...,n; calculate the comprehensive efficacy of different combinations of medicinal materials on key targets; 7.20 / 1 combinatorial optimization, constructing a 0 / 1 optimization objective function: f(S) = a × Coverage(S,T) + b × Synergy(S) + c × Evidence(S) − d × Cost(S), where the weights a, b, c, and d can be user-defined. 7.3 The above 0 / 1 combinatorial optimization problem is solved by an optimization algorithm based on maximum likelihood estimation, and the optimized combination of medicinal materials, corresponding active ingredients, target sites, expected pathways and quantitative evaluation indicators are output. S8, Offline Enrichment Analysis and Interactive Visualization 8.1 Integrates an offline enrichment engine to perform ORA and GSEA analyses, and outputs pathway enrichment results, statistical parameters (p-value, FDR, NES), and visualization charts; 8.2 Generates multi-layer interactive multi-layer network diagrams, enrichment analysis bubble charts, target interaction network diagrams, and component-target heatmaps, supporting drag-and-drop filtering on the front end; 8.3 Supports exporting analysis results in multiple formats such as PDF reports, CSV data tables, PNG images, and complete JSON parameter files. All files include data version, source, and running logs to ensure the reproducibility of the research.

[0023] Example 1: Symptom-based AI-based diagnosis and prescription recommendation Step 1. The user enters a natural language description of the symptoms, such as: "knee pain, limited range of motion, morning stiffness, worsened by cold"; Step 2. The system performs entity recognition and identifies four symptom entities: "knee pain", "limited mobility", "morning stiffness", and "worsening with cold". Step 3. The system performs candidate recall along the "symptom-syndrome-prescription" path, matches possible syndromes based on symptom combinations, such as "wind-cold-dampness bi syndrome" and "liver and kidney deficiency", and then recalls classic prescriptions and modern prescriptions corresponding to the syndromes to generate a set of candidate prescriptions. Step 4. The system constructs a structured Prompt, inputs the candidate prescription set and its corresponding symptom-syndrome-prescription association evidence into the large model, and requires the model to perform dialectical reasoning and prescription recommendation within the candidate set; Step 5. The large model outputs structured JSON results, including the name of the recommended prescription, the reason for the recommendation, such as "This prescription is a classic prescription for treating wind-cold-dampness arthralgia. XX medicinal material in the prescription has anti-inflammatory and analgesic effects. Modern research has confirmed that its active ingredients can act on inflammatory targets such as COX-2 and IL-6", the source of evidence, and the confidence score. Step 6. The system performs reverse verification on the recommended prescriptions output by the model, and queries the knowledge graph to confirm whether the prescription composition, active ingredients of medicinal materials and targets are consistent with the database records; after the verification is passed, the confidence score is enhanced and the source link is attached. Step 7. The system outputs the final recommendation results, including the prescription composition, mechanism of action, level of evidence, and uncertainty explanation. Users can click on each evidence link to view the original database record and literature source.

[0024] Example 2: Analysis of the positive mechanism based on prescription Step 1. The user enters the prescription name "Guizhi Tang" and the system confirms the prescription entity through entity dictionary matching; Step 2. The system unfolds a positive correlation network along the path of "prescription-medicinal material-compound-target-pathway". First, the medicinal materials contained in the prescription (cinnamon twig, white peony root, ginger, jujube, licorice) are obtained, then the active ingredients of each medicinal material are obtained, then the target of each compound is obtained, and finally the biological pathways involved by the target are obtained. Step 3. The system scores and performs topology analysis on the network, calculates the degree centrality and PageRank value of each node, and identifies key targets (TNF, IL-6, NF-kappaB, etc.) and key pathways (NF-kappaB signaling pathway, TNF signaling pathway, etc.). Step 4. The system performs offline enrichment analysis of KEGG / GO / DO and outputs a list of significantly enriched pathways (p<0.05, FDR<0.25). Step 5. The system generates an interactive network visualization diagram, which users can view and interact with in a browser; in the network diagram, the prescription node is located at the center, the medicinal material nodes surround the prescription, the compound nodes connect the medicinal materials, the target node connects the compound, and the pathway node connects the target; the size of the node indicates its importance in the network, and the thickness of the edge indicates the strength of the association; Step 6. Users export the analysis results as PDF reports, CSV data tables, and PNG images. The reports contain complete analysis parameters, data sources, and version information.

[0025] Example 3: Reverse Candidate Discovery Based on Disease Step 1. The user enters the disease name "osteoarthritis". The system confirms the disease entity through entity dictionary matching and obtains the set of known targets related to the disease. Step 2. The system constructs a correlation network in reverse along the "target-compound-medicinal material-formula" path; first, it obtains all compounds that act on osteoarthritis-related targets, then it obtains medicinal materials containing these compounds, and finally it obtains formulas containing these medicinal materials. Step 3. The system performs multi-dimensional scoring on candidate prescriptions; the target coverage score calculates the coverage ratio of the active ingredients of the candidate prescriptions to osteoarthritis targets; the pathway enrichment significance score is calculated through hypergeometric distribution test; the evidence level score is based on the number of literature reports and experimental verifications; the network modularity score is based on the network clustering coefficient. Step 4. The system sorts the above scores and outputs the TOP-10 candidate prescriptions, such as "Du Huo Ji Sheng Tang" and "Bu Shen Huo Xue Tang". Step 5. The system analyzes the formulation rules of candidate prescriptions, extracts high-frequency co-occurring herb pairs, such as "Angelica pubescens-Taxillus chinensis" and "Eucommia ulmoides-Achyranthes bidentata", and compares them with classical prescription theory to provide an explanation for the rationality of candidate prescriptions; Step 6. Users can further verify the positive mechanism of the candidate prescriptions to see the specific molecular mechanism by which they act on the osteoarthritis target.

[0026] Example 4: Multi-component synergistic quantification and formulation optimization Step 1. The user inputs a set of key targets (such as IL-1beta, TNF-alpha, MMP-13, COX-2) and a set of candidate medicinal materials (such as Angelica pubescens, Taxillus chinensis, Eucommia ulmoides, Achyranthes bidentata, Gentiana macrophylla, Saposhnikovia divaricata). Step 2. The system obtains the active ingredients of each candidate medicinal material and their probability of action on each target. Step 3. The system performs probability joint scoring calculations to evaluate the comprehensive effects of different combinations of medicinal materials on key targets; for example, the probability of the combined effect of the combination "Angelica pubescens + Taxillus chinensis + Eucommia ulmoides" on IL-1beta is 0.85, and the probability of the combined effect on TNF-alpha is 0.82. Step 4. System settings optimization objectives: maximize target coverage, maximize synergy, maximize evidence level, minimize drug quantity, with weighting coefficients set to a=0.4, b=0.3, c=0.2, d=0.1; Step 5. The system uses a 0 / 1 combinatorial optimization algorithm to solve the problem and output the optimal combination of medicinal materials "Angelica pubescens + Taxillus chinensis + Eucommia ulmoides + Achyranthes bidentata". This combination contains 12 active ingredients, covering all 4 key targets. The average probability of combined action is 0.78 and the total level of evidence is 35 (out of 50). Step 6. The system outputs the optimization results, including recommended herbal combinations, a list of active ingredients, a matrix of target targets, expected pathways, quantitative evaluation indicators, and a comparative analysis with classic prescriptions.

[0027] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0028] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0029] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0030] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0031] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A traditional Chinese medicine intelligent analysis system based on knowledge graph constrained reasoning and bidirectional network pharmacology, characterized in that: The architecture adopts a five-layer design, consisting of, from top to bottom: application service layer, computation and analysis layer, inference engine layer, knowledge graph layer, and multi-source data layer; among which: The multi-source data layer is used to integrate publicly available TCM databases such as TCMID, TCMSP, SymMap, and BATMAN-TCM to complete entity normalization, conflict verification, evidence annotation, and output standardized raw datasets. The knowledge graph layer is used to construct a multi-layered association knowledge graph that connects symptoms, syndromes, prescriptions, medicinal materials, compounds, protein targets, pathways, and diseases. The graph relationships include prescription-medicinal materials, medicinal materials-compounds, compounds-targets, targets-pathways, syndromes-prescriptions, symptoms-syndromes, and targets-diseases. Each relationship is accompanied by source database, evidence level, and original literature tracing fields. The inference engine layer is used for trusted AI identification. It sequentially performs multimodal input entity recognition and synonym normalization, knowledge graph candidate evidence recall, graph-constrained large model structured reasoning, and underlying database reverse verification, and outputs the identification prescription recommendation results with confidence and traceability information. The computational analysis layer integrates a forward network pharmacology computation module, a reverse network pharmacology computation module, a multi-component synergistic quantification optimization module, and an interactive visualization module; among which, the multi-component synergistic quantification optimization module includes a probability joint scoring submodule and a 0 / 1 herbal combination optimization submodule; The application service layer provides three delivery forms: a web-based visual interactive interface, a RESTful API call interface, and a Docker offline private deployment image, and exposes all search, calculation, optimization, and export functions to the outside world; The system executes the following integrated analysis process: multimodal input → entity recognition and synonym normalization → knowledge graph candidate recall → knowledge graph constrained large model reasoning → database reverse verification → bidirectional network pharmacology calculation → multi-component synergistic quantification and 0 / 1 combination optimization → enrichment visualization and standardized report export.

2. A method for intelligent analysis of traditional Chinese medicine based on knowledge graph-constrained reasoning and bidirectional network pharmacology, executed using the intelligent analysis system for traditional Chinese medicine based on knowledge graph-constrained reasoning and bidirectional network pharmacology as described in claim 1, characterized in that... Includes the following steps: S1. Multi-source data governance and multi-layer knowledge graph construction; S2, Enhanced entity recognition and candidate evidence recall; S3, structured dialectical reasoning with knowledge graph constraints on large-scale models; S4. Reverse verification of the underlying database; S5, Forward Network Pharmacological Computation; S6, Reverse Network Pharmacology Calculation; S7, Multi-component synergistic quantization and 0 / 1 combinatorial optimization; S8, offline enrichment analysis and interactive visualization.

3. The intelligent analysis method for traditional Chinese medicine based on knowledge graph constrained reasoning and bidirectional network pharmacology according to claim 2, characterized in that, Multi-source data governance and multi-layered knowledge graph construction specifically include the following steps: (1) The data layer integrates publicly available TCM database resources and establishes a unified entity dictionary, relational schema and evidence field. The entity dictionary includes six core entities: prescriptions, medicinal materials, compounds, protein targets, diseases and syndromes. For each entity, a mapping relationship is established between Chinese and English names, pinyin, aliases and standard IDs. The standard IDs include PubChem ID, Gene Symbol, OMIM or ICD code. (2) By deduplication, synonym merging, conflict labeling, missing data checking, and source recording, the data version, original source, and processing process are preserved. For entities with the same name but different characteristics, they are distinguished and labeled according to the source database and the level of evidence; for relationships with inconsistent levels of evidence, the highest level is given priority, and conflict information is recorded for user reference. (3) Construct a multi-layer knowledge graph with "prescription-medicine-chemical-target" as the core, connecting disease, syndrome and pathway information. The graph association includes prescription-medicinal material, medicinal material-compound, compound-target, target-pathway, syndrome-prescription, symptom-syndrome and target-disease. Each relationship is accompanied by source database, evidence level and original literature tracing field. The association is divided into three levels of evidence according to experimental verification, literature report and database prediction.

4. The intelligent analysis method for traditional Chinese medicine based on knowledge graph constrained reasoning and bidirectional network pharmacology according to claim 3, characterized in that, Enhanced entity recognition and candidate evidence recall include the following steps: (1) Receive any one or more inputs such as natural language symptoms, prescriptions, names of medicinal materials, compounds, target IDs, gene names, and disease names. First, perform word segmentation and entity recognition on the input text, and use a pre-built entity dictionary for precise matching and fuzzy matching. For symptom descriptions, use a large language model for semantic understanding and entity extraction to identify symptom entities, syndrome candidates, and implicit prescription intentions. (2) The identified entities are normalized by synonyms, and aliases, common names and historical names are uniformly mapped to standard names; for entities that cannot be matched precisely, Levenshtein edit distance and Jaccard similarity are used to expand the candidates and generate a candidate list of similar entities. (3) Based on entity type and relational paths in the knowledge graph, retrieve candidate prescriptions, medicinal materials, ingredients, targets, and their associated evidence from the knowledge graph; the retrieval process follows these rules: For symptom / syndrome input, candidate prescriptions are recalled along the "symptom-syndrome-prescription" path; For prescription input, an association network is developed along the path of "prescription-medicinal material-compound-target-pathway"; For disease / target inputs, candidate prescriptions are traced back along the "target-compound-medicinal material-prescription" path, and the data source, matching method and confidence score of each related prescription are recorded simultaneously.

5. The intelligent analysis method for traditional Chinese medicine based on knowledge graph constrained reasoning and bidirectional network pharmacology according to claim 4, characterized in that, The structured dialectical reasoning of large-scale models constrained by knowledge graphs includes the following steps: (1) Using the recall candidate entity set, associated evidence and knowledge graph subgraph as context, construct structured prompt words. The prompt words clearly limit the model to reasoning and output only within the candidate set, and prohibit the model from generating conclusions beyond the candidate range based on prior knowledge in the training corpus. (2) Call a large model compatible with the OpenAI interface to output structured JSON format results; the JSON format results include a list of recommended prescriptions, reasons for recommendation, sources of evidence, confidence scores, and associated pathway information. The model output strictly follows JSONSchema constraints to ensure field integrity and type consistency; (3) Post-process the model output. The processing steps include JSON format verification, field integrity check, and entity name matching verification with the knowledge graph. For each recommended prescription output by the model, extract its corresponding medicinal material composition, active ingredients and target points, and prepare for reverse verification in step S4.

6. The intelligent analysis method for traditional Chinese medicine based on knowledge graph constrained reasoning and bidirectional network pharmacology according to claim 5, characterized in that, Reverse verification of the underlying database includes the following steps: (1) For the recommended prescriptions, medicinal materials, components and targets output by the model in step 3, perform reverse query verification against the underlying database one by one; the verification content includes: whether the prescription exists in the database, whether the composition of the medicinal materials is consistent with the database record, whether the target prediction of the components is supported by literature, and whether the pathway enrichment results are consistent with the known mechanisms. (2) For items that fail verification, mark them as "pending confirmation" and prompt the user; for items that pass verification, enhance their confidence score and attach database source links and original literature citations; (3) Generate a verification report, which includes the recommended conclusions, data basis, evidence level, source traceability and uncertainty explanation; all results retain the source, version, parameters and operation log to ensure the reproducibility of the results.

7. The intelligent analysis method for traditional Chinese medicine based on knowledge graph constrained reasoning and bidirectional network pharmacology according to claim 6, characterized in that, Forward network pharmacology calculations specifically include the following steps: (1) For the prescription or medicinal materials input by the user, construct a positive correlation network along the path of "prescription or medicinal material-component-target-pathway". First, obtain all the medicinal materials contained in the prescription, then obtain all the compounds corresponding to each medicinal material, then obtain the predicted or verified target corresponding to each compound, and finally obtain the biological pathways in which the target participates. (2) Nodes and edges in the network are scored; the weight of compound-target edge is calculated based on the number of literature reports, database prediction score and experimental verification level; the weight of target-pathway edge is calculated based on the enrichment significance p-value and pathway participation in the KEGG and GO databases. (3) Perform topological analysis on the network, calculate the degree centrality, betweenness centrality and PageRank value of each node, and identify key nodes, hub genes and key pathways in the network; at the same time, perform KEGG / GO / DO enrichment analysis through hypergeometric distribution test or Fisher's exact test, and output a list of significantly enriched pathways and their statistical parameters.

8. The intelligent analysis method for traditional Chinese medicine based on knowledge graph constrained reasoning and bidirectional network pharmacology according to claim 7, characterized in that, Reverse network pharmacology calculations specifically include the following steps: (1) For the disease name, target ID or gene name input by the user, construct a reverse association network along the path of "disease / target-component-medicinal material-prescription"; first obtain the relevant target from the disease, or directly from the target, then obtain the compound that acts on the corresponding target, then obtain the medicinal material containing the corresponding compound, and finally obtain the prescription containing the corresponding medicinal material. (2) The candidate prescriptions in the reverse network are scored in multiple dimensions, including target coverage, pathway enrichment significance, network modularity and evidence level; a comprehensive score is generated by weighted summation, and the TOP-N candidate prescriptions are output after sorting; the target coverage is the proportion of the active ingredients contained in the candidate prescriptions that cover the input target. (3) Analyze the formulation rules of candidate prescriptions, extract high-frequency co-occurring medicinal material pairs and drug group patterns, compare and verify them with classical prescription theory, and provide an explanation for the rationality of candidate prescriptions.

9. The intelligent analysis method for traditional Chinese medicine based on knowledge graph constrained reasoning and bidirectional network pharmacology according to claim 8, characterized in that, Multi-component synergistic quantization and 0 / 1 combinatorial optimization specifically include the following steps: (1) Probability joint scoring; For multiple active ingredients in the candidate prescription, calculate the comprehensive probability of their effect on key targets. Suppose that prescription F contains n active ingredients C1, C2, ..., Cn, and the set of key targets is T = {t1, t2, ..., tm}. Then the probability of active ingredient Ci acting on target tj is p(Ci->tj), which is obtained by normalizing the predicted scores based on literature reports and databases. The combined effect probability of multiple components on target tj was calculated using a probabilistic joint scoring model: where i = 1, 2, ..., n; This formula is based on the principle of the complement of independent event probabilities. It assumes that the effects of each component on the target are independent of each other. Then the probability that at least one component acts on the target is 1 minus the probability that none of the components act on the target. (2) 0 / 1 combinatorial optimization; the selection of candidate medicinal materials is transformed into a 0 / 1 combinatorial optimization problem: given a set of candidate medicinal materials H={h1,h2,...,hk}, each medicinal material hi contains a set of active ingredients, select a subset of medicinal materials S that is included in H, so as to maximize the objective function; the objective function is defined as: f(S)=a×Coverage(S,T)+b×Synergy(S)+c×Evidence(S)-d×Cost(S); Where Coverage(S,T) represents the coverage of subset S to the set of key targets T, Synergy(S) represents the synergistic strength of components within subset S, Evidence(S) represents the sum of evidence levels for the association between components and targets within subset S, and Cost(S) represents the cost of medicinal materials in subset S. a, b, c, and d are weighting coefficients, which can be adjusted by the user according to research needs; (3) The above 0 / 1 combinatorial optimization problem is solved by using an optimization algorithm based on maximum likelihood estimation. For a small candidate set of k<=20, exhaustive search is used to ensure global optimum. For a large candidate set of k>20, genetic algorithm or simulated annealing algorithm is used for approximate solution. The number of iterations and convergence threshold are set to ensure the quality of the solution. The optimized combination of medicinal materials, corresponding active ingredients, target sites, expected pathways and quantitative evaluation indicators are output.

10. The intelligent analysis method for traditional Chinese medicine based on knowledge graph constrained reasoning and bidirectional network pharmacology according to claim 9, characterized in that, Offline enrichment analysis and interactive visualization include the following steps: (1) It integrates an offline KEGG / GO / DO enrichment analysis engine, which does not rely on external network services, ensuring the stability and reproducibility of the computing environment; it supports two enrichment methods, ORA and GSEA, and outputs pathway enrichment results, statistical parameters and visualization charts. The statistical parameters include p-value, FDR and NES. (2) Generate interactive network visualization diagrams, including multi-layer network diagrams of prescription-medicinal materials-compound-target-pathway, enrichment analysis bubble diagrams, target interaction network diagrams, and component-target heatmaps; users can zoom, drag, filter, and export through the Web interface; (3) Supports exporting analysis results in multiple formats, including PDF reports, CSV data tables, PNG images and JSON parameter files. All exported files contain analysis parameters, data sources and version information to ensure the reproducibility of the research process.

Images