A knowledge graph construction system of a drug database

Abstract

The application relates to the technical field of drug knowledge graph, and provides a knowledge graph construction system of a drug database, which comprises a data acquisition module, an analysis module, a graph module, a risk assessment module and a verification module; key characteristic data information and graph abnormal characteristic parameter information of target compound drugs of various compound structure types in the drug database are acquired; drug key characteristic factors and graph abnormal characteristic representation values are analyzed; whether the system operation conforms to the standard is determined through the graph module; if the standard is not met, the verification module is started based on the graph structure risk tendency type; and corresponding processing modes are adopted based on the determined reasons for not meeting the standard. Discrete drug data is converted into a calculable and explorable relationship network, a structured knowledge graph is provided for drug research and development and clinical research, and the system efficiency and precision are improved.

Description

Technical Field

[0001] This invention relates to the field of drug knowledge graph technology, and more particularly to a knowledge graph construction system for a drug database. Background Technology

[0002] The drug development field has long faced core challenges, including difficulties in integrating multi-source heterogeneous data and insufficient mining of complex relationships. Traditional drug databases, such as DrugBank and PubChem, store massive amounts of information, including compound structures, target interactions, and clinical trial results. However, the data is often presented in tabular or isolated entries, making it difficult to intuitively represent the multidimensional relationships between drugs, targets, diseases, and side effects. For example, drug interaction analysis requires manual searching across multiple databases, which is inefficient and prone to missing implicit correlations. Therefore, there is an urgent need for software that can integrate and correlate multi-source drug data.

[0003] Chinese Patent Publication No. CN120375910A discloses a method and system for constructing a pharmaceutical knowledge graph based on artificial intelligence. This method collects the molecular structures of drugs and their corresponding target protein sequences through database queries. The molecular structure data is numerically encoded to extract molecular fingerprints and protein domain features. These features are then combined with the drug's chemical properties and protein sequence characteristics to form a drug-protein feature set. Through precise analysis of drug molecular structures and their target protein sequences, this innovative approach significantly enhances the understanding of drug-protein interactions. Researchers can directly extract key features from the data and monitor the dynamic changes in drug effects. This not only accelerates the drug development process and optimizes treatment plans, but also provides strong data support for personalized medicine by dynamically tracking the interaction between drug side effects and pathological features.

[0004] Chinese Patent Publication No. CN117827898A discloses a data stream processing method for dynamically updating knowledge graphs. This method acquires updated data streams from multiple data channels to form a real-time data stream. The real-time data stream is then parsed, validated, and processed to form drug feature information. Drug information within the feature information group is transformed using a bag-of-words model or word embedding method to form a vector-based data group. Multiple directed edge models are then established on this vector-based data group using a graph database or graph algorithm to simulate the multifaceted interactions between drugs and drug features, creating a set of directed edge models. This set is then matched against existing directed edge models in the system to generate directed edge query results. The set of directed edge models is dynamically updated based on these query results, achieving real-time updates of drug relationships regarding drug interactions and side effects.

[0005] However, the following problems still exist in the existing technology:

[0006] Existing technologies do not consider the data isolation of drug databases, making it difficult to reflect the relationship between drugs, targets, diseases, and side effects from multiple dimensions, as well as the differences in the factors that influence the relationship, resulting in low system efficiency and low accuracy. Summary of the Invention

[0007] To address this, the present invention provides a knowledge graph construction system for drug databases, which overcomes the problems of low efficiency and low accuracy in existing technologies that do not consider the data isolation of drug databases, make it difficult to reflect the correlation between drugs, targets, diseases, and side effects in multiple dimensions, and the differences in the degree of correlation.

[0008] To achieve the above objectives, the present invention provides a knowledge graph construction system for drug databases, comprising:

[0009] The data acquisition module is used to collect key characteristic data and abnormal spectral characteristic parameters of target compound drugs of various compound structure types in the drug database.

[0010] An analysis module, which is connected to the data acquisition module, is used to analyze key drug characteristic factors based on the key characteristic data information and to analyze the abnormal characteristic characterization value of the spectrum based on the abnormal characteristic parameter information of the spectrum.

[0011] The spectral module, which is connected to the analysis module, is used to determine whether the system operation meets the standard based on the difference between a predetermined threshold value for spectral anomaly characteristics based on the compound structure type and the actual spectral anomaly characteristics.

[0012] The risk assessment module is connected to the analysis module and the spectrum module respectively. It is used to determine the risk tendency type of the spectrum structure based on the key drug characteristic factors of the compound structure type, and to determine the activation verification module based on the risk tendency type of the spectrum structure.

[0013] The verification module, which is connected to the risk assessment module, is used to determine the reasons for the system's non-compliance with standards and the handling methods based on the difference between the drug's key characteristic factors and the predetermined thresholds for the drug's key characteristic factors: determining the magnitude of the decrease in the threshold of the abnormal feature characterization value of the spectrum, or determining the magnitude of the increase in the threshold of the drug's key characteristic factors.

[0014] The key feature data information includes: the number of drug entity nodes and the number of drug-related edges; the graph anomaly feature parameter information includes drug duplication rate and drug error rate.

[0015] Furthermore, the number of drug entity nodes is determined based on the number of target gene information, the number of disease associations, the number of side effect records, and the number of clinical trials.

[0016] Furthermore, the number of drug-associated edges is determined based on the number of drug compounds binding to the target, the number of diseases treated by the drug, and the number of side effects caused by the drug.

[0017] Furthermore, the key drug characteristic factors are determined based on the sum of a first key characteristic factor and a second key characteristic factor, wherein:

[0018] The first key feature factor is determined based on the ratio of the number of drug entity nodes to a predetermined threshold for the number of drug entity nodes.

[0019] The second key feature factor is determined based on the ratio of the number of drug-related edges to a predetermined threshold for the number of drug-related edges.

[0020] Furthermore, the anomaly feature characterization value of the map is determined based on the sum of the first anomaly feature factor and the second anomaly feature factor, wherein:

[0021] The first abnormal feature factor is determined based on the ratio of a predetermined drug repetition rate threshold to the actual drug repetition rate.

[0022] The second abnormal feature factor is determined based on the ratio of a predetermined drug error rate threshold to the actual drug error rate.

[0023] Preferably, the system operation does not meet the standard when the difference between the predetermined threshold value of the abnormal spectral feature characterization value of the target compound structure type and the actual value of the abnormal spectral feature characterization value is greater than the predetermined difference threshold value.

[0024] Preferably, the criterion for determining the risk tendency type of the spectral structure is the compound structure type where the drug key feature factor is greater than a predetermined drug key feature factor threshold.

[0025] Preferably, the reasons why the system does not meet the standard include a first reason label and a second reason label, wherein:

[0026] If the difference between the key drug feature factor and the predetermined key drug feature factor threshold is less than or equal to the predetermined difference threshold, it is determined as the first cause label and the first processing method is executed.

[0027] If the difference between the key drug feature factor and the predetermined key drug feature factor threshold is greater than the predetermined difference threshold, it is determined as the second cause label and the second processing method is executed.

[0028] Preferably, the processing method includes a first processing method and a second processing method, wherein:

[0029] The first processing method is to determine the reduction range of the threshold for the characterization value of the abnormal feature of the spectrum based on the difference between the key feature factor of the drug and the predetermined threshold of the key feature factor of the drug.

[0030] The second processing method involves determining the increase in the threshold of key drug characteristic factors.

[0031] Preferably, the system operates under the following condition: the difference between the predetermined threshold value of the abnormal spectral feature characterization value of the target compound structure type and the actual abnormal spectral feature characterization value is less than or equal to the predetermined difference threshold value.

[0032] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention provides a knowledge graph construction system for drug databases, including: a data acquisition module, an analysis module, a graph module, a risk assessment module, and a verification module. It collects key characteristic data information of target compounds of various compound structural types from the drug database, as well as abnormal characteristic parameter information of the graph. It analyzes key drug characteristic factors and abnormal graph characteristic values. The graph module determines whether the system operation meets the standards. If it does not meet the standards, it determines the risk tendency type of the graph structure and initiates the verification module. Based on the difference between the key drug characteristic factors and a predetermined threshold for key drug characteristic factors, it determines the reason for the system's non-compliance and the handling method: determining whether to lower the threshold for abnormal graph characteristic values ​​or determine the extent to which to increase the threshold for key drug characteristic factors. This overcomes the problems of isolated data in drug databases, making it difficult to represent the multi-dimensional relationships between drugs, targets, diseases, and side effects, and the resulting low system efficiency and accuracy due to differences in the degree of correlation. It transforms discrete drug data into a computable and explorable relational network, providing a structured knowledge foundation for drug development and clinical research, and improving the system's operational efficiency and accuracy.

[0033] In particular, it supports the parsing of compound structures from drug databases, including SDF, SMILES formats, bioactivity data, target gene information, disease associations, side effect records, clinical trials, and other multi-source heterogeneous data. Through standardized cleaning, molecular structure normalization, redundancy removal, and entity recognition (such as drug names, indications, and molecular formulas), a unified data model is constructed. This model integrates chemical structure libraries, such as ChEMBL, bioactivity databases, and patent literature, achieving semantic association and consistent mapping across source data. Drug entities—compounds, targets, diseases, and side effects—are abstracted as nodes, and their relationships, such as compound-target binding, drug treatment of diseases, and drug-induced side effects, are abstracted as edges, forming a multi-dimensional drug knowledge graph. This graph supports incremental updates, automatically expanding nodes and edges when new data, such as new drug patents or clinical trials, is added, maintaining the timeliness of the graph.

[0034] Furthermore, this invention uses Python to quickly preprocess data from various drug databases, employs graph databases such as Neo4j to store data, uses the Cypher query language to achieve efficient relational retrieval, supports complex relational traversal such as multi-hop queries, and has a built-in data preprocessing module that supports molecular structure standardization, redundancy removal, and classification indexing. It also integrates an IK word segmenter to expand the medical dictionary and improve the accuracy of entity links; and combines biomedical ontology, such as disease ontology, to strengthen semantic associations.

[0035] In particular, evaluating the quality of drug data integration by quantifying the graph topology can improve the standardization of metrics and key factor extraction in the knowledge graph construction process. Key feature factors are generated by standardizing the drug entity node count and drug association edge count. This enables objective evaluation of the construction process, transforming abstract data integrity and association richness features into calculable numerical metric fusion dimensions, reducing bias from subjective judgment. Continuous monitoring of changes in key feature factors allows for rapid diagnosis of weaknesses in the data integration process and timely warnings of graph structure risks. Adaptive optimization scientifically adjusts data cleaning and entity alignment strategies or thresholds, further improving the knowledge graph's coverage, connectivity, and reliability, ensuring its quality as a basis for analytical decision-making.

[0036] Furthermore, by extracting and evaluating key drug characteristic factors for individual compound structure types, a highly efficient automated risk assessment mechanism was constructed. Through refined risk localization, the monitoring granularity is refined from the overall spectrum to specific compound structure types, enabling precise identification of which specific types of compound data have insufficient coverage or missing correlations. By comparing key factors with predetermined thresholds, the system can automatically classify a large number of compound types into risk-prone and non-risk-prone categories. For high-risk types, in-depth validation or data completion processes can be triggered first; for low-risk types, routine monitoring is maintained to ensure system operational efficiency. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the knowledge graph construction system for drug databases according to an embodiment of the present invention;

[0038] Figure 2 This is a flowchart illustrating the steps of a knowledge graph construction method for a drug database according to an embodiment of the present invention.

[0039] Figure 3 This is a flowchart illustrating the steps involved in determining key drug characteristic factors in an embodiment of the present invention.

[0040] Figure 4 This is a logic block diagram for determining the risk tendency type of the graph structure in an embodiment of the invention. Detailed Implementation

[0041] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0042] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0043] It should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0044] Please see Figure 1 As shown, Figure 1 This is a schematic diagram of a knowledge graph construction system for a drug database according to an embodiment of the present invention. The present invention provides a knowledge graph construction system for a drug database, comprising:

[0045] The data acquisition module is used to collect key characteristic data and abnormal spectral characteristic parameters of target compound drugs of various compound structure types in the drug database.

[0046] An analysis module, which is connected to the data acquisition module, is used to analyze key drug characteristic factors based on the key characteristic data information and to analyze the abnormal characteristic characterization value of the spectrum based on the abnormal characteristic parameter information of the spectrum.

[0047] The spectral module, which is connected to the analysis module, is used to determine whether the system operation meets the standard based on the difference between a predetermined threshold value for spectral anomaly characteristics based on the compound structure type and the actual spectral anomaly characteristics.

[0048] The risk assessment module is connected to the analysis module and the spectrum module respectively. It is used to determine the risk tendency type of the spectrum structure based on the key drug characteristic factors of the compound structure type, and to determine the activation verification module based on the risk tendency type of the spectrum structure.

[0049] The verification module, which is connected to the risk assessment module, is used to determine the reasons for the system's non-compliance with standards and the handling methods based on the difference between the drug's key characteristic factors and the predetermined thresholds for the drug's key characteristic factors: determining the magnitude of the decrease in the threshold of the abnormal feature characterization value of the spectrum, or determining the magnitude of the increase in the threshold of the drug's key characteristic factors.

[0050] The key feature data information includes: the number of drug entity nodes and the number of drug-related edges; the graph anomaly feature parameter information includes drug duplication rate and drug error rate.

[0051] It is understandable that there are no limitations on the construction and parsing of the drug database. The key characteristic data and abnormal feature parameters of target compounds of various drug compound structure types can be collected. The drug database for examples includes compound structures, and examples are in SDF / SMILES format, bioactivity data, target gene information, disease associations, side effect records, clinical trials, and other multi-source heterogeneous data. Then, through standardized cleaning—molecular structure normalization, redundancy removal, and entity recognition—drug name, indication, and molecular formula—a unified data model is constructed. This model integrates the chemical structure library ChEMBL, bioactivity databases, patent literature, etc., to achieve semantic association and consistent mapping across source data. Drug entities—compounds, targets, diseases, and side effects—are abstracted as nodes, and relationships—compound-target binding, drug treatment of diseases, and drug-induced side effects—are abstracted as edges, forming a multi-dimensional drug knowledge graph. This graph supports incremental updates, automatically expanding nodes and edges when new data—new drug patents or clinical trials—is added, maintaining the timeliness of the graph.

[0052] It is understood that the classification of compound structure types in the drug database is not limited. The examples are categorized as follows: small molecule drugs and large molecule drugs. Small molecule drugs typically have a molecular weight less than 900 Da. These are further classified as: aromatic and heteroaromatic compounds (e.g., aspirin, gefitinib); aliphatic compounds (e.g., statins – lipid-lowering drugs); steroids (e.g., hydrocortisone, estrogens); alkaloids (e.g., quinine); antibiotics (e.g., β-lactams – penicillin, macrolides – azithromycin, tetracyclines); peptides and peptide mimics (composed of amino acids, with molecular weights between small and large molecules, such as insulin, some anticoagulants); and large molecule drugs: monoclonal antibodies (e.g., pembrolizumab – Keytruda); recombinant proteins (e.g., insulin, erythropoietin); antibody-drug conjugates (e.g., DS-8201); and nucleic acid drugs (e.g., antisense oligonucleotides, siRNA, mRNA vaccines). Other methods may also be used, but will not be elaborated upon here.

[0053] The embodiment collects drug-related data from multi-source heterogeneous data, medical literature, clinical trial reports, and drug instructions. Through rule matching and outlier detection, duplicate and erroneous data are removed, and the data is standardized. Based on a deep learning model and a drug domain dictionary, entities such as drugs, diseases, targets, and adverse reactions, as well as their interrelationships, are identified. The extracted entities and relationships are stored in a structured manner, and a drug knowledge graph is constructed using a graph database. It provides functions such as visualization, intelligent question answering, and association analysis, supporting users to quickly query drug action mechanisms and drug interaction information.

[0054] Understandably, the input file contains information files from various drug databases, such as DrugBank and PubChem, and the storage path for the graph database knowledge graph. The processing method includes: reading the input file, converting the downloaded drug database files into JSON format, parsing the drug entity information, targets, and corresponding disease relationship pairs after obtaining the JSON file, organizing and merging the relationship pairs converted from different drug databases, importing the organized relationship pairs into the Neo4j database, and outputting a drug relationship knowledge graph integrating multiple drug databases.

[0055] It is understandable that there are no restrictions on the method of collecting key feature data information, as long as the number of drug entity nodes and the number of drug-related edges can be obtained. Other methods can also be used, which will not be elaborated here.

[0056] It is understandable that there are no restrictions on the method of collecting the abnormal feature parameter information of the spectrum. As long as the drug repetition rate and drug error rate in drug rule matching and outlier detection can be obtained, other methods can also be used, which will not be elaborated here.

[0057] This embodiment describes a knowledge graph construction system for drug databases, comprising a data acquisition module, an analysis module, a graph module, a risk assessment module, and a validation module. It collects key characteristic data and abnormal graph parameter information of target compounds of various compound structural types from the drug database. It analyzes key drug characteristic factors and abnormal graph characteristic values. The graph module determines whether the system operation meets standards. If it does not meet standards, it identifies the risk tendency type of the graph structure and initiates the validation module. Based on the difference between the key drug characteristic factors and predetermined thresholds, it determines the reason for the system's non-compliance and the corresponding handling method: adjusting the threshold for abnormal graph characteristic values ​​or adjusting the threshold for key drug characteristic factors. This overcomes the problems of isolated data in drug databases, making it difficult to represent the multi-dimensional relationships between drugs, targets, diseases, and side effects, leading to low system efficiency and accuracy. It transforms discrete drug data into a computable and explorable relational network, providing a structured knowledge foundation for drug development and clinical research, thus improving system efficiency and accuracy.

[0058] Please see Figure 2 The diagram shows a flowchart of the steps in a knowledge graph construction method for a drug database according to an embodiment of the present invention. The steps of the knowledge graph construction method for a drug database according to the embodiment include:

[0059] Step S1: Obtain key feature data information of target compound drugs of various compound structure types in the drug database, and analyze key feature factors of the drug based on the key feature data information;

[0060] Step S2: Obtain the spectral anomaly feature parameter information of target compound drugs of each compound structure type in the drug database, and analyze the spectral anomaly feature characterization value based on the spectral anomaly feature parameters;

[0061] Step S3: Determine whether the system operation meets the standard by comparing the difference between the predetermined threshold value of the abnormal spectral features based on the compound structure type and the actual value of the abnormal spectral features.

[0062] Step S4: If it is initially determined that the system operation does not meet the standard, determine the risk tendency type of the spectrum structure based on the key drug characteristic factors of the compound structure type, and determine whether to start the verification module based on the risk tendency type of the spectrum structure.

[0063] Step S5: Determine the reasons for the system's non-compliance with standards and the handling methods based on the difference between the key drug characteristic factors and the predetermined threshold of the key drug characteristic factors: determine the reduction range of the threshold value of the abnormal feature in the spectrum, or determine the increase range of the threshold value of the key drug characteristic factors.

[0064] The embodiment establishes a dynamic assessment-diagnosis-regulation closed loop through intelligent quality monitoring and adaptive optimization of the drug knowledge graph. By comparing abnormal feature values ​​and thresholds of the graph in real time, deviations and anomalies in the data integration or mapping process are accurately identified, ensuring that the quality of the graph construction meets standards. Key feature factors are used to determine the type of structural risk, and intelligent decisions are made on whether to initiate verification to prevent the spread of erroneous associations. By quantifying the deviation between key factor thresholds and abnormal characterization values, the system threshold parameters are automatically adjusted to improve the robustness and accuracy of the knowledge graph and ensure its reliability in incremental updates.

[0065] Specifically, the number of drug entity nodes is determined based on the number of target gene information, the number of disease associations, the number of side effect records, and the number of clinical trials.

[0066] Preferably, the number of drug entity nodes in the embodiment is the sum of the number of corresponding drug target gene information, the number of disease associations, the number of side effect records, and the number of clinical trials.

[0067] Specifically, the number of drug-associated edges is determined based on the number of drug compounds binding to targets, the number of diseases treated by the drug, and the number of side effects caused by the drug.

[0068] Preferably, the number of drug-related edges in the embodiment is the sum of the number of drug compounds binding to the target, the number of diseases treated by the drug, and the number of side effects caused by the drug.

[0069] The embodiments use Python to quickly preprocess data from various drug databases, employ graph databases such as Neo4j to store data, and use the Cypher query language to achieve efficient relational retrieval, supporting complex relationship traversal, such as multi-hop queries. The built-in data preprocessing module supports molecular structure standardization, redundancy removal, and classification indexing. It integrates an IK word segmenter to expand the medical dictionary and improve the accuracy of entity links. It also combines biomedical ontology, such as disease ontology, to strengthen semantic associations.

[0070] Specifically, the key drug characteristic factors are determined based on the sum of a first key characteristic factor and a second key characteristic factor, wherein:

[0071] The first key feature factor is determined based on the ratio of the number of drug entity nodes to a predetermined threshold for the number of drug entity nodes.

[0072] The second key feature factor is determined based on the ratio of the number of drug-related edges to a predetermined threshold for the number of drug-related edges.

[0073] Please see Figure 3 The diagram shown is a flowchart illustrating the steps for determining key drug characteristic factors according to an embodiment of the present invention. The steps for determining key drug characteristic factors in the embodiment include:

[0074] Extract the number of drug entity nodes and drug association edges for target compound drugs of each compound structure type from the drug database;

[0075] The ratio of the number of drug entity nodes to a predetermined threshold for the number of drug entity nodes is determined as the first key feature factor.

[0076] The ratio of the number of drug-related edges to a predetermined threshold for the number of drug-related edges is determined as the second key feature factor;

[0077] The sum of the first key feature factor and the second key feature factor is determined as the drug key feature factor.

[0078] The example abstracts drug entities—compounds, targets, diseases, side effects, and number of clinical trials—as nodes, and the relationships—compound-target binding, drug treatment of diseases, and drug-induced side effects—as edges, forming a multidimensional drug knowledge graph.

[0079] It is understood that the predetermined threshold for the number of drug entity nodes is predetermined. Specifically, the average number of drug entity nodes for the target compound drug in each compound structure type in the drug database is recorded in advance. The product of the average number of drug entity nodes and the accuracy coefficient is determined as the threshold for the number of drug entity nodes. The accuracy coefficient is selected in the range [0.95, 0.99], preferably 0.98.

[0080] Similarly, the predetermined threshold for the number of drug-related edges is predetermined. The average number of drug-related edges for the target compound drug in each compound structure type in the drug database is recorded in advance. The sum of the average number of drug-related edges and the deviation coefficient is determined as the threshold for the number of drug-related edges. The deviation coefficient is selected in the range [1.05, 1.15], preferably 1.09.

[0081] This embodiment evaluates the quality of drug data integration by quantifying the topological structure of the knowledge graph, which improves the standardization of metrics and key factor extraction in the knowledge graph construction process. Key feature factors are generated by standardizing the number of drug entity nodes and drug association edges. This enables objective evaluation of the construction process, transforming abstract data integrity and association richness features into calculable numerical indicators, reducing bias from subjective judgments. Continuous monitoring of changes in key feature factors allows for rapid diagnosis of weaknesses in the data integration process and timely warnings of graph structure risks. Adaptive optimization scientifically adjusts data cleaning and entity alignment strategies or thresholds, further improving the coverage, connectivity, and reliability of the knowledge graph, ensuring its quality as a basis for analytical decision-making.

[0082] Specifically, the anomaly feature representation value of the map is determined based on the sum of a first anomaly feature factor and a second anomaly feature factor, wherein:

[0083] The first abnormal feature factor is determined based on the ratio of a predetermined drug repetition rate threshold to the actual drug repetition rate.

[0084] The second abnormal feature factor is determined based on the ratio of a predetermined drug error rate threshold to the actual drug error rate.

[0085] It is understood that the predetermined drug repetition rate threshold is predetermined. The average number of drug repetitions of the target compound drug in each compound structure type in the drug database is recorded in advance. The product of the average number of drug repetitions in the system and the accuracy coefficient is determined as the drug repetition rate. The accuracy coefficient is selected in the range [0.95, 0.99], preferably 0.98.

[0086] Similarly, the predetermined drug error rate threshold is pre-determined, wherein the average number of drug errors for the target compound drug in each compound structure type in the drug database is pre-recorded, and the product of the average number of drug errors in the system and the accuracy coefficient is determined as the drug error rate. The accuracy coefficient is selected in the range [0.95, 0.99], preferably 0.98.

[0087] This implementation focuses on data duplication and errors to achieve precise quantitative evaluation and anomaly monitoring of the data cleaning effect of a drug knowledge graph. An efficient and computable data integrity early warning mechanism is established. Through two measurable indicators—duplication rate and error rate—comparable anomaly characteristic factors are generated. By continuously calculating the deviation between actual values ​​and thresholds, redundant entities or mapping errors generated during data integration and standardization cleaning are monitored in real time, triggering alarms promptly. A comprehensive graph anomaly characteristic representation value serves as an overall quality score, providing an objective and unified benchmark for judging whether the system's operating status meets standards, reducing misjudgments based on subjective experience. Adaptive optimization lays the foundation; feature representation can accurately pinpoint whether a problem stems from data redundancy or data errors, guiding the system to activate either the deduplication module or the entity verification module, achieving precise scheduling of processing strategies and improving the accuracy and reliability of the knowledge graph.

[0088] Specifically, the system operation does not meet the standard when the difference between the predetermined threshold value of the abnormal spectral feature characterization value of the target compound structure type and the actual abnormal spectral feature characterization value is greater than the predetermined difference threshold value.

[0089] The example extracts the difference between a predetermined threshold value for spectral anomaly characteristics of the target compound structure type from the drug database and the actual spectral anomaly characteristic values.

[0090] If the difference is greater than the predetermined difference threshold, it is preliminarily determined that the system operation does not meet the standard, and the risk tendency type of the drug key characteristic factor determination map is based on the compound structure type.

[0091] If the difference is less than or equal to a predetermined difference threshold, the operation is determined to meet the standard, and monitoring continues.

[0092] In this embodiment, the threshold value for the characterization value of the abnormal feature of the spectrum is selected within the range of [2.05, 2.35]. Preferably, the threshold value for the characterization value of the abnormal feature of the spectrum is 2.25.

[0093] This embodiment establishes an automated and intelligent quality monitoring and decision-making closed loop. By calculating the deviation between the anomaly characteristic values ​​of the spectrum and the threshold in real time, it can accurately quantify and judge the system's operating status. When the difference exceeds a predetermined threshold, the system can automatically trigger a risk warning and intelligently diagnose the risk type based on key drug characteristic factors, thus providing a scientific basis for deciding whether to initiate a high-cost validation module. Conversely, if the difference is less than or equal to the predetermined threshold, it indicates that the system is operating well and requires continuous monitoring.

[0094] Specifically, the criterion for determining the risk tendency type of the spectrum structure is the compound structure type where the drug key feature factor is greater than a predetermined drug key feature factor threshold.

[0095] Please see Figure 4 As shown, this is a logic block diagram for determining the risk tendency type of the graph structure according to an embodiment of the present invention. The process for determining the risk tendency type of the graph structure according to an embodiment of the present invention includes:

[0096] Extract key drug characteristic factors for individual compound structure types from drug databases;

[0097] If the key characteristic factor of a drug is greater than the predetermined threshold for key characteristic factors of a drug, it is determined to be a risk tendency type of the spectrum structure, and the verification module is activated.

[0098] If the key characteristic factor of a drug is less than or equal to the predetermined threshold of key characteristic factor of a drug, it is determined to be a non-risk-prone type of the spectrum structure, and monitoring continues.

[0099] The predetermined threshold for key drug characteristic factors is selected within the range [2.05, 2.35]. In this embodiment, the threshold for key drug characteristic factors is preferably 2.15.

[0100] The core effect of the embodiments lies in achieving precise and refined risk screening and classification management of the data quality of compounds with specific structural types in the drug knowledge graph.

[0101] This embodiment constructs a highly efficient automated risk assessment mechanism by extracting and evaluating key drug characteristic factors for individual compound structure types. Through refined risk localization, the monitoring granularity is refined from the overall spectrum to specific compound structure types, enabling precise identification of specific types of compound data with insufficient coverage or missing correlations. By comparing key factors with predetermined thresholds, the system can automatically classify a large number of compound types into risk-prone and non-risk-prone categories. For high-risk types, in-depth validation or data completion processes can be triggered first; for low-risk types, routine monitoring is maintained to ensure system operational efficiency.

[0102] Specifically, the reasons why the system does not meet the standards include a first reason label and a second reason label, wherein:

[0103] If the difference between the key drug feature factor and the predetermined key drug feature factor threshold is less than or equal to the predetermined difference threshold, it is determined as the first cause label and the first processing method is executed.

[0104] If the difference between the key drug feature factor and the predetermined key drug feature factor threshold is greater than the predetermined difference threshold, it is determined as the second cause label and the second processing method is executed.

[0105] Specifically, the processing method includes a first processing method and a second processing method, wherein:

[0106] The first processing method is to determine the reduction range of the threshold for the characterization value of the abnormal feature of the spectrum based on the difference between the key feature factor of the drug and the predetermined threshold of the key feature factor of the drug.

[0107] The second processing method involves determining the increase in the threshold of key drug characteristic factors.

[0108] It is understood that no limit is placed on the difference threshold. In this embodiment, the difference threshold is selected within the range [0.05, 0.45], and preferably the difference threshold is 0.15.

[0109] The embodiment constructs a highly intelligent, adaptive feedback adjustment closed-loop system capable of accurately diagnosing and self-optimizing anomalies during the drug knowledge graph construction process, significantly improving the system's automation level and reliability. By categorizing non-compliance reasons into primary and secondary labels, the primary label indicates that while the data coverage and richness of key drug feature factors are close to the expected goals, the graph contains errors and duplications that do not meet the standards. This signifies problems in the data cleaning and standardization processes, such as deduplication failures and entity alignment errors, resulting in insufficient data accuracy. The secondary label indicates that key feature factors do not meet the standards, providing a solid foundation for subsequent differentiated self-correction.

[0110] The first processing method in the embodiment is to reduce the anomaly threshold. When a data quality problem is diagnosed, the threshold of the anomaly feature representation value of the map is actively reduced to increase the tolerance benchmark for data errors and duplications. In subsequent operation, the system will improve the accuracy of data cleaning quality.

[0111] The second processing method in the embodiment is to increase the threshold of key factors. By increasing the threshold of key drug characteristic factors, the accuracy of the target of data volume and correlation richness is improved. The system will actively incentivize the data integration and expansion process to find and absorb data from more sources to fill the data gap of this compound structure type, thereby improving the accuracy of the system.

[0112] Specifically, the system operation in the embodiment meets the standard condition that the difference between the predetermined threshold value of the abnormal spectral feature characterization value of the target compound structure type and the actual abnormal spectral feature characterization value is less than or equal to the predetermined difference threshold value.

[0113] This example establishes a clear and quantifiable compliance judgment benchmark for the quality monitoring system of a drug knowledge graph. By comparing the difference between a predetermined threshold for abnormal feature representation values ​​and the actual values ​​with a pre-defined allowable difference threshold, the system achieves objective and automated binary judgment of operational status. This reduces the occurrence of subjective assumptions and misjudgments, ensuring the uniqueness and repeatability of the evaluation results. By transforming the complex concept of system operational quality into a simple mathematical comparison problem, the stability and accuracy of the knowledge graph construction process are improved.

[0114] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A knowledge graph construction system for drug databases, characterized in that: The data acquisition module is used to collect key characteristic data and abnormal spectral characteristic parameters of target compound drugs of various compound structure types in the drug database. An analysis module, which is connected to the data acquisition module, is used to analyze key drug characteristic factors based on the key characteristic data information and to analyze the abnormal characteristic characterization value of the spectrum based on the abnormal characteristic parameter information of the spectrum. The spectral module, which is connected to the analysis module, is used to determine whether the system operation meets the standard based on the difference between a predetermined threshold value for spectral anomaly characteristics based on the compound structure type and the actual spectral anomaly characteristics. The risk assessment module is connected to the analysis module and the spectrum module respectively. It is used to determine the risk tendency type of the spectrum structure based on the key drug characteristic factors of the compound structure type, and to determine the activation verification module based on the risk tendency type of the spectrum structure. The verification module, which is connected to the risk assessment module, is used to determine the reasons for the system's non-compliance with standards and the handling methods based on the difference between the drug's key characteristic factors and the predetermined thresholds for the drug's key characteristic factors: determining the magnitude of the decrease in the threshold of the abnormal feature characterization value of the spectrum, or determining the magnitude of the increase in the threshold of the drug's key characteristic factors. The key feature data information includes: the number of drug entity nodes and the number of drug association edges; the graph anomaly feature parameter information includes drug duplication rate and drug error rate. The key characteristic factors of the drug are determined based on the sum of the first key characteristic factor and the second key characteristic factor, wherein: The first key feature factor is determined based on the ratio of the number of drug entity nodes to a predetermined threshold for the number of drug entity nodes. The second key feature factor is determined based on the ratio of the number of drug-related edges to a predetermined threshold for the number of drug-related edges; The anomaly feature characterization value of the map is determined based on the sum of the first anomaly feature factor and the second anomaly feature factor, wherein: The first abnormal feature factor is determined based on the ratio of a predetermined drug repetition rate threshold to the actual drug repetition rate. The second abnormal feature factor is determined based on the ratio of a predetermined drug error rate threshold to the actual drug error rate. 2.The knowledge graph construction system for a drug database of claim 1, wherein, The number of drug entity nodes is determined based on the number of target gene information, the number of disease associations, the number of side effect records, and the number of clinical trials. 3.The knowledge graph construction system for a drug database of claim 2, wherein, The number of drug-associated edges is determined based on the number of drug compounds binding to targets, the number of diseases treated by the drug, and the number of side effects caused by the drug. 4.The knowledge graph construction system for a drug database of claim 1, wherein, The system operation does not meet the standard when the difference between the predetermined threshold value of the abnormal spectral feature characterization value of the target compound structure type and the actual abnormal spectral feature characterization value is greater than the predetermined difference threshold value. 5.The knowledge graph construction system for a drug database of claim 1, wherein, The criteria for determining the risk tendency type of the spectral structure are compound structure types whose key drug characteristic factors are greater than a predetermined threshold for key drug characteristic factors. 6.The knowledge graph construction system for a drug database of claim 4, wherein, The reasons why the system does not meet the standards include a first reason label and a second reason label, wherein: If the difference between the key drug feature factor and the predetermined key drug feature factor threshold is less than or equal to the predetermined difference threshold, it is determined as the first cause label and the first processing method is executed. If the difference between the drug key characteristic factor and the predetermined drug key characteristic factor threshold value is greater than the predetermined difference threshold value, a second cause label is determined, and a second processing mode is executed. 7.The knowledge graph construction system for a drug database of claim 6, wherein, The processing mode includes a first processing mode and a second processing mode, wherein: The first processing mode is to determine a reduction range of the atlas abnormal characteristic feature value threshold value based on the difference between the drug key characteristic factor and the predetermined drug key characteristic factor threshold value; The second processing mode is to determine an increase range of the drug key characteristic factor threshold value. 8.The knowledge graph construction system for a drug database of claim 1, wherein, The system running condition conforming to the standard is that the difference between the predetermined atlas abnormal characteristic feature value threshold value of the target compound structure type and the actual atlas abnormal characteristic feature value is less than or equal to the predetermined difference threshold value.

Images