Drug interaction intelligent prediction method and system based on knowledge graph

Abstract

The invention provides an intelligent drug interaction prediction method and system based on a knowledge graph, and relates to the technical field of medical information, and the method comprises the steps: extracting drug molecular structure and metabolic data, calculating the conformation overlapping degree, determining a metabolic pathway overlapping site, calculating the competitive binding probability, analyzing the drug distribution state, and judging a metabolic cycle overlapping interval. Determining a concentration accumulation range, evaluating a metabolic inhibition level, generating early warning data and updating the knowledge graph. The drug interaction can be accurately predicted, the medication safety is improved, and the adverse reaction risk is reduced.

Description

Technical Field

[0001] The present invention relates to medical information technology, and in particular to a method and system for intelligently predicting drug interactions based on a knowledge graph. Background Art

[0002] With the advancement of medical technology, the use of multiple drugs in clinical practice is increasing, and the problem of adverse reactions caused by drug interactions is becoming increasingly prominent. Currently, drug interactions are mainly predicted based on physician experience or by consulting drug instructions. This method suffers from poor real-time performance and low accuracy, making it difficult to adapt to the complexity of clinical drug use.

[0003] Existing drug interaction prediction technologies are primarily based on statistical analysis and machine learning methods, which use mathematical models to predict drug interactions. However, these methods often overlook the influence of drug molecular structure and metabolic mechanisms, have limited predictive power for novel drug combinations, and lack interpretability of prediction results, making it difficult to provide effective guidance for clinical decision-making.

[0004] While knowledge graph technology has been applied in pharmaceutical research, existing technologies have yet to establish an intelligent prediction system based on drug molecular structure and metabolic mechanisms, and are unable to effectively integrate multi-dimensional information such as drug spatial conformation, metabolic sites, and metabolic kinetics. Therefore, a technical approach that can combine drug knowledge graphs and metabolic mechanisms to achieve intelligent prediction of drug interactions is urgently needed. Summary of the Invention

[0005] The embodiments of the present invention provide a method and system for intelligent prediction of drug interactions based on knowledge graphs, which can solve the problems in the prior art.

[0006] A first aspect of an embodiment of the present invention provides a method for intelligently predicting drug interactions based on a knowledge graph, comprising:

[0007] Extract the molecular structure data and metabolic data of the drug to be tested from the drug knowledge graph, calculate the spatial conformation overlap of the drug to be tested, determine the overlapping sites of the metabolic pathway through graph reasoning, and generate a metabolic site distribution map;

[0008] Based on the metabolic site distribution map, the competitive binding probability of the drug to be tested at the coincident site is calculated. The coincident site inhibition coefficient is calculated based on the competitive binding probability. The drug distribution state under different inhibition coefficients is analyzed using the drug diffusion calculation method, and a time series curve reflecting the dynamic changes of drug distribution is generated.

[0009] Determine the overlapping interval of the metabolic cycle of the drug to be tested based on the time series curve, determine the concentration accumulation range of the drug to be tested based on the overlapping interval of the metabolic cycle, calculate the interaction coefficient of the drug to be tested based on the concentration accumulation range, and obtain the blood drug concentration change trend of the drug to be tested based on the interaction coefficient;

[0010] Determine the metabolic inhibition level of the drug to be tested based on the trend of changes in blood drug concentration, compare the metabolic inhibition level with the safety threshold preset in the drug knowledge graph, obtain the risk assessment result of the drug to be tested, and generate metabolic abnormality warning data for the drug to be tested based on the risk assessment result;

[0011] The metabolic abnormality warning data is written into the drug knowledge graph, and the interaction feature records of the drugs to be tested in the drug knowledge graph are updated to form new metabolic reference data for the drugs to be tested.

[0012] In an optional embodiment,

[0013] Extract the molecular structure data and metabolic data of the drug to be tested from the drug knowledge graph, calculate the spatial conformation overlap of the drug to be tested, determine the overlapping sites of the metabolic pathway through graph reasoning, and generate a metabolic site distribution map including:

[0014] Constructing a drug knowledge graph, setting molecular structure data as graph node attributes, setting metabolic data as graph edge relationships, using a recursive query method to identify the molecular structure data of the drug node to be tested in the drug knowledge graph, traversing the graph edge relationships to obtain metabolic data, and combining the obtained molecular structure data and metabolic data into a drug feature set;

[0015] Identifying structural matching fragments of the drug to be tested based on the molecular structure data in the drug feature set, spatially superimposing the drug to be tested using the structural matching fragments, and calculating the spatial conformational overlap of the drug to be tested;

[0016] Analyzing the metabolic reaction pathways of the drug knowledge graph based on the metabolic data in the drug feature set, identifying target nodes with common metabolic features through graph reasoning, screening the target nodes based on the spatial conformational overlap to obtain candidate overlapping sites, calculating the spatial distribution parameters of the candidate overlapping sites based on the spatial conformational overlap, and determining the candidate overlapping sites whose spatial distribution parameters exceed a preset distribution threshold as metabolic pathway overlapping sites;

[0017] The spatial distribution parameters of the coincident sites of the metabolic pathway and the metabolic data in the drug feature set are integrated to generate a metabolic site distribution map.

[0018] In an optional embodiment,

[0019] The metabolic reaction pathways of the drug knowledge graph are analyzed based on the metabolic data in the drug feature set. Target nodes with common metabolic features are identified through graph reasoning. Target nodes are screened based on the spatial conformational overlap, and candidate overlap sites are obtained, including:

[0020] Constructing a metabolic reaction network based on the metabolic data in the drug feature set, setting the metabolic reactions as directed edges, setting the reactants and products as network nodes, and calculating the reaction weight values ​​of the directed edges based on the substrate affinity and catalytic efficiency of the metabolic enzymes;

[0021] Performing a random walk search on the metabolic reaction network, setting the drug node to be tested as the starting point, determining the node transfer probability according to the reaction weight value, introducing a path length attenuation factor to perform path search, and obtaining a node access sequence;

[0022] Building a metabolic pattern feature library based on the node access sequence, setting substrate-enzyme-product combinations and metabolic intermediate sharing relationships as pattern features, matching pattern features in the metabolic reaction network, and identifying target nodes with common metabolic features;

[0023] constructing a spatial scoring function using the obtained spatial conformation overlap, and calculating the spatial compatibility value of the target node based on the spatial scoring function;

[0024] Calculate the similarity of the metabolic characteristics of the target node and the drug to be tested, construct a multi-objective optimization function, take the metabolic characteristic similarity as the first optimization goal, and take the spatial compatibility value as the second optimization goal. Calculate the comprehensive score of the target node according to the multi-objective optimization function, and select the atlas nodes whose comprehensive score exceeds the preset threshold as candidate coincidence sites.

[0025] In an optional embodiment,

[0026] Based on the metabolic site distribution map, the competitive binding probability of the drug to be tested at the coincident site is calculated. The coincident site inhibition coefficient is calculated based on the competitive binding probability. The drug diffusion calculation method is used to analyze the drug distribution state under different inhibition coefficients. The time series curve reflecting the dynamic changes of drug distribution is generated, including:

[0027] Extract the spatial structure data and chemical environment data of the overlapping sites from the metabolic site distribution map, use molecular dynamics to simulate the conformational changes of the drug to be tested, and obtain the binding energy value between the drug to be tested and the overlapping sites;

[0028] Calculating the initial competitive binding probability of the test drug at each overlapping site based on the binding energy value, determining the maximum binding capacity of each overlapping site, and establishing a mass action equation;

[0029] Substituting the concentration data of the drug to be tested into the mass action equation, calculating the actual competitive binding probability of the coincident site, and determining the ratio of the actual competitive binding probability to the initial competitive binding probability as the coincident site inhibition coefficient;

[0030] The coincidence site inhibition coefficient is used as the diffusion resistance factor to calculate the diffusion rate of the drug to be tested in the tissue space, and the drug concentration value of each area is obtained. The calculation time interval is dynamically adjusted according to the change amplitude of the drug concentration value, and an iterative operation is used to obtain the drug distribution data of multiple time points. The drug distribution data is converted into a time series to generate a drug distribution dynamic change curve.

[0031] In an optional embodiment,

[0032] Determine the overlapping interval of the metabolic cycle of the drug to be tested based on the time series curve, determine the concentration accumulation range of the drug to be tested based on the overlapping interval of the metabolic cycle, calculate the interaction coefficient of the drug to be tested based on the concentration accumulation range, and obtain the blood drug concentration change trend of the drug to be tested based on the interaction coefficient, including:

[0033] Scanning the time series curve using a sliding time window, extracting normalized concentration values ​​of the drug to be tested at different time points according to the time series curve, and constructing a concentration data sequence of the drug to be tested according to the normalized concentration values ​​in a time series;

[0034] Acquiring half-life characteristics of the drug to be tested based on the concentration data sequence, constructing an adaptive threshold judgment rule according to the half-life characteristics, performing segmented comparison on the concentration data sequence, and determining an overlapping interval of the metabolic cycle of the drug to be tested;

[0035] Calculating a concentration vector of the drug to be tested within an overlapping interval of the metabolic cycle, constructing a multidimensional concentration accumulation tensor based on the concentration vector, performing a boundary integral operation on the concentration accumulation tensor to obtain a cumulative volume, and determining a concentration accumulation range of the drug to be tested based on the cumulative volume;

[0036] The cumulative effect intensity of the drug to be tested in the overlapping interval of the metabolic cycle is calculated according to the concentration accumulation range, the interaction coefficient of the drug to be tested is calculated based on the cumulative effect intensity, the concentration distribution characteristics of the drug to be tested are analyzed according to the interaction coefficient, the concentration distribution characteristics are fitted and verified with the measured data, the fitting error is calculated, and the adaptive threshold judgment rule is dynamically corrected according to the fitting error to obtain the blood drug concentration change trend of the drug to be tested.

[0037] In an optional embodiment,

[0038] Calculating the concentration vector of the drug to be tested within the overlapping interval of the metabolic cycle, constructing a multidimensional concentration accumulation tensor based on the concentration vector, performing a boundary integral operation on the concentration accumulation tensor to obtain a cumulative volume, and determining the concentration accumulation range of the drug to be tested based on the cumulative volume includes:

[0039] Establishing a three-dimensional coordinate system in the overlapping intervals of the metabolic cycle, the three-dimensional coordinate system including a spatial position axis, a metabolic pathway axis, and a time series axis, obtaining a normalized concentration value at each coordinate point, and constructing a drug concentration vector;

[0040] Analyzing the activity intensity of the metabolic enzyme to obtain the substrate affinity between the metabolic enzyme and the drug molecule, calculating the metabolic pathway weight factor based on the activity intensity and substrate affinity, and performing a tensor operation on the metabolic pathway weight factor and the drug concentration vector to generate a drug accumulation tensor;

[0041] Determining a main characteristic direction of the drug cumulative tensor, slicing the drug cumulative tensor along the main characteristic direction, constructing a computational grid based on a concentration gradient distribution of the drug cumulative tensor, and performing adaptive partitioning;

[0042] A piecewise integral operation is performed on the drug cumulative tensor within the divided computational grid to obtain the local cumulative value of each grid. The local cumulative values ​​are summarized to obtain the cumulative volume of the drug. The spatial distribution characteristics of the drug in the overlapping intervals of the metabolic cycle are analyzed based on the cumulative volume, and the concentration accumulation range of the drug is determined in combination with the spatial distribution characteristics.

[0043] In an optional embodiment,

[0044] The metabolic inhibition level of the drug to be tested is determined based on the trend of changes in blood drug concentration. The metabolic inhibition level is compared with the safety threshold preset in the drug knowledge graph to obtain the risk assessment result of the drug to be tested. Based on the risk assessment result, metabolic abnormality warning data of the drug to be tested is generated, including:

[0045] Performing time-frequency decomposition on the blood drug concentration change trend, extracting the frequency characteristics and amplitude characteristics of the blood drug concentration change trend, identifying abnormal fluctuation points of the blood drug concentration based on the frequency characteristics and amplitude characteristics, and generating a blood drug concentration fluctuation characteristic map;

[0046] Acquiring dynamic change data of metabolic enzyme activity, constructing a metabolic activity distribution matrix based on the dynamic change data of metabolic enzyme activity, calculating the eigenvalue change rate of the metabolic activity distribution matrix, and determining the metabolic inhibition level of the drug to be tested according to the eigenvalue change rate;

[0047] Obtaining metabolic pathway information and metabolic enzyme-related data of the drug to be tested from the drug knowledge graph, calculating the safety threshold range, and comparing the metabolic inhibition level with the safety threshold range to obtain a risk assessment result of the drug to be tested;

[0048] The risk assessment result is associated with the blood drug concentration fluctuation characteristic map, the metabolic abnormality type of the drug to be tested is determined according to the association analysis result, and metabolic abnormality warning data of the drug to be tested is generated according to the metabolic abnormality type.

[0049] A second aspect of an embodiment of the present invention provides a drug interaction intelligent prediction system based on a knowledge graph, comprising:

[0050] The first unit is used to extract the molecular structure data and metabolic data of the drug to be tested from the drug knowledge graph, calculate the spatial conformation overlap of the drug to be tested, determine the overlapping sites of the metabolic pathway through graph reasoning, and generate a metabolic site distribution map;

[0051] The second unit is used to calculate the competitive binding probability of the drug to be tested at the coincident site based on the metabolic site distribution map, calculate the coincident site inhibition coefficient based on the competitive binding probability, analyze the drug distribution state under different inhibition coefficients using the drug diffusion calculation method, and generate a time series curve reflecting the dynamic changes of drug distribution;

[0052] The third unit is used to determine the overlapping interval of the metabolic cycle of the drug to be tested based on the time series curve, determine the concentration accumulation range of the drug to be tested based on the overlapping interval of the metabolic cycle, calculate the interaction coefficient of the drug to be tested based on the concentration accumulation range, and obtain the blood drug concentration change trend of the drug to be tested based on the interaction coefficient;

[0053] The fourth unit is used to determine the metabolic inhibition level of the drug to be tested based on the trend of changes in blood drug concentration, compare the metabolic inhibition level with the safety threshold preset in the drug knowledge graph, obtain the risk assessment result of the drug to be tested, and generate metabolic abnormality warning data for the drug to be tested based on the risk assessment result;

[0054] The fifth unit is used to write metabolic abnormality warning data into the drug knowledge graph, update the interaction feature records of the drugs to be tested in the drug knowledge graph, and form new metabolic reference data for the drugs to be tested.

[0055] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0056] processor;

[0057] a memory for storing processor-executable instructions;

[0058] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0059] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0060] In this embodiment, by analyzing the molecular structure and metabolic data of drugs, calculating the spatial conformation overlap and the probability of competitive binding of metabolic sites, accurate prediction of drug interactions is achieved, the accuracy of drug interaction risk assessment is improved, and the incidence of adverse reactions in clinical medication is reduced. A drug metabolism dynamic analysis model based on knowledge graph is established. By calculating the time series curve and concentration accumulation range, it can dynamically monitor the metabolic process of drugs in the body, provide doctors with real-time medication reference, and effectively improve the safety and effectiveness of clinical medication. The automatic update and knowledge accumulation of drug interaction data are realized. By writing the metabolic abnormality warning data back into the knowledge graph, a closed-loop management system for drug metabolism is formed, and the drug interaction prediction model is continuously optimized, which improves the system's self-learning ability and prediction accuracy, providing strong support for precision medicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 Schematic diagram of the process of the intelligent prediction method of drug interactions based on knowledge graph according to an embodiment of the present invention;

[0062] Figure 2 This is a scatter plot of spatial conformation overlap analysis of drugs with different structures according to the present invention;

[0063] Figure 3 This is a comparison chart of the normalization processing effects of drug concentration in the sliding time window of an embodiment of the present invention. DETAILED DESCRIPTION

[0064] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0065] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0066] Figure 1 FIG. 1 is a flow chart of an intelligent prediction method for drug interactions based on a knowledge graph according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0067] Extract the molecular structure data and metabolic data of the drug to be tested from the drug knowledge graph, calculate the spatial conformation overlap of the drug to be tested, determine the overlapping sites of the metabolic pathway through graph reasoning, and generate a metabolic site distribution map;

[0068] Based on the metabolic site distribution map, the competitive binding probability of the drug to be tested at the coincident site is calculated. The coincident site inhibition coefficient is calculated based on the competitive binding probability. The drug distribution state under different inhibition coefficients is analyzed using the drug diffusion calculation method, and a time series curve reflecting the dynamic changes of drug distribution is generated.

[0069] Determine the overlapping interval of the metabolic cycle of the drug to be tested based on the time series curve, determine the concentration accumulation range of the drug to be tested based on the overlapping interval of the metabolic cycle, calculate the interaction coefficient of the drug to be tested based on the concentration accumulation range, and obtain the blood drug concentration change trend of the drug to be tested based on the interaction coefficient;

[0070] Determine the metabolic inhibition level of the drug to be tested based on the trend of changes in blood drug concentration, compare the metabolic inhibition level with the safety threshold preset in the drug knowledge graph, obtain the risk assessment result of the drug to be tested, and generate metabolic abnormality warning data for the drug to be tested based on the risk assessment result;

[0071] The metabolic abnormality warning data is written into the drug knowledge graph, and the interaction feature records of the drugs to be tested in the drug knowledge graph are updated to form new metabolic reference data for the drugs to be tested.

[0072] In an optional embodiment, extracting molecular structure data and metabolic data of the drug to be tested from the drug knowledge graph, calculating the spatial conformational overlap of the drug to be tested, determining the overlapping sites of the metabolic pathway through graph reasoning, and generating a metabolic site distribution map includes:

[0073] Constructing a drug knowledge graph, setting molecular structure data as graph node attributes, setting metabolic data as graph edge relationships, using a recursive query method to identify the molecular structure data of the drug node to be tested in the drug knowledge graph, traversing the graph edge relationships to obtain metabolic data, and combining the obtained molecular structure data and metabolic data into a drug feature set;

[0074] Identifying structural matching fragments of the drug to be tested based on the molecular structure data in the drug feature set, spatially superimposing the drug to be tested using the structural matching fragments, and calculating the spatial conformational overlap of the drug to be tested;

[0075] Analyzing the metabolic reaction pathways of the drug knowledge graph based on the metabolic data in the drug feature set, identifying target nodes with common metabolic features through graph reasoning, screening the target nodes based on the spatial conformational overlap to obtain candidate overlapping sites, calculating the spatial distribution parameters of the candidate overlapping sites based on the spatial conformational overlap, and determining the candidate overlapping sites whose spatial distribution parameters exceed a preset distribution threshold as metabolic pathway overlapping sites;

[0076] The spatial distribution parameters of the coincident sites of the metabolic pathway and the metabolic data in the drug feature set are integrated to generate a metabolic site distribution map.

[0077] This embodiment provides a method for extracting drug molecular structure data and metabolic data from a drug knowledge graph, calculating the overlap of drug spatial conformations, and determining the overlapping sites of metabolic pathways to generate a metabolic site distribution map.

[0078] In the process of constructing a drug knowledge graph, an RDF triple model is used to map drug nodes to their attributes and relationships. Each drug node contains molecular structure data attributes, such as SMILES strings, molecular weight, and number of functional groups. Graph edges contain metabolic data, such as the type of metabolizing enzyme, metabolic rate, and metabolites. For example, for drug A, its node attributes might be recorded as: "Drug A has the structure SMILES: CC(C)NCC(O)COc1cccc2ccccc12," and its edge relationships might be recorded as: "Drug A is metabolized by the metabolizing enzyme CYP2D6" and "Drug A has a metabolite, metabolite A."

[0079] To identify the molecular structure data of the drug node under test, a recursive query method using the Structured Query Language (SQL) was employed. Taking the drug "phenytoin sodium" as an example, the recursive query statement can be constructed as follows: starting with the drug node, with a recursive depth of 3, all directly connected structural attribute nodes are retrieved. The complete molecular structure data is then retrieved through relationships such as "has attribute" and "has structure." The query results will return the molecular structure attributes of phenytoin sodium, including the concise linear representation of the molecular input "O=C1NC(=O)C(c2ccccc2)(c2ccccc2)N1," a molecular weight of 252.27, and three ring structures.

[0080] When traversing the metabolic data, we extract information about phenytoin sodium metabolizing enzymes (such as CYP2C9), metabolic sites (such as C4 and C5), and metabolite products (such as 4-hydroxyphenytoin) along the edges of the "has metabolic pathway" and "is metabolized" relationships. By combining the acquired molecular structure data with the metabolic data, we form a complete drug feature set, including detailed molecular characterization information and metabolic characteristics.

[0081] When identifying structurally matching fragments based on a drug feature set, the drug under test is structurally compared to drugs with known metabolic mechanisms in the knowledge graph. Molecular fingerprinting techniques, such as extended ring structure fingerprints or molecular access control fragment keys, are used to calculate molecular similarity, with a similarity threshold of 0.7. For example, the comparison revealed similarities between phenytoin sodium and carbamazepine in their five-membered ring structure. Both contain double benzene ring substituents, confirming the diphenylamide structure as a matching fragment.

[0082] A flexible molecular alignment algorithm is used to spatially align the test drugs. First, a stable conformation of phenytoin sodium is obtained through energy minimization. Subsequently, based on the identified matching fragments, phenytoin sodium is three-dimensionally aligned with known drugs with similar structures (such as carbamazepine). During the alignment process, the root mean square deviation of the matching fragments is minimized by rotating dihedral angles and optimizing atomic positions. In this example, the root mean square deviation of the double benzene ring of phenytoin sodium and carbamazepine after alignment is 1.2, indicating a good spatial structural match.

[0083] The spatial conformational overlap calculated using the molecular volume overlap coefficient method resulted in a calculated spatial volume overlap of 0.82 for phenytoin sodium and carbamazepine, exceeding the preset threshold of 0.75, indicating a high degree of conformational similarity between the two. Furthermore, the calculated electron cloud distribution similarity was 0.78, demonstrating a high degree of consistency in electronic properties, which is of great significance for predicting metabolic sites.

[0084] When analyzing metabolic reaction pathways, a metabolic network was constructed based on the metabolic information in the graph. Starting from the phenytoin sodium node, a recursive search for edges with metabolic reactions yielded the complete metabolic pathway: phenytoin sodium → CYP2C9 catalysis → 4-hydroxylation → 4-hydroxyphenytoin product. Graph reasoning revealed that carbamazepine also generates the 10,11-epoxide catalyzed by CYP3A4, indicating a similar aromatic ring hydroxylation metabolic mechanism between the two drugs.

[0085] During the screening of target nodes, a combined assessment of spatial conformational overlap and metabolic enzyme overlap was performed. Phenytoin and carbamazepine share a conformational overlap of 0.82, and share CYP2C9 and CYP3A4 as metabolizing enzymes, with an enzyme overlap of 0.6. Based on these criteria, four candidate overlapping sites were identified: the C4 and C5 positions of phenytoin, and the C2 and C3 positions of carbamazepine. Spatial distribution parameters for these sites were further calculated, including distance from the molecular center, surface exposure, and electrostatic potential. Specifically, the C4 site of phenytoin exhibited a surface exposure of 87% and an electrostatic potential of -8.2 kilojoules per mole; the C5 site exhibited a surface exposure of 72% and an electrostatic potential of -7.5 kilojoules per mole. Both exceeded the established distribution thresholds (surface exposure of at least 60% and electrostatic potential of no greater than -5 kilojoules per mole), thus confirming them as metabolic pathway overlap sites.

[0086] When generating the metabolic site distribution map, a heat map was used to integrate spatial distribution parameters and metabolic data. The surface of the molecular structure model of phenytoin sodium was marked with a color gradient to indicate metabolic activity: red areas indicated the C4 site (metabolism probability 0.85), and orange areas indicated the C5 site (metabolism probability 0.65). The directions of action of the major metabolizing enzymes CYP2C9 and CYP3A4 were also annotated, with arrows pointing to the corresponding sites. The heat map clearly displays the metabolic hotspots on the drug molecule surface, providing a visual reference for drug metabolism research and drug-drug interaction analysis.

[0087] In this example, by constructing a drug knowledge graph based on the RDF triple model and combining it with recursive queries, molecular fingerprint comparison, flexible molecular alignment, and spatial conformation analysis using structured query language, high-precision identification and matching of the molecular structure and metabolic characteristics of the drug under test was achieved. This method can effectively explore the intrinsic relationship between drug structure and metabolic mechanism, accurately predict potential metabolic sites and their enzymatic pathways, and improve the sensitivity and visualization level of metabolic reaction identification. Compared with traditional two-dimensional structure analysis methods, this solution performs better in terms of spatial overlap and electronic distribution consistency judgment, improving the accuracy and reliability of metabolic prediction.

[0088] Figure 2 This is a scatter plot of spatial conformation overlap analysis of drugs with different structures according to the present invention. Figure 2 As shown in the figure, the analysis results of the spatial conformational overlap of drugs with different structural complexities are shown, and the performance difference between the present technical solution and the traditional molecular docking method is compared. The horizontal axis represents the molecular complexity index (range 0-1000), and the vertical axis represents the percentage of spatial conformational overlap (0-100%). A 70% overlap threshold line is set in the figure as a criterion for judging effective overlap. The spatial conformational overlap data points of the present technical solution (represented by rectangular marks) are distributed between 82.4% and 93.8%, all above the threshold line, showing high stability. The specific data points are: 82.4% at a molecular complexity of 120, 88.5% at 190, 84.2% at 260, 92.6% at 330, 89.1% at 400, 93.8% at 470, 87.9% at 540, 91.5% at 610, and 83.7% at 680. In contrast, the overlap data points for the traditional molecular docking method (indicated by circular markers) ranged from 39.8% to 54.9%, all below the threshold, and fluctuated significantly across different complexity levels. Comparison of the two methods reveals that the proposed method achieves higher conformational overlap than the traditional method at all complexity levels, demonstrating its significant advantage in spatial alignment of drug molecules.

[0089] In an optional embodiment, the metabolic reaction pathways of the drug knowledge graph are analyzed based on the metabolic data in the drug feature set, target nodes with common metabolic features are identified through graph reasoning, and the target nodes are screened based on the spatial conformational overlap, and the candidate overlapping sites obtained include:

[0090] Constructing a metabolic reaction network based on the metabolic data in the drug feature set, setting the metabolic reactions as directed edges, setting the reactants and products as network nodes, and calculating the reaction weight values ​​of the directed edges based on the substrate affinity and catalytic efficiency of the metabolic enzymes;

[0091] Performing a random walk search on the metabolic reaction network, setting the drug node to be tested as the starting point, determining the node transfer probability according to the reaction weight value, introducing a path length attenuation factor to perform path search, and obtaining a node access sequence;

[0092] Building a metabolic pattern feature library based on the node access sequence, setting substrate-enzyme-product combinations and metabolic intermediate sharing relationships as pattern features, matching pattern features in the metabolic reaction network, and identifying target nodes with common metabolic features;

[0093] constructing a spatial scoring function using the obtained spatial conformation overlap, and calculating the spatial compatibility value of the target node based on the spatial scoring function;

[0094] Calculate the similarity of the metabolic characteristics of the target node and the drug to be tested, construct a multi-objective optimization function, take the metabolic characteristic similarity as the first optimization goal, and take the spatial compatibility value as the second optimization goal. Calculate the comprehensive score of the target node according to the multi-objective optimization function, and select the atlas nodes whose comprehensive score exceeds the preset threshold as candidate coincidence sites.

[0095] In this embodiment, the metabolic reaction pathways of the drug knowledge graph are analyzed based on the metabolic data in the drug feature set, target nodes with common metabolic features are identified through graph reasoning, and the target nodes are screened based on the spatial conformational overlap to obtain candidate overlapping sites.

[0096] When constructing a metabolic reaction network based on metabolic data from a drug feature set, metabolic reactions are set as directed edges, and reactants and products are set as network nodes. For each metabolic reaction, the reaction weight of the directed edge is calculated based on the substrate affinity and catalytic efficiency of the metabolic enzyme. For example, for a reaction R catalyzed by enzyme E, in which substrate S is converted to product P, the reaction weight can be determined by the enzyme's affinity constant for the substrate, Km, and the maximum catalytic rate, Vmax. In practical applications, assuming that for the drug methyldopamine, which is metabolized by monoamine oxidase (MAO), the calculated reaction weight is 1.67 when the Km value is 15 μM and the Vmax is 25 nmol / min / mg. The higher the reaction weight, the more important the metabolic pathway.

[0097] During the random walk search of the metabolic reaction network, the drug node to be tested is set as the starting point, and the node transfer probability is determined according to the reaction weight value. In each step of transfer, the probability of moving from the current node to the adjacent node is proportional to the reaction weight value of the connecting edge. At the same time, a path length attenuation factor is introduced to control the search depth and avoid information dilution caused by excessively long paths. For example, setting the attenuation factor to 0.85 means that as the path extends, the importance of the node decreases by 0.85 times. Taking atorvastatin as an example, starting from its starting node, the node access sequence obtained after 3 steps of random walk includes: atorvastatin → hydroxyatorvastatin → hydroxyatorvastatin glucuronide conjugate → final metabolite. Through multiple random walks, a representative set of node access sequences is obtained.

[0098] When constructing a metabolic pattern feature library based on node access sequences, substrate-enzyme-product combinations and shared metabolic intermediate relationships are set as pattern features. For example, patterns such as "substrate S is metabolized by enzyme E1 to intermediate I, and intermediate I is metabolized by enzyme E2 to product P" are extracted. In practical applications, for simvastatin, extracted pattern features include "simvastatin-CYP3A4-hydroxysimvastatin-UDP-glucuronosyltransferase-hydroxysimvastatin glucuronic acid conjugate." When matching pattern features within a metabolic reaction network, a subgraph matching algorithm is used to identify target nodes with shared metabolic features. If two drug nodes share similar metabolic patterns, such as being metabolized by CYP3A4 and producing similar metabolic intermediates, they are identified as target nodes with shared metabolic features.

[0099] When constructing a spatial scoring function using the obtained spatial conformational overlap, the three-dimensional structural similarity of the molecules is taken into account. The spatial scoring function is based on factors such as molecular shape similarity, matching of key functional group positions, and charge distribution similarity. For example, by analyzing the molecular structural overlap of rosiglitazone and pioglitazone, a spatial overlap of 0.78 was calculated. The spatial compatibility value of the target node is calculated based on the spatial scoring function. The higher the spatial compatibility value, the more similar the two molecules are in three-dimensional space and the more likely they are to share the same site of action.

[0100] When calculating the similarity of the metabolic characteristics of the target node and the drug to be tested, a similarity calculation model is constructed based on dimensions such as metabolic pathway pattern, degree of sharing of metabolic enzymes, and similarity of intermediate metabolites. For example, for two statins, if they share 80% of metabolic enzymes and 75% of metabolic intermediates, their metabolic characteristic similarity may reach 0.77. When constructing a multi-objective optimization function, the metabolic characteristic similarity is used as the first optimization target, and the spatial compatibility value is used as the second optimization target. A weight coefficient is set between the two targets, for example, the metabolic characteristic similarity weight is 0.6, and the spatial compatibility value weight is 0.4. The comprehensive score of the target node is obtained by weighted calculation. The preset threshold is set to 0.75, and the map nodes with a comprehensive score exceeding the threshold are selected as candidate overlap sites.

[0101] In a specific application scenario, the antihypertensive drug losartan was analyzed. The aforementioned method identified irbesartan, which has a metabolic similarity of 0.82 and a spatial compatibility of 0.78. The calculated overall score was 0.804, exceeding the preset threshold of 0.75. Therefore, irbesartan was selected as a candidate overlap site for losartan. Further experimental verification showed that the two drugs indeed have similar metabolic behaviors and targets in the human body, demonstrating the effectiveness of this method.

[0102] In this embodiment, by constructing a metabolic reaction network and introducing a random walk search and path attenuation mechanism, dynamic mining and characteristic pattern recognition of complex metabolic pathways are achieved, and target nodes with common metabolic characteristics with the drug to be tested can be accurately extracted. A spatial scoring function is constructed in combination with spatial conformational overlap, and a multi-objective optimization method is introduced to comprehensively consider metabolic similarity and spatial compatibility to achieve accurate screening of candidate overlapping sites. This solution improves the sensitivity and specificity of metabolic mechanism matching, effectively overcomes the adaptation barriers of traditional methods to drugs with large structural differences, and provides a more discriminative screening method for new drug metabolism prediction and drug interaction analysis.

[0103] In an optional embodiment, the competitive binding probability of the drug to be tested at the coincident site is calculated based on the metabolic site distribution map, the coincident site inhibition coefficient is calculated based on the competitive binding probability, and the drug distribution state under different inhibition coefficients is analyzed using a drug diffusion calculation method. The generation of a time series curve reflecting the dynamic change of drug distribution includes:

[0104] Extract the spatial structure data and chemical environment data of the overlapping sites from the metabolic site distribution map, use molecular dynamics to simulate the conformational changes of the drug to be tested, and obtain the binding energy value between the drug to be tested and the overlapping sites;

[0105] Calculating the initial competitive binding probability of the test drug at each overlapping site based on the binding energy value, determining the maximum binding capacity of each overlapping site, and establishing a mass action equation;

[0106] Substituting the concentration data of the drug to be tested into the mass action equation, calculating the actual competitive binding probability of the coincident site, and determining the ratio of the actual competitive binding probability to the initial competitive binding probability as the coincident site inhibition coefficient;

[0107] The coincidence site inhibition coefficient is used as the diffusion resistance factor to calculate the diffusion rate of the drug to be tested in the tissue space, and the drug concentration value of each area is obtained. The calculation time interval is dynamically adjusted according to the change amplitude of the drug concentration value, and an iterative operation is used to obtain the drug distribution data of multiple time points. The drug distribution data is converted into a time series to generate a drug distribution dynamic change curve.

[0108] This embodiment provides a drug competition binding analysis method based on a metabolic site distribution map. The method calculates the competitive binding probability and inhibition coefficient of drugs at overlapping sites, thereby analyzing the dynamic changes in drug distribution.

[0109] To extract coincident site data from a metabolic site map, the map is first digitized, and high-resolution microscopy is used to capture the three-dimensional spatial structure of the tissue sample. For liver tissue, three-dimensional spatial data is acquired by scanning with a laser confocal microscope at a resolution of 0.5 μm, forming a coincident site distribution data matrix containing spatial coordinates (x, y, z). For each coincident site, its coordinate position, geometric parameters (such as pocket depth and volume), and local chemical environment data (such as pH, hydrophobic region distribution, and charged amino acid residue distribution) are recorded. For the CYP3A4 enzyme site, the active pocket depth of 1.2 × 10 -9 m, with a volume of approximately 1.2 nm 3 There is a hydrophobic region around it, accounting for 65%, 5 negatively charged amino acid residues and 3 positively charged amino acid residues.

[0110] During molecular dynamics simulations of the test drug, a force field parameterization method is used to construct a molecular model of the drug. Taking atorvastatin as an example, its standard chemical structure is first obtained. The molecule is parameterized using the AMBER force field. The simulation temperature is set to 310K and the pressure is 1atm. Molecular dynamics simulations are performed in an aqueous environment for 100ns with a sampling interval of 10ps, resulting in a total of 10,000 conformations. Cluster analysis is used to extract a representative set of conformations, generally selecting the top five cluster center conformations that cover more than 90% of the conformational space. These conformations are then molecularly docked with the previously extracted overlapping site structures, and the binding free energy is calculated. The docking results for atorvastatin at the CYP3A4 site show a binding free energy of -8.5 kcal / mol, with two hydrogen bonds and multiple hydrophobic interactions.

[0111] When calculating competitive binding probabilities based on binding energy values, all overlapping sites are ranked according to their binding affinity for the drug under test. For example, the binding free energies for atorvastatin to the three major liver metabolizing enzymes, CYP3A4, CYP2C9, and CYP2D6, are -8.5, -6.2, and -5.1 kcal / mol, respectively. Initial competitive binding probabilities were calculated using the Boltzmann distribution principle, showing initial binding probabilities of 75%, 20%, and 5% for atorvastatin at these three sites, respectively. The maximum binding capacity for each site is determined based on the enzyme's expression level. For example, the expression levels of CYP3A4, CYP2C9, and CYP2D6 in hepatocytes are approximately 137, 60, and 10 pmol / mg protein, respectively. The maximum binding capacity for each site was calculated based on this.

[0112] During the calculation of the actual competitive binding probability, a mass-action relationship model based on drug concentration and site capacity was established. When the plasma concentration of atorvastatin was 100 ng / ml, the actual binding probabilities at the CYP3A4, CYP2C9, and CYP2D6 sites were substituted into the mass-action equation and calculated to be 62%, 25%, and 13%, respectively. Comparing the actual binding probabilities with the initial binding probabilities, the inhibition coefficients for the three sites were calculated to be 0.83 (62% / 75%), 1.25 (25% / 20%), and 2.6 (13% / 5%), respectively. This indicates that under actual dosing conditions, the CYP3A4 site is partially inhibited, while the relative importance of the CYP2C9 and CYP2D6 sites increases.

[0113] When applying the coincident site inhibition coefficient to drug diffusion calculations, a three-dimensional tissue spatial grid model was constructed with a grid size of 1 μm × 1 μm × 1 μm. The numerical model was established based on Fick's diffusion law, with the initial condition set to the boundary concentration of the drug entering the tissue from the blood vessels. The initial concentration of atorvastatin was 100 ng / ml. Without considering the inhibition coefficient, the standard diffusion coefficient of atorvastatin in liver tissue is 3.6 × 10-7 cm 2 After the coincident site inhibition coefficient is introduced, the local diffusion coefficient is adjusted according to the distribution density of the site in the tissue. For example, in the CYP3A4 enzyme-enriched area, the local diffusion coefficient is reduced to 3.0×10 -7 cm 2 / s(3.6×10 -7 ×0.83); in the CYP2D6-enriched area, diffusion almost stagnated, and the local diffusion coefficient dropped to 9.4×10 -8 cm 2 / s(3.6×10 -7 / 2.6).

[0114] During the iterative calculation process, the initial time step is set to 0.01 seconds and adaptively adjusted based on the rate of change of drug concentration. If the local concentration change rate exceeds 10% per step, the time step is halved. If the concentration change rate is less than 1% per step for 10 consecutive steps, the time step is doubled, up to a maximum of 1 second. In a typical 100-second simulation, approximately 1000 iterative steps are performed to obtain time-varying 3D distribution data of the drug in the tissue.

[0115] The resulting dynamic drug distribution curves included drug concentration-time curves at different tissue depths, drug enrichment curves for specific metabolic enzyme regions, and a time-varying curve for the total drug volume in the entire tissue. The results showed that atorvastatin rapidly penetrated the tissue surface within 10 seconds after administration, reached the middle tissue layer within 30 seconds, and reached a steady-state distribution after 60 seconds. The drug concentration in the CYP3A4-enriched region reached a maximum of 138 ng / ml at 40 seconds, while the CYP2D6 region reached a peak of 63 ng / ml at 80 seconds. These differences in spatiotemporal distribution directly reflect the competitive binding characteristics of different metabolic sites and the differences in local drug metabolic conversion rates, providing an important reference for clinical drug dosage adjustments and drug interaction assessments.

[0116] This technical solution accurately extracts the spatial structure and chemical environment characteristics of overlapping sites by fusing metabolic site distribution maps with molecular dynamics simulations, significantly improving the accuracy of binding probability estimation compared to existing methods that rely on static two-dimensional structures or empirical rules. On this basis, the binding energy value is introduced to construct the mass action equation, and the competitive binding probability and inhibition coefficient are dynamically calculated to achieve quantitative modeling of the competitive behavior of drugs at metabolic sites, breaking through the qualitative analysis limitations of traditional methods for predicting drug interactions. Furthermore, by taking the inhibition coefficient as a diffusion resistance factor and combining the drug concentration with time changes to construct a drug diffusion model, the distribution state of the drug in the tissue space is dynamically simulated, and a distribution curve with time series characteristics is generated, which effectively makes up for the lack of a dynamic response mechanism in the existing methods in spatial distribution prediction. Starting from multidimensional characteristics, this application introduces dynamic calculation and time control mechanisms to achieve spatial precision and time series continuity in drug distribution prediction, providing a more scientific and reliable modeling tool for drug metabolic behavior evaluation and interaction mechanism analysis.

[0117] In an optional embodiment, determining the overlapping interval of the metabolic cycle of the drug to be tested based on the time series curve, determining the concentration accumulation range of the drug to be tested based on the overlapping interval of the metabolic cycle, calculating the interaction coefficient of the drug to be tested using the concentration accumulation range, and obtaining the blood drug concentration change trend of the drug to be tested based on the interaction coefficient includes:

[0118] Scanning the time series curve using a sliding time window, extracting normalized concentration values ​​of the drug to be tested at different time points according to the time series curve, and constructing a concentration data sequence of the drug to be tested according to the normalized concentration values ​​in a time series;

[0119] Acquiring half-life characteristics of the drug to be tested based on the concentration data sequence, constructing an adaptive threshold judgment rule according to the half-life characteristics, performing segmented comparison on the concentration data sequence, and determining an overlapping interval of the metabolic cycle of the drug to be tested;

[0120] Calculating a concentration vector of the drug to be tested within an overlapping interval of the metabolic cycle, constructing a multidimensional concentration accumulation tensor based on the concentration vector, performing a boundary integral operation on the concentration accumulation tensor to obtain a cumulative volume, and determining a concentration accumulation range of the drug to be tested based on the cumulative volume;

[0121] The cumulative effect intensity of the drug to be tested in the overlapping interval of the metabolic cycle is calculated according to the concentration accumulation range, the interaction coefficient of the drug to be tested is calculated based on the cumulative effect intensity, the concentration distribution characteristics of the drug to be tested are analyzed according to the interaction coefficient, the concentration distribution characteristics are fitted and verified with the measured data, the fitting error is calculated, and the adaptive threshold judgment rule is dynamically corrected according to the fitting error to obtain the blood drug concentration change trend of the drug to be tested.

[0122] This embodiment discloses a method for determining the overlapping intervals of the metabolic cycles of a drug to be tested based on a time series curve, determining the cumulative concentration range of the drug to be tested through the overlapping intervals of the metabolic cycles, calculating the interaction coefficient of the drug to be tested using the cumulative concentration range, and obtaining the blood drug concentration change trend of the drug to be tested based on the interaction coefficient.

[0123] In practical applications, the metabolic process of a drug under test in the human body can be tracked by collecting blood samples and measuring drug concentrations to obtain a time series curve. This method uses a sliding time window technique to scan this time series curve, with a window size set to 30 minutes and a sliding step size of 5 minutes. During the scanning process, the system extracts the concentration value of the drug under test every 5 minutes and normalizes it by dividing the measured concentration value by the theoretical maximum concentration of the drug to obtain a normalized concentration value. For example, for an antibiotic drug, the measured initial concentration is 8 mg / L and the theoretical maximum concentration is 10 mg / L, so the normalized concentration value is 0.8. These normalized concentration values ​​are arranged in chronological order to construct a drug concentration data sequence, such as [0.8, 0.75, 0.72, 0.68, 0.65, 0.62, 0.58...].

[0124] Based on the constructed concentration data sequence, the system calculates the half-life characteristics of the drug to be tested by the concentration decline rate. The specific implementation method is: select the concentration values ​​of adjacent time points in the sequence, calculate their ratio, and when the ratio is stable within a certain range (such as between 0.88-0.92) for three consecutive times, the average value of the ratio is used as the drug concentration decay coefficient. The time required for the drug concentration to decay from the initial value to half of the initial value is used as the half-life. For example, if the decay coefficient of a drug is 0.9 and the sampling time interval is 5 minutes, its half-life is calculated to be approximately 33 minutes. Based on this half-life characteristic, an adaptive threshold judgment rule is constructed: when the detected concentration value is lower than 20% of the concentration at the corresponding time point of the previous metabolic cycle, it is determined that a new metabolic cycle has begun; when the concentration curves of two metabolic cycles overlap within a certain time period, and the concentration value of the overlapping area exceeds 15% of the maximum concentration of a single dose, the time period is determined to be the metabolic cycle overlap interval.

[0125] Within the defined overlapping intervals of the metabolic cycle, the concentration vector of the drug to be tested is calculated. Assuming there are five time points within the overlapping interval, and concentration data is collected at each time point for three dimensions (e.g., plasma drug concentration, tissue drug concentration, and metabolite concentration), a 5×3 set of concentration vectors can be constructed. These concentration vectors are then used to construct a multidimensional concentration accumulation tensor, where each element represents the drug concentration value at a specific time point and measurement dimension. For example, for a given anticoagulant drug, the plasma concentrations collected at the five overlapping time points are [0.45, 0.52, 0.58, 0.49, 0.43] mg / L, the tissue concentrations are [0.32, 0.38, 0.41, 0.37, 0.30] mg / L, and the metabolite concentrations are [0.15, 0.19, 0.22, 0.24, 0.25] mg / L.

[0126] The constructed concentration accumulation tensor is subjected to boundary integral operation, and the trapezoidal integration method is used to calculate the cumulative concentration area in each dimension, and then the area of ​​each dimension is combined to calculate the cumulative volume. Taking the above anticoagulant drug as an example, the integral area of ​​the plasma concentration dimension is 2.47 mg·h / L, the integral area of ​​the tissue concentration dimension is 1.78 mg·h / L, and the integral area of ​​the metabolite concentration dimension is 1.05 mg·h / L. The cumulative volume calculated by combining the three dimensions is 4.61 mg 2 ·h 2 / L 2 The system determines the drug concentration accumulation range based on this cumulative volume, such as [4.2, 5.0] mg 2 ·h 2 / L 2 .

[0127] Based on the calculated concentration accumulation range, the system further calculates the cumulative effect strength of the drug under test in the overlapping interval of the metabolic cycle. The calculation method of the cumulative effect strength is to perform a weighted processing on the ratio of the median of the concentration accumulation range to the lower limit of the effective therapeutic concentration of the drug. For example, if the median of the concentration accumulation range of a drug is 4.6mg 2 ·h 2 / L 2 , the lower limit of effective therapeutic concentration is 2.5 mg / L, then the cumulative effect strength can be calculated as 4.6 / 2.5×0.8=1.472. Based on the cumulative effect strength, the interaction coefficient of the drug to be tested is calculated, which represents the cumulative effect of the drug during continuous administration. The calculation of the interaction coefficient takes into account the half-life of the drug, the dosing interval and the cumulative effect strength. For example, if the cumulative effect strength is 1.472, the half-life is 33 minutes, and the dosing interval is 6 hours, the interaction coefficient can be calculated as 1.472×(1-0.9( 6 ×60min / 5) )=1.47.

[0128] The concentration distribution characteristics of the drug to be tested are analyzed based on the calculated interaction coefficient. When the interaction coefficient is greater than 1.2, it indicates that the drug has a significant cumulative effect in the body and the dosing regimen needs to be adjusted. The concentration distribution characteristics obtained by analysis are fitted and verified with the actual collected blood drug concentration data, and the fitting error is calculated. If the fitting error exceeds the preset threshold (such as 10%), the adaptive threshold judgment rule is dynamically corrected according to the error size and direction, for example, the overlap judgment threshold is adjusted from 15% to 17%. Through this iterative optimization method, the accurate blood drug concentration change trend of the drug to be tested is finally obtained, providing a scientific basis for the formulation of clinical medication plans.

[0129] In this embodiment, by constructing a concentration data sequence of the drug to be tested within the metabolic cycle, and combining the sliding time window scan with the adaptive threshold judgment rule, accurate identification of the overlapping intervals of the metabolic cycle is achieved. Compared with the existing technology that relies more on static threshold setting or single time point data to infer the interaction strength, this solution constructs a multi-dimensional concentration accumulation tensor through concentration vectors, and uses the boundary integral method to obtain the concentration accumulation range, making the modeling of the cumulative effect of the drug within the metabolic cycle more dynamic and detailed. On this basis, the judgment threshold is dynamically corrected by further combining the fitting error feedback mechanism, thereby improving the accuracy of the concentration distribution prediction. The overall solution effectively improves the temporal response capability and individual adaptability of the drug interaction intensity assessment, and provides strong support for the realization of high-precision blood drug concentration prediction and individual difference sensitivity analysis.

[0130] Figure 3 This is a comparison chart of the normalization processing effect of the sliding time window drug concentration in the embodiment of the present invention. Figure 3 As shown in the figure, the effect of normalizing the drug concentration curve using a sliding time window is demonstrated. The present technical solution (marked with black squares) achieves more accurate normalized concentration calculations through adaptive window size analysis. Compared with the traditional cubic spline interpolation method (marked with white circles), it shows obvious advantages in the peak area (12 hours), with the normalized peak concentration reaching 0.855, while the traditional method only reaches 0.82. In the drug half-life area (18 hours), the half-life calculated by the present technical solution is 8.4 hours, which is more accurate than the 9.1 hours of the traditional method, reducing the error by about 7.7%. At the 30-hour time point, the normalized concentration value of the present technical solution is 0.3, which is closer to the actual observed data of 0.31, while the traditional method is 0.27, with a relative error of up to 12.9%. The sliding window (blue dotted box) can adaptively adjust the sampling density, providing higher sampling accuracy in areas with rapid concentration changes (10-14 hours), ensuring that the temporal characteristics of the data sequence are preserved, and laying the foundation for subsequent half-life feature extraction and metabolic cycle analysis.

[0131] In an optional embodiment, the concentration vector of the drug to be tested is calculated within the overlapping interval of the metabolic cycle, a multidimensional concentration accumulation tensor is constructed using the concentration vector, a boundary integral operation is performed on the concentration accumulation tensor to obtain a cumulative volume, and determining the concentration accumulation range of the drug to be tested based on the cumulative volume includes:

[0132] Establishing a three-dimensional coordinate system in the overlapping intervals of the metabolic cycle, the three-dimensional coordinate system including a spatial position axis, a metabolic pathway axis, and a time series axis, obtaining a normalized concentration value at each coordinate point, and constructing a drug concentration vector;

[0133] Analyzing the activity intensity of the metabolic enzyme to obtain the substrate affinity between the metabolic enzyme and the drug molecule, calculating the metabolic pathway weight factor based on the activity intensity and substrate affinity, and performing a tensor operation on the metabolic pathway weight factor and the drug concentration vector to generate a drug accumulation tensor;

[0134] Determining a main characteristic direction of the drug cumulative tensor, slicing the drug cumulative tensor along the main characteristic direction, constructing a computational grid based on a concentration gradient distribution of the drug cumulative tensor, and performing adaptive partitioning;

[0135] A piecewise integral operation is performed on the drug cumulative tensor within the divided computational grid to obtain the local cumulative value of each grid. The local cumulative values ​​are summarized to obtain the cumulative volume of the drug. The spatial distribution characteristics of the drug in the overlapping intervals of the metabolic cycle are analyzed based on the cumulative volume, and the concentration accumulation range of the drug is determined in combination with the spatial distribution characteristics.

[0136] This embodiment provides a method for calculating the concentration vector of the drug to be tested in the overlapping interval of the metabolic cycle, which can accurately evaluate the distribution characteristics and accumulation of the drug in the human body. When establishing a three-dimensional coordinate system in the overlapping interval of the metabolic cycle, the spatial position axis is used to represent the spatial distribution of the drug in different tissues and organs, the metabolic pathway axis is used to represent the different metabolic pathways experienced by the drug, and the time series axis is used to represent the dynamic process of the drug changing over time in the body. The collection time interval is set to 15 minutes, and a total of 48 time points are collected to form a 12-hour overlapping interval of the metabolic cycle. Five key metabolic tissues (liver, kidney, fat, muscle and blood) are selected for each spatial position, and 10 sampling points are set for each tissue. For the metabolic pathway, the four metabolic pathways that the drug mainly experiences (oxidation, reduction, hydrolysis and combination) are selected.

[0137] To obtain normalized concentration values, high-performance liquid chromatography-mass spectrometry was used to determine drug concentrations at each sampling point. For a subject taking a standard 100 mg dose of the drug, the initial concentration measured at the first sampling point in liver tissue, within the oxidative metabolic pathway, at time point 1 (15 minutes after administration) was 8.45 μg / mL. This value was normalized to the maximum in vivo concentration of 12.6 μg / mL for the drug, yielding a normalized concentration of 0.67. This method was used to process the data from all sampling points, constructing a drug concentration vector containing 5 × 10 × 4 × 48 = 9600 elements.

[0138] To analyze the activity of metabolic enzymes, enzyme kinetics experiments were used to determine the maximum reaction rate (Vmax) and Michaelis constant (Km) of each metabolic enzyme. For the liver CYP3A4 enzyme, for example, the Vmax was measured to be 45.2 nmol / min / mg protein, and the Km was 5.6 μM. The enzyme activity index (IQ) was calculated as 8.1 based on the relationship between the enzymatic reaction rate and substrate concentration. Molecular docking was also used to calculate the substrate affinity of the metabolic enzyme for the drug molecule. For example, the binding energy of CYP3A4 for the test drug was -8.7 kcal / mol, which translated into an affinity score of 7.8. Combining the enzyme activity and substrate affinity, weight factors were calculated as 0.35 for the oxidative metabolic pathway, 0.25 for the reductive pathway, 0.22 for the hydrolytic pathway, and 0.18 for the binding pathway. These weight factors were combined with the drug concentration vector for tensor operations to generate a drug accumulation tensor with a shape of 5 × 10 × 4 × 48.

[0139] The main eigenvectors of the drug accumulation tensor were determined using tensor decomposition techniques. Using high-order singular value decomposition, the principal eigenvectors were calculated to point toward the liver region on the spatial position axis, favor the oxidative pathway on the metabolic pathway axis, and be concentrated in the 2-4 hour period after drug administration on the time series axis. The drug accumulation tensor was sliced ​​along these main eigenvectors to produce a series of two-dimensional concentration distribution cross-sections. A computational grid was constructed based on the concentration gradient distribution, with a finer mesh used in areas with drastic gradient changes. Specifically, the liver region was divided into a 100×100 fine mesh, and other tissue regions were divided into 50×50 meshes. On the time axis, the interval 2-4 hours after drug administration was refined at 5-minute intervals, while the remaining time periods were kept at 15-minute intervals.

[0140] When performing piecewise integration of the drug accumulation tensor within the divided computational grid, an improved trapezoidal integration method is used. For example, a grid in the liver region, located in the central region of the liver, in the oxidative metabolic pathway, at 180 minutes after drug administration, measures a concentration gradient of 0.056 / mm. The local accumulation value for this grid is calculated to be 0.235 μg·h / mm. 3Similar calculations were performed on all grids and summarized to obtain a cumulative volume of 147.8 μg·h / cm during the entire metabolic cycle. 3 .

[0141] Analyzing the spatial distribution characteristics of the drug during overlapping metabolic cycles revealed that the drug accumulated in the liver, accounting for 68.4% of the total, in the kidneys, 15.7%, in the bloodstream, 10.2%, in muscle tissue, 3.5%, and 2.2% in adipose tissue. In terms of temporal distribution, the main accumulation phase occurred between 2 and 4 hours after dosing, accounting for 57.3% of the total. Regarding metabolic pathway distribution, the oxidative metabolic pathway contributed 62.5% of the drug metabolism, followed by the reduction pathway (21.8%), the hydrolysis pathway (9.4%), and the conjugation pathway (6.3%).

[0142] Combined with the above spatial distribution characteristics, the drug's cumulative concentration range was determined to be 5.2-9.8 μg / mL in the liver, 2.1-3.7 μg / mL in the kidney, 0.8-2.5 μg / mL in the blood, 0.4-0.9 μg / mL in muscle tissue, and 0.2-0.5 μg / mL in adipose tissue. These data indicate that the drug is primarily accumulated in the liver and metabolized primarily through the oxidative metabolism pathway of the liver CYP3A4 enzyme. The drug reaches its highest accumulation level 2-4 hours after administration, with a half-life of approximately 8.5 hours. Attention should be paid to the potential for drug accumulation with multiple dosing.

[0143] In this embodiment, a three-dimensional coordinate system containing spatial position, metabolic pathway and time series is established in the overlapping interval of metabolic cycle, normalized concentration value is obtained and concentration vector is constructed, and on this basis, metabolic enzyme activity intensity and substrate affinity are fused to generate weighted tensor, which effectively depicts the metabolic accumulation characteristics of drugs in multidimensional space. The existing technology usually uses static time point sampling to estimate drug concentration changes, ignores the combined influence of metabolic enzyme activity and spatial distribution, and is difficult to accurately describe complex metabolic processes. The present application realizes dynamic spatial slice analysis of drug concentration changes by constructing drug accumulation tensor and extracting its main characteristic direction, further combines concentration gradient adaptive division calculation grid, and performs slice integration operation in each grid, so as to achieve accurate acquisition of drug cumulative volume and concentration distribution in the overlapping interval of metabolic cycle. This scheme takes the spatial dynamic change of metabolic behavior as the starting point, improves the characterization method of the association between metabolic intensity and distribution, significantly improves the prediction accuracy of concentration accumulation range, and helps to comprehensively evaluate the potential interaction risk of drugs.

[0144] In an optional embodiment, the metabolic inhibition level of the drug to be tested is determined based on the trend of changes in blood drug concentration, and the metabolic inhibition level is compared with a preset safety threshold in the drug knowledge graph to obtain a risk assessment result of the drug to be tested. The metabolic abnormality warning data of the drug to be tested is generated according to the risk assessment result, including:

[0145] Performing time-frequency decomposition on the blood drug concentration change trend, extracting the frequency characteristics and amplitude characteristics of the blood drug concentration change trend, identifying abnormal fluctuation points of the blood drug concentration based on the frequency characteristics and amplitude characteristics, and generating a blood drug concentration fluctuation characteristic map;

[0146] Acquiring dynamic change data of metabolic enzyme activity, constructing a metabolic activity distribution matrix based on the dynamic change data of metabolic enzyme activity, calculating the eigenvalue change rate of the metabolic activity distribution matrix, and determining the metabolic inhibition level of the drug to be tested according to the eigenvalue change rate;

[0147] Obtaining metabolic pathway information and metabolic enzyme-related data of the drug to be tested from the drug knowledge graph, calculating the safety threshold range, and comparing the metabolic inhibition level with the safety threshold range to obtain a risk assessment result of the drug to be tested;

[0148] The risk assessment result is associated with the blood drug concentration fluctuation characteristic map, the metabolic abnormality type of the drug to be tested is determined according to the association analysis result, and metabolic abnormality warning data of the drug to be tested is generated according to the metabolic abnormality type.

[0149] For example, after a patient takes a drug, blood samples are collected at multiple time points, and the blood drug concentration is measured by high performance liquid chromatography-mass spectrometry to form blood drug concentration-time curve data. In one embodiment, after a patient takes the anti-epileptic drug carbamazepine, blood samples are collected at 0.5h, 1h, 2h, 4h, 8h, 12h, and 24h after taking the drug, and the measured blood drug concentrations are 2.1μg / mL, 4.3μg / mL, 6.8μg / mL, 5.7μg / mL, 4.2μg / mL, 3.1μg / mL, and 1.6μg / mL, respectively.

[0150] The obtained blood drug concentration change trend was subjected to time-frequency decomposition. Using the wavelet transform method, Morlet wavelet was selected as the mother wavelet to perform multi-scale decomposition on the blood drug concentration-time curve, decomposing the signal into different frequency components. Taking carbamazepine as an example, through 6-layer wavelet decomposition, sub-band signals with frequency ranges of 0.01-0.05Hz, 0.05-0.1Hz, 0.1-0.2Hz, 0.2-0.4Hz, 0.4-0.8Hz and 0.8-1.6Hz were obtained. Frequency features were extracted for each sub-band signal, including the main frequency, energy proportion, and frequency offset rate; at the same time, amplitude features were extracted, including peak amplitude, root mean square amplitude, and amplitude fluctuation coefficient.

[0151] Based on the extracted frequency and amplitude features, an abnormality detection rule was established: when the frequency deviation rate exceeded 15% or the amplitude fluctuation coefficient exceeded 20%, it was identified as an abnormal fluctuation point. This rule identified that carbamazepine blood concentrations abnormally increased 4 hours after taking the drug and abnormally decreased 8 hours later. These abnormal points were visualized in the three-dimensional space of time-concentration-frequency to generate a blood concentration fluctuation characteristic map. Different colors indicate the magnitude of the fluctuation: red indicates highly abnormal areas, yellow indicates moderately abnormal areas, and green indicates normal areas.

[0152] Dynamic changes in metabolic enzyme activity were also obtained. The activity of CYP3A4, the primary metabolizer of carbamazepine, was measured using a fluorescent substrate assay. In an in vitro experiment, CYP3A4 was extracted from human liver microsomes and treated with varying concentrations of carbamazepine (0 μM, 10 μM, 20 μM, 50 μM, and 100 μM), and enzyme inhibition was measured. The measured enzyme activities at each concentration were 100%, 92%, 78%, 45%, and 22%, respectively. Based on these data, a metabolic activity distribution matrix was constructed, with rows representing different drug concentrations and columns representing different time points. Matrix element values ​​represent the percentage of enzyme activity at specific concentrations and time conditions.

[0153] Eigenvalue decomposition was performed on the metabolic activity distribution matrix to obtain a time-varying curve of the principal eigenvalues. The eigenvalue change rate, defined as the change in eigenvalue per unit time, was calculated. In the case of carbamazepine, the eigenvalue change rate was -0.15 / h in the first 2 hours, -0.28 / h from 2 to 8 hours, and -0.08 / h from 8 to 24 hours. The metabolic inhibition level was determined based on the maximum eigenvalue change rate. The -0.28 / h value was converted to a standardized inhibition index of 76%, indicating that carbamazepine inhibited CYP3A4 by 76%.

[0154] The metabolic pathway information for carbamazepine was obtained from a pre-built drug knowledge graph. The knowledge graph contains drug nodes, enzyme nodes, receptor nodes, and their relationships. The query revealed that carbamazepine is primarily metabolized by CYP3A4, CYP2C8, and CYP2C19, with CYP3A4 contributing 70%, CYP2C8 20%, and CYP2C19 10%. Relevant data for these metabolizing enzymes was obtained, including inhibition thresholds, induction thresholds, and the influence coefficient of genetic polymorphisms. Based on historical clinical data and literature reports, a safety threshold range was calculated. The upper limit of the safety threshold for CYP3A4 inhibition is 65%, and the lower limit is 15%. In other words, inhibition levels within the range of 15%-65% are considered safe.

[0155] Comparing the measured carbamazepine metabolic inhibition level of 76% with the upper limit of the CYP3A4 safety threshold of 65%, it was found to exceed the safety range and enter the moderate-risk range (65%-85%), but did not reach the high-risk threshold (>85%). Therefore, the risk assessment result was "moderate metabolic inhibition risk."

[0156] The risk assessment results were correlated with the blood drug concentration fluctuation profile. A decision tree algorithm was used to establish a correlation model. Inputs included the metabolic inhibition level, the location of abnormal blood drug concentration fluctuation points, and the degree of frequency characteristic shift. The output was the metabolic abnormality type. Analysis of carbamazepine revealed a metabolic inhibition level of 76%, an abnormal increase in blood drug concentration 4 hours after administration, and a 17% main frequency shift, consistent with the characteristic pattern of "saturated inhibition of metabolic enzymes." The metabolic abnormality type was determined to be "decreased drug clearance due to saturated inhibition of the CYP3A4 enzyme."

[0157] Based on the identified metabolic abnormality type, a metabolic abnormality warning data package is generated. This warning data includes a description of the abnormality type, risk level (moderate risk), possible clinical consequences (such as cumulative blood drug concentrations and increased risk of adverse reactions), and recommended interventions (such as adjusting the dose to 70% of the original dose or increasing the frequency of blood drug concentration monitoring). This warning data is distributed to clinicians through the hospital information system to support medication safety decision-making.

[0158] In this embodiment, by performing time-frequency decomposition on the trend of changes in blood drug concentration, extracting frequency and amplitude features, identifying abnormal fluctuation points, generating a fluctuation feature map, and combining the distribution matrix constructed by the dynamic changes in metabolic enzyme activity, the drug metabolism inhibition level is accurately assessed. The existing technology usually only makes risk judgments based on static or average concentration values, fails to consider the dynamic changes of metabolic enzyme activity and blood drug concentration in multi-dimensional time and space, and easily misses potential metabolic abnormalities. This application introduces the rate of change of metabolic enzyme characteristic values ​​as a dynamic discrimination basis, and sets safety thresholds with the help of the metabolic pathways and enzyme data associated in the drug knowledge graph, thereby realizing personalized, path-dependent risk comparison analysis. By correlating the concentration fluctuation map with the risk assessment results, the type of metabolic abnormality can be accurately identified and targeted early warning data can be generated. This technology takes improving the sensitivity and accuracy of metabolic risk prediction as its starting point, and has made improvements in time-frequency domain recognition capabilities and knowledge graph-assisted reasoning, effectively improving the early detection and intervention capabilities of drug metabolic abnormalities.

[0159] A second aspect of an embodiment of the present invention provides a drug interaction intelligent prediction system based on a knowledge graph, the system comprising:

[0160] The first unit is used to extract the molecular structure data and metabolic data of the drug to be tested from the drug knowledge graph, calculate the spatial conformation overlap of the drug to be tested, determine the overlapping sites of the metabolic pathway through graph reasoning, and generate a metabolic site distribution map;

[0161] The second unit is used to calculate the competitive binding probability of the drug to be tested at the coincident site based on the metabolic site distribution map, calculate the coincident site inhibition coefficient based on the competitive binding probability, analyze the drug distribution state under different inhibition coefficients using the drug diffusion calculation method, and generate a time series curve reflecting the dynamic changes of drug distribution;

[0162] The third unit is used to determine the overlapping interval of the metabolic cycle of the drug to be tested based on the time series curve, determine the concentration accumulation range of the drug to be tested based on the overlapping interval of the metabolic cycle, calculate the interaction coefficient of the drug to be tested based on the concentration accumulation range, and obtain the blood drug concentration change trend of the drug to be tested based on the interaction coefficient;

[0163] The fourth unit is used to determine the metabolic inhibition level of the drug to be tested based on the trend of changes in blood drug concentration, compare the metabolic inhibition level with the safety threshold preset in the drug knowledge graph, obtain the risk assessment result of the drug to be tested, and generate metabolic abnormality warning data for the drug to be tested based on the risk assessment result;

[0164] The fifth unit is used to write metabolic abnormality warning data into the drug knowledge graph, update the interaction feature records of the drugs to be tested in the drug knowledge graph, and form new metabolic reference data for the drugs to be tested.

[0165] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0166] processor;

[0167] a memory for storing processor-executable instructions;

[0168] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0169] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0170] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0171] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent prediction method for drug interactions based on knowledge graph, characterized by: include: Extract the molecular structure data and metabolic data of the drug to be tested from the drug knowledge graph, calculate the spatial conformation overlap of the drug to be tested, determine the overlapping sites of the metabolic pathway through graph reasoning, and generate a metabolic site distribution map; Based on the metabolic site distribution map, the competitive binding probability of the drug to be tested at the coincident site is calculated. The coincident site inhibition coefficient is calculated based on the competitive binding probability. The drug distribution state under different inhibition coefficients is analyzed using the drug diffusion calculation method, and a time series curve reflecting the dynamic changes of drug distribution is generated. Determine the overlapping interval of the metabolic cycle of the drug to be tested based on the time series curve, determine the concentration accumulation range of the drug to be tested based on the overlapping interval of the metabolic cycle, calculate the interaction coefficient of the drug to be tested based on the concentration accumulation range, and obtain the blood drug concentration change trend of the drug to be tested based on the interaction coefficient; Determine the metabolic inhibition level of the drug to be tested based on the trend of changes in blood drug concentration, compare the metabolic inhibition level with the safety threshold preset in the drug knowledge graph, obtain the risk assessment result of the drug to be tested, and generate metabolic abnormality warning data for the drug to be tested based on the risk assessment result; The metabolic abnormality warning data is written into the drug knowledge graph, and the interaction feature records of the drugs to be tested in the drug knowledge graph are updated to form new metabolic reference data for the drugs to be tested.

2. The method according to claim 1, characterized in that Extract the molecular structure data and metabolic data of the drug to be tested from the drug knowledge graph, calculate the spatial conformation overlap of the drug to be tested, determine the overlapping sites of the metabolic pathway through graph reasoning, and generate a metabolic site distribution map including: Constructing a drug knowledge graph, setting molecular structure data as graph node attributes, setting metabolic data as graph edge relationships, using a recursive query method to identify the molecular structure data of the drug node to be tested in the drug knowledge graph, traversing the graph edge relationships to obtain metabolic data, and combining the obtained molecular structure data and metabolic data into a drug feature set; Identifying structural matching fragments of the drug to be tested based on the molecular structure data in the drug feature set, spatially superimposing the drug to be tested using the structural matching fragments, and calculating the spatial conformational overlap of the drug to be tested; Analyzing the metabolic reaction pathways of the drug knowledge graph based on the metabolic data in the drug feature set, identifying target nodes with common metabolic features through graph reasoning, screening the target nodes based on the spatial conformational overlap to obtain candidate overlapping sites, calculating the spatial distribution parameters of the candidate overlapping sites based on the spatial conformational overlap, and determining the candidate overlapping sites whose spatial distribution parameters exceed a preset distribution threshold as metabolic pathway overlapping sites; The spatial distribution parameters of the coincident sites of the metabolic pathway and the metabolic data in the drug feature set are integrated to generate a metabolic site distribution map.

3. The method according to claim 2, characterized in that The metabolic reaction pathways of the drug knowledge graph are analyzed based on the metabolic data in the drug feature set. Target nodes with common metabolic features are identified through graph reasoning. Target nodes are screened based on the spatial conformational overlap, and candidate overlap sites are obtained, including: Constructing a metabolic reaction network based on the metabolic data in the drug feature set, setting the metabolic reactions as directed edges, setting the reactants and products as network nodes, and calculating the reaction weight values ​​of the directed edges based on the substrate affinity and catalytic efficiency of the metabolic enzymes; Performing a random walk search on the metabolic reaction network, setting the drug node to be tested as the starting point, determining the node transfer probability according to the reaction weight value, introducing a path length attenuation factor to perform path search, and obtaining a node access sequence; Building a metabolic pattern feature library based on the node access sequence, setting substrate-enzyme-product combinations and metabolic intermediate sharing relationships as pattern features, matching pattern features in the metabolic reaction network, and identifying target nodes with common metabolic features; constructing a spatial scoring function using the obtained spatial conformation overlap, and calculating the spatial compatibility value of the target node based on the spatial scoring function; Calculate the similarity of the metabolic characteristics of the target node and the drug to be tested, construct a multi-objective optimization function, take the metabolic characteristic similarity as the first optimization goal, and take the spatial compatibility value as the second optimization goal. Calculate the comprehensive score of the target node according to the multi-objective optimization function, and select the atlas nodes whose comprehensive score exceeds the preset threshold as candidate coincidence sites.

4. The method according to claim 1, wherein Based on the metabolic site distribution map, the competitive binding probability of the drug to be tested at the coincident site is calculated. The coincident site inhibition coefficient is calculated based on the competitive binding probability. The drug diffusion calculation method is used to analyze the drug distribution state under different inhibition coefficients. The time series curve reflecting the dynamic changes of drug distribution is generated, including: Extract the spatial structure data and chemical environment data of the overlapping sites from the metabolic site distribution map, use molecular dynamics to simulate the conformational changes of the drug to be tested, and obtain the binding energy value between the drug to be tested and the overlapping sites; Calculating the initial competitive binding probability of the test drug at each overlapping site based on the binding energy value, determining the maximum binding capacity of each overlapping site, and establishing a mass action equation; Substituting the concentration data of the drug to be tested into the mass action equation, calculating the actual competitive binding probability of the coincident site, and determining the ratio of the actual competitive binding probability to the initial competitive binding probability as the coincident site inhibition coefficient; The coincidence site inhibition coefficient is used as the diffusion resistance factor to calculate the diffusion rate of the drug to be tested in the tissue space, and the drug concentration value of each area is obtained. The calculation time interval is dynamically adjusted according to the change amplitude of the drug concentration value, and an iterative operation is used to obtain the drug distribution data of multiple time points. The drug distribution data is converted into a time series to generate a drug distribution dynamic change curve.

5. The method according to claim 1, characterized in that Determine the overlapping interval of the metabolic cycle of the drug to be tested based on the time series curve, determine the concentration accumulation range of the drug to be tested based on the overlapping interval of the metabolic cycle, calculate the interaction coefficient of the drug to be tested based on the concentration accumulation range, and obtain the blood drug concentration change trend of the drug to be tested based on the interaction coefficient, including: Scanning the time series curve using a sliding time window, extracting normalized concentration values ​​of the drug to be tested at different time points according to the time series curve, and constructing a concentration data sequence of the drug to be tested according to the normalized concentration values ​​in a time series; Acquiring half-life characteristics of the drug to be tested based on the concentration data sequence, constructing an adaptive threshold judgment rule according to the half-life characteristics, performing segmented comparison on the concentration data sequence, and determining an overlapping interval of the metabolic cycle of the drug to be tested; Calculating a concentration vector of the drug to be tested within an overlapping interval of the metabolic cycle, constructing a multidimensional concentration accumulation tensor based on the concentration vector, performing a boundary integral operation on the concentration accumulation tensor to obtain a cumulative volume, and determining a concentration accumulation range of the drug to be tested based on the cumulative volume; The cumulative effect intensity of the drug to be tested in the overlapping interval of the metabolic cycle is calculated according to the concentration accumulation range, the interaction coefficient of the drug to be tested is calculated based on the cumulative effect intensity, the concentration distribution characteristics of the drug to be tested are analyzed according to the interaction coefficient, the concentration distribution characteristics are fitted and verified with the measured data, the fitting error is calculated, and the adaptive threshold judgment rule is dynamically corrected according to the fitting error to obtain the blood drug concentration change trend of the drug to be tested.

6. The method according to claim 5, characterized in that Calculating the concentration vector of the drug to be tested within the overlapping interval of the metabolic cycle, constructing a multidimensional concentration accumulation tensor based on the concentration vector, performing a boundary integral operation on the concentration accumulation tensor to obtain a cumulative volume, and determining the concentration accumulation range of the drug to be tested based on the cumulative volume includes: Establishing a three-dimensional coordinate system in the overlapping intervals of the metabolic cycle, the three-dimensional coordinate system including a spatial position axis, a metabolic pathway axis, and a time series axis, obtaining a normalized concentration value at each coordinate point, and constructing a drug concentration vector; Analyzing the activity intensity of the metabolic enzyme to obtain the substrate affinity between the metabolic enzyme and the drug molecule, calculating the metabolic pathway weight factor based on the activity intensity and substrate affinity, and performing a tensor operation on the metabolic pathway weight factor and the drug concentration vector to generate a drug accumulation tensor; Determining a main characteristic direction of the drug cumulative tensor, slicing the drug cumulative tensor along the main characteristic direction, constructing a computational grid based on a concentration gradient distribution of the drug cumulative tensor, and performing adaptive partitioning; A piecewise integral operation is performed on the drug cumulative tensor within the divided computational grid to obtain the local cumulative value of each grid. The local cumulative values ​​are summarized to obtain the cumulative volume of the drug. The spatial distribution characteristics of the drug in the overlapping intervals of the metabolic cycle are analyzed based on the cumulative volume, and the concentration accumulation range of the drug is determined in combination with the spatial distribution characteristics.

7. The method according to claim 1, characterized in that The metabolic inhibition level of the drug to be tested is determined based on the trend of changes in blood drug concentration. The metabolic inhibition level is compared with the safety threshold preset in the drug knowledge graph to obtain the risk assessment result of the drug to be tested. Based on the risk assessment result, metabolic abnormality warning data of the drug to be tested is generated, including: Performing time-frequency decomposition on the blood drug concentration change trend, extracting the frequency characteristics and amplitude characteristics of the blood drug concentration change trend, identifying abnormal fluctuation points of the blood drug concentration based on the frequency characteristics and amplitude characteristics, and generating a blood drug concentration fluctuation characteristic map; Acquiring dynamic change data of metabolic enzyme activity, constructing a metabolic activity distribution matrix based on the dynamic change data of metabolic enzyme activity, calculating the eigenvalue change rate of the metabolic activity distribution matrix, and determining the metabolic inhibition level of the drug to be tested according to the eigenvalue change rate; Obtaining metabolic pathway information and metabolic enzyme-related data of the drug to be tested from the drug knowledge graph, calculating the safety threshold range, and comparing the metabolic inhibition level with the safety threshold range to obtain a risk assessment result of the drug to be tested; The risk assessment result is associated with the blood drug concentration fluctuation characteristic map, the metabolic abnormality type of the drug to be tested is determined according to the association analysis result, and metabolic abnormality warning data of the drug to be tested is generated according to the metabolic abnormality type.

8. A drug interaction intelligent prediction system based on knowledge graph, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to extract the molecular structure data and metabolic data of the drug to be tested from the drug knowledge graph, calculate the spatial conformation overlap of the drug to be tested, determine the overlapping sites of the metabolic pathway through graph reasoning, and generate a metabolic site distribution map; The second unit is used to calculate the competitive binding probability of the drug to be tested at the coincident site based on the metabolic site distribution map, calculate the coincident site inhibition coefficient based on the competitive binding probability, analyze the drug distribution state under different inhibition coefficients using the drug diffusion calculation method, and generate a time series curve reflecting the dynamic changes of drug distribution; The third unit is used to determine the overlapping interval of the metabolic cycle of the drug to be tested based on the time series curve, determine the concentration accumulation range of the drug to be tested based on the overlapping interval of the metabolic cycle, calculate the interaction coefficient of the drug to be tested based on the concentration accumulation range, and obtain the blood drug concentration change trend of the drug to be tested based on the interaction coefficient; The fourth unit is used to determine the metabolic inhibition level of the drug to be tested based on the trend of changes in blood drug concentration, compare the metabolic inhibition level with the safety threshold preset in the drug knowledge graph, obtain the risk assessment result of the drug to be tested, and generate metabolic abnormality warning data for the drug to be tested based on the risk assessment result; The fifth unit is used to write metabolic abnormality warning data into the drug knowledge graph, update the interaction feature records of the drugs to be tested in the drug knowledge graph, and form new metabolic reference data for the drugs to be tested.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images