Intelligent evaluation system and method for safety of combined medication of traditional Chinese medicine and western medicine

Abstract

The invention discloses a traditional Chinese medicine and western medicine combined medication safety intelligent evaluation system and method, and belongs to the technical field of medicine safety evaluation. The system comprises a traditional Chinese medicine component analysis module, a hypergraph convolution prediction module, a medicine effect collaborative evaluation module and an individualized decision module; a quaternary hypergraph network of traditional Chinese medicine components-western medicine-metabolic enzyme-target spots is constructed, and a hypergraph convolutional neural network and a graph attention mechanism are adopted to predict pharmacokinetic interaction; evaluating pharmacodynamic interaction by constructing a target-pathway double-layer network, and establishing a closed-loop feedback collaboration mechanism of pharmacokinetics and pharmacodynamics; cYP450 enzyme system gene polymorphism and transporter gene variation information of a patient are integrated, individualized medication decision and dosage adjustment are achieved, the problems that in the prior art, multi-component multi-target interaction prediction is not accurate, pharmacokinetics-pharmacodynamics collaborative analysis is lacked, and the individualized guiding ability is insufficient are solved, and the application prospect is wide. And the safety and effectiveness of combination of Chinese and western medicines are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of drug safety assessment technology, and in particular to an intelligent assessment system and method for the safety of combined use of traditional Chinese medicine and Western medicine based on hypergraph convolutional neural networks and pharmacokinetic-pharmacodynamic dual-layer mechanism analysis. Background Technology

[0002] The integration of traditional Chinese medicine (TCM) and Western medicine is becoming increasingly common in clinical practice, and the combined use of TCM compound prescriptions and chemical drugs has demonstrated unique advantages in the treatment of various diseases. However, TCM compound prescriptions typically contain dozens or even hundreds of chemical components, which may interact complexly with chemical drugs, leading to reduced efficacy or increased toxicity. Statistics show that adverse reactions caused by the combined use of TCM and Western medicine account for 15%–20% of all adverse drug reactions in clinical practice, seriously threatening patient medication safety.

[0003] Existing technology CN117995430A discloses a method for predicting multi-drug therapy responses based on changes in blood viscosity. This method dynamically detects blood rheological parameters using a fully automated blood viscosity monitor, constructs a knowledge graph of interactions between multiple drugs, and collects data on drug-drug, drug-food, and drug-disease interactions. However, this technology has the following shortcomings: First, it only focuses on blood viscosity as a single physiological indicator, failing to comprehensively reflect the multi-dimensional characteristics of interactions between traditional Chinese medicine and Western medicine; second, it fails to systematically analyze the multi-component and multi-target characteristics of traditional Chinese medicine compound prescriptions, lacking in-depth research on the molecular-level interaction mechanisms between single components of traditional Chinese medicine and chemical drugs; third, it does not consider the significant impact of individual genetic polymorphism on drug metabolism, failing to provide truly personalized medication guidance; fourth, it lacks a synergistic analysis mechanism of pharmacokinetic and pharmacodynamic interactions, making it difficult to accurately predict the clinical outcomes of combined drug use.

[0004] From the perspective of drug interaction mechanisms, the safety issues of combining traditional Chinese medicine (TCM) and Western medicine mainly stem from two levels: pharmacokinetic interactions and pharmacodynamic interactions. Pharmacokinetic interactions refer to the influence of TCM components on the absorption, distribution, metabolism, and excretion of chemical drugs in the body, particularly through competitive inhibition or induction at the level of drug-metabolizing enzymes and transporters, which may lead to abnormally high or low blood concentrations of chemical drugs. Pharmacodynamic interactions refer to the synergistic or antagonistic effects produced by TCM components and chemical drugs acting on the same or related pharmacological targets and signaling pathways. Current technologies often analyze only one level in isolation, failing to establish the correlation and feedback mechanisms between the two.

[0005] Furthermore, the chemical composition of traditional Chinese medicine (TCM) compound prescriptions is complex and varies greatly from person to person; the content of active ingredients in the same TCM herb from different origins and batches may differ significantly. Simultaneously, individual patients exhibit genetic polymorphisms in drug-metabolizing enzyme genes (such as the CYP450 enzyme system) and drug transporter genes, leading to substantial differences in metabolic capacity and susceptibility to adverse reactions to the same combination of TCM and Western medicine among different patients. Current technologies lack a comprehensive consideration of these individualized factors, resulting in insufficient targeting and accuracy in medication recommendations.

[0006] From a technical perspective, traditional drug interaction prediction mainly relies on clinical observation, in vitro experiments, and animal experiments. These methods are time-consuming, costly, and struggle to cover the vast possibilities of combining traditional Chinese medicine (TCM) formulas with chemical drugs. While artificial intelligence (AI) has made significant progress in drug development in recent years, existing drug interaction prediction models are mostly designed for interactions between chemical drugs, with limited ability to model multi-component, multi-target systems like TCM formulas. In particular, traditional graph neural network models struggle to effectively capture the complex semantic dependencies between TCM components, chemical drugs, metabolic enzymes, and targets.

[0007] Therefore, there is an urgent need to develop an intelligent assessment system that can systematically analyze the interaction mechanisms between multiple components of traditional Chinese medicine compound and chemical drugs at both the pharmacokinetic and pharmacodynamic levels, and combine this system with individual patient genotype information to achieve accurate prediction and personalized medication guidance, so as to improve the safety and effectiveness of integrated traditional Chinese and Western medicine treatment. Summary of the Invention

[0008] The purpose of this invention is to provide an intelligent safety assessment system and method for the combined use of traditional Chinese medicine and Western medicine. It constructs a quaternary hypergraph network of traditional Chinese medicine components, Western medicine, metabolic enzymes, and targets, combined with a hypergraph convolutional neural network and graph attention mechanism, to achieve accurate prediction of pharmacokinetic interactions. It also constructs a target-pathway bilayer network to analyze pharmacodynamic interactions, establishing a closed-loop feedback and synergistic mechanism between pharmacokinetics and pharmacodynamics. Furthermore, it integrates patient gene polymorphism information to achieve personalized medication decisions, addressing the problems of insufficient accuracy in predicting multi-component and multi-target interactions, lack of pharmacokinetic-pharmacodynamic synergistic analysis, and inadequate personalized guidance capabilities in existing technologies.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] This invention provides an intelligent safety assessment system for combined use of traditional Chinese medicine (TCM) and Western medicine, comprising a TCM component analysis module, a hypergraph convolutional prediction module, a pharmacodynamic synergy assessment module, and a personalized decision-making module. The TCM component analysis module receives information about TCM compound prescriptions and, based on a TCM chemical component database, decomposes the TCM compound prescriptions into individual chemical components. Mass spectrometry analysis is used to identify the molecular structure and content of each individual chemical component, generating a TCM component feature vector. The hypergraph convolutional prediction module constructs a four-way hypergraph network of TCM components, Western medicine, metabolic enzymes, and targets. Hyperedges of the hypergraph network connect TCM individual component nodes, Western medicine molecule nodes, drug-metabolizing enzyme nodes, and target nodes. High-order semantic dependency learning is performed through a hypergraph convolutional neural network, and the competitive inhibitory strength of TCM components and Western medicine molecules on drug-metabolizing enzymes is calculated based on a graph attention mechanism, predicting the risk level of changes in Western medicine blood concentration caused by pharmacokinetic interactions. The pharmacodynamic synergy assessment module acquires target information of traditional Chinese medicine components and Western medicine molecules, constructs a target-pathway two-layer network, and calculates target overlap and pathway crosstalk strength through network topology analysis to determine the type of synergistic or antagonistic pharmacodynamic interactions. When the pharmacokinetic interaction risk level is high or the pharmacodynamic interaction type is antagonistic, a risk warning signal is generated and the prediction parameters of the hypergraph convolution prediction module are adjusted in reverse to form a closed-loop feedback mechanism. The individualized decision-making module receives patient genotype data, predicts the patient's metabolic capacity coefficients for traditional Chinese medicine components and Western medicine molecules, calculates an individualized medication risk score, and automatically generates a dosage adjustment plan when the risk score exceeds a threshold. Actual adverse reaction data is then fed back to the hypergraph convolution prediction module for incremental model learning.

[0011] The beneficial effects of this invention are as follows:

[0012] First, by constructing a quaternary hypergraph network and using a hypergraph convolutional neural network for modeling, we can effectively capture the high-order complex relationships between multiple components of traditional Chinese medicine, chemical drugs, metabolic enzymes, and targets. Compared with traditional pairwise relationship modeling methods, hypergraph networks can simultaneously represent the associations between multiple entities, more accurately reflect the essential characteristics of the synergistic effects of multiple components in traditional Chinese medicine compound formulas, and significantly improve the accuracy of pharmacokinetic interaction prediction.

[0013] Second, a closed-loop feedback synergy mechanism for pharmacokinetic interaction prediction and pharmacodynamic interaction assessment was established. The pharmacodynamic synergy assessment module, based on the interaction type determined by target overlap and pathway crosstalk intensity, can adjust the prediction parameters of the hypergraph convolution prediction module in reverse, achieving dynamic optimization between the two modules. This synergy mechanism ensures that the combined use of traditional Chinese medicine and Western medicine considers both changes in blood drug concentration and pharmacological effects, avoiding prediction biases caused by isolated analysis of a single aspect in existing technologies.

[0014] Third, by integrating patients' CYP450 enzyme system gene polymorphisms and drug transporter gene variation information, truly personalized medication decisions are achieved. The metabolic capacity coefficient calculated based on genotype can quantify the metabolic differences of different patients to drug combinations, automatically generating targeted dosage adjustment plans. Compared to the group-based recommendations of existing technologies, the personalized guidance of this invention can significantly reduce the incidence of adverse reactions and improve medication safety.

[0015] Fourth, the graph attention mechanism is used to calculate the contribution of different nodes to the prediction results, providing interpretability for the prediction results. Clinicians can intuitively understand which Chinese medicine components and which metabolic enzymes have the greatest impact on the interactions, which helps to optimize the compatibility of Chinese medicine and select alternative drugs, thus enhancing the clinical applicability of the system.

[0016] Fifth, by using actual adverse reaction case data for incremental model learning, the system can continuously optimize prediction accuracy. With the accumulation of clinical application data, the hypergraph convolutional prediction model continuously improves itself, forming a virtuous cycle of prediction-application-feedback-optimization, ensuring the system always maintains optimal performance. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall structure of the intelligent assessment system for the safety of combined use of traditional Chinese medicine and Western medicine according to the present invention;

[0018] Figure 2 This is a schematic diagram of the structure of the herbal component analysis module of the present invention;

[0019] Figure 3 This is a flowchart of the individualized decision-making module of the present invention;

[0020] Figure 4 This is a flowchart of the intelligent safety assessment method for the combined use of traditional Chinese medicine and Western medicine according to the present invention. Detailed Implementation

[0021] Please refer to the attached document. Figures 1-4 The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0022] Reference Figure 1 This invention provides an intelligent safety assessment system for the combined use of traditional Chinese medicine and Western medicine, including a traditional Chinese medicine component analysis module 1, a hypergraph convolution prediction module 2, a drug efficacy synergy assessment module 3, and an individualized decision-making module 4.

[0023] Reference Figure 2 The Traditional Chinese Medicine (TCM) component analysis module 1 is used to receive TCM compound information and perform component breakdown and feature extraction. This module includes a component database unit, a mass spectrometry identification unit, and a feature extraction unit.

[0024] The component database unit stores chemical component information covering more than 500 commonly used Chinese medicinal herbs, including detailed data on the names, molecular formulas, molecular weights, molecular structures, pharmacological effects, and metabolic pathways of over 10,000 monomeric chemical components. The database information is sourced from authoritative data sources such as the Chinese Materia Medica Database (TCMID), the Traditional Chinese Medicine Systems Pharmacology Database (TCMSP), and the Traditional Chinese Medicine Database (TCM Database@Taiwan), and has undergone manual verification and integration to ensure the accuracy and completeness of the data.

[0025] The mass spectrometry identification unit (MSI) employs liquid chromatography-mass spectrometry (LC-MS) to analyze traditional Chinese medicine (TCM) compound samples. Specifically, after extraction, concentration, and filtration pretreatment, the TCM compound samples are injected into a high-performance liquid chromatography (HPLC) system for separation. The chromatographic conditions are: a C18 reversed-phase column (250 mm × 4.6 mm, 5 μm), a water-acetonitrile gradient elution mobile phase, a flow rate of 1.0 mL / min, and a column temperature of 35 °C. The separated components are then detected by a mass spectrometer using an electrospray ionization (ESI) source, with simultaneous scanning in both positive and negative ion modes, and a mass scan range of 100–1500 m / z. The MSI acquires the mass-to-charge ratio, retention time, and mass spectrometric fragmentation mode of each monomeric chemical component.

[0026] The feature extraction unit intelligently matches the data obtained by the mass spectrometry identification unit with the standard data in the component database unit. The matching algorithm employs multi-dimensional similarity calculation, comprehensively considering the deviation of mass-to-charge ratio (tolerance ±0.01 Da), the consistency of retention time (tolerance ±0.5 min), and the matching degree of secondary mass spectrometry fragments (similarity >80%). For successfully matched compounds, the feature extraction unit further obtains their peak areas and calculates the absolute or relative content of the compound in the traditional Chinese medicine compound using external or internal standard methods. The feature extraction unit ultimately generates a feature vector of traditional Chinese medicine components, where each dimension of the vector corresponds to a monomeric chemical component, and the numerical value represents the content of that component.

[0027] In a preferred embodiment of the present invention, for the classic traditional Chinese medicine compound Liuwei Dihuang Wan, the traditional Chinese medicine component analysis module 1 identifies 83 major chemical components, including mononoside, loganin, cornus officinalis glycoside, paeonol, and paeoniflorin, generating an 83-dimensional feature vector. The content of mononoside is 2.5 mg / g, the content of paeonol is 1.8 mg / g, and the content of paeoniflorin is 3.2 mg / g.

[0028] Hypergraph convolution prediction module 2 is used to predict pharmacokinetic interactions. This module first constructs a quaternary hypergraph network of traditional Chinese medicine components, Western medicine, metabolic enzymes, and targets.

[0029] A hypergraph network G=(V, E) consists of a node set V and a hyperedge set E. The node set V includes four types of nodes: nodes containing single components of traditional Chinese medicine. Western medicine molecular nodes Drug metabolism enzyme nodes and target nodes Each hyperedge in the set E can connect multiple nodes of different types, representing higher-order relationships between them. For example, a hyperedge can simultaneously connect a node of a traditional Chinese medicine component, a node of a Western medicine molecule, and a node of a metabolic enzyme, indicating that the traditional Chinese medicine component and the Western medicine molecule have a competitive metabolic relationship with the metabolic enzyme.

[0030] The hypergraph network was constructed based on the following data sources: First, information on the metabolic enzymes and target sites of Western medicine molecules was obtained from the DrugBank database; second, information on the substrates of drug-metabolizing enzymes of traditional Chinese medicine components was obtained from a database of traditional Chinese medicine components; third, the binding affinity between traditional Chinese medicine components and metabolic enzymes, and between Western medicine molecules and metabolic enzymes, was calculated using molecular docking technology. Molecular docking was performed using AutoDock Vina software. Protein receptor structures were obtained from the Protein Data Bank (PDB), and ligand molecular structures were drawn using ChemDraw and converted to pdbqt format using Open Babel. The docking box was set to a 20Å × 20Å × 20Å region covering the active sites of metabolic enzymes, and the search precision was set to 8. Ligand-receptor pairs with docking scores (in kcal / mol) below -6.0 were considered to have effective binding.

[0031] Based on hypergraph networks, the hypergraph convolutional prediction module 2 uses a hypergraph convolutional neural network (HGCN) for feature learning and interaction prediction. The core idea of ​​HGCN is to capture high-order semantic dependencies by aggregating and transmitting information through nodes connected by hyperedges.

[0032] In the innovative algorithm of this invention, the hypergraph convolution operation is defined as follows:

[0033] ,

[0034] in, For the first The feature matrix of the layer node has a dimension of , The total number of nodes. For the first Layer feature dimension; The hypergraph incidence matrix has dimensions . , Let be the total number of superedges, if the node Belongs to superedge but ,otherwise ; The diagonal hyperedge weight matrix has dimensions of . diagonal elements Indicates the superedge The weights; This is a diagonal matrix with node degree, and the diagonal elements are... ; This is a hyper-edge-degree diagonal matrix, with diagonal elements. ; For the first The trainable weight matrix of the layer; The ReLU function is used as the activation function. This formula implements feature aggregation through nodes connected by hyperedges, where... The normalized Laplacian matrix operation of the hypergraph was partially completed, ensuring balanced information propagation among nodes and hyperedges of different degrees.

[0035] To further enhance the model's expressive power, the hypergraph convolutional prediction module 2 introduces a graph attention mechanism to calculate the contribution of different nodes to the prediction results. The formula for calculating the attention weights is:

[0036] ,

[0037] in, For nodes For nodes Attention weights; and They are nodes and nodes eigenvectors; It is a linear transformation matrix; This is the attention coefficient vector; This represents a vector concatenation operation; For nodes The set of neighboring nodes; LeakyReLU is a linear rectified function with leakage, and the slope for negative values ​​is set to 0.2. This attention mechanism can adaptively learn the importance of different nodes for interaction prediction, improving the accuracy and interpretability of the model.

[0038] Based on updated node features, the hypergraph convolution prediction module 2 predicts the competitive inhibition intensity of traditional Chinese medicine components and Western medicine molecules on drug-metabolizing enzymes. Competitive inhibition intensity. The calculation comprehensively considers both the molecular docking binding energy and the node embedding representation learned by the hypergraph network:

[0039] ,

[0040] in, This refers to the binding energy between components of traditional Chinese medicine and metabolic enzymes. This refers to the docking and binding energy between Western medicine molecules and metabolic enzymes. , , These are the embedding vectors of traditional Chinese medicine component nodes, Western medicine molecule nodes, and metabolic enzyme nodes after hypergraph convolution, respectively. and These are weighting coefficients, with preferred values ​​of 0.6 and 0.4, respectively. The competition intensity calculation function is based on the docking binding energy. , This is a similarity calculation function based on embedded vectors, using cosine similarity.

[0041] Based on the competitive inhibition strength, the hypergraph convolution prediction module 2 further predicts the magnitude of changes in the blood concentration of Western medicine. For cases where drug-metabolizing enzymes are inhibited, the increase in blood drug concentration can be predicted using the following formula:

[0042] ,

[0043] in, This refers to the steady-state blood concentration of the Western medicine when there is no interaction. The inhibition fraction of traditional Chinese medicine components at metabolic sites, through Calculation, where This represents the concentration of components in traditional Chinese medicine at metabolic sites. This is the suppression constant. When A risk level exceeding 30% is considered high-risk; a risk level between 10% and 30% is considered medium-risk; and a risk level below 10% is considered low-risk.

[0044] In a specific embodiment of the present invention, when Liuwei Dihuang Pills are used in combination with warfarin, the hypergraph convolution prediction module 2 identifies that paeonol has a competitive inhibitory effect on CYP2C9 enzyme (the docking binding energy is -7.2 kcal / mol, while the docking binding energy between warfarin and CYP2C9 is -8.1 kcal / mol), calculates the competitive inhibition intensity as 0.68, predicts a 35% increase in warfarin blood concentration, and classifies the risk level as high risk.

[0045] Module 3, the pharmacodynamic synergy assessment module, is used to evaluate pharmacodynamic interactions. This module constructs a target-pathway two-layer network to achieve a comprehensive analysis of the relationships between traditional Chinese medicine components and Western medicine molecules at the pharmacological level.

[0046] The target-pathway bilayer network consists of two sub-networks. The first layer is the Target Interaction Network (TIN), where nodes represent protein targets and edges represent interactions between proteins, including physical binding, phosphorylation, and regulation. Target interaction data were obtained from the STRING and BioGRID databases, retaining only experimentally validated interactions with a confidence score greater than 0.7. The second layer is the Signaling Pathway Network (SPN), where nodes represent signaling pathways and edges represent crosstalk between pathways. Pathway information was obtained from the KEGG and Reactome databases, and pathway crosstalk was determined by analyzing shared key protein nodes between pathways.

[0047] The synergistic efficacy evaluation module 3 first calculates the set of target sites for the components of traditional Chinese medicine. Set of molecular targets of Western medicine overlap :

[0048] ,

[0049] in, The number of elements in the intersection of the two target sets. and These represent the number of elements in the two target sets, respectively. Geometric mean normalization is used to avoid bias caused by differences in the number of target points. When the value is greater than 0.3 (the third preset threshold), it is considered that there is significant target overlap.

[0050] For overlapping targets, the synergistic efficacy assessment module 3 further analyzes the action directions of traditional Chinese medicine components and Western medicine molecules on these common targets. Action directions are categorized into agonist and antagonist types. If the action directions of traditional Chinese medicine components and Western medicine molecules on a target are consistent (both agonist or both antagonist), then that target contributes positively to synergistic enhancement; if the action directions are opposite (one agonist, the other antagonist), then that target contributes positively to mutual antagonism. The synergistic efficacy assessment module 3 calculates the synergistic score. :

[0051] ,

[0052] in, Target The importance weights are determined by the degree centrality of the target site in the protein-protein interaction network. This is a directional consistency index; it takes a value of 1 when the directions of action are consistent, and a value of -1 when the directions of action are opposite. When it is determined to be a synergistic effect type, when It is determined to be a mutually antagonistic type.

[0053] In addition to target-level analysis, the pharmacodynamic synergy assessment module 3 also calculates the crosstalk intensity at the pathway level. This is relevant for the set of signaling pathways activated by traditional Chinese medicine components. Collection of signaling pathways activated by Western medicine molecules Crosstalk strength Calculations were made by analyzing key protein nodes shared between pathways:

[0054] ,

[0055] in, and respectively pathway and The key protein set, For pathway and The crosstalk indicator function between the two paths takes a value of 1 if a known crosstalk relationship exists, and a value of 0 otherwise. When the value exceeds 1.5 (the fourth preset threshold), it is considered that there is significant pathway crosstalk, and the pharmacodynamic interaction type determination result based on target overlap needs to be corrected.

[0056] The innovation of the pharmacodynamic synergy assessment module 3 lies in establishing a closed-loop feedback mechanism with the hypergraph convolution prediction module 2. When the pharmacodynamic synergy assessment module 3 determines that the pharmacodynamic interaction type is antagonistic, or detects a high-risk level for the pharmacokinetic interaction, the module generates a risk warning signal and adjusts the prediction parameters of the hypergraph convolution prediction module 2 accordingly. Specifically, the pharmacodynamic synergy assessment module 3 calculates a feedback adjustment factor based on the target overlap and pathway crosstalk intensity. :

[0057] ,

[0058] in, The sensitivity coefficient is preferred, with a value of 0.15. The weighting factor for the impact of path crosstalk on feedback is 0.08, with an optimal value of 0.08. This adjustment factor is used to correct the hyperedge weight matrix in the hypergraph convolutional neural network. The diagonal element, i.e. This allows the information propagation intensity of the hypergraph convolution to be adjusted, enabling the prediction model to dynamically adapt to the interaction characteristics at the pharmacodynamic level.

[0059] In a preferred embodiment of the present invention, when Danshen injection is used in combination with aspirin, the pharmacodynamic synergy evaluation module 3 found that both the main component of Danshen, tanshinone IIA, and aspirin act on platelet aggregation-related targets such as COX-1 and thromboxane A2 receptors, with a target overlap of 0.45 and consistent action directions (both are inhibitory), resulting in a synergy score of 0.72, classifying it as a synergistic enhancement type. However, significant crosstalk was also detected in the signaling pathways activated by both (arachidonic acid metabolism pathway and platelet activation pathway). This may enhance the anticoagulant effect, leading to a bleeding risk. Therefore, the pharmacodynamic synergy assessment module 3 generates a risk warning signal and feeds back the regulatory factors. The weights of the relevant hyperedges are passed to the hypergraph convolution prediction module 2 to increase the prediction sensitivity to the side effects caused by such synergistic effects.

[0060] Reference Figure 3 The personalized decision-making module 4 integrates patient genotype information to enable personalized medication decisions. This module receives patient genotype data, including CYP450 enzyme system gene polymorphism information and drug transporter gene variation information.

[0061] The CYP450 enzyme system is the main enzyme system for drug metabolism, and its genetic polymorphisms lead to significant differences in drug metabolism capacity among individuals. Personalized decision-making module 4 focuses on analyzing single nucleotide polymorphisms (SNPs) in key metabolic enzyme genes such as CYP2C9, CYP2C19, CYP2D6, and CYP3A4. For each gene, the module classifies patients into ultrarapid metabolizers (UM), extensive metabolizers (EM), intermediate metabolizers (IM), or poor metabolizers (PM) based on the detected genotype.

[0062] Taking the CYP2C9 gene as an example, this gene has multiple functional variant sites such as CYP2C92 (R144C) and CYP2C93 (I359L). Wild-type CYP2C91 / 1 is classified as the normal metabolizer, heterozygous variants such as CYP2C91 / 2 or CYP2C91 / 3 are classified as intermediate metabolizers, and homozygous variants such as CYP2C92 / 2, CYP2C93 / 3, or compound heterozygous variants such as CYP2C92 / 3 are classified as slow metabolizers. The individualized decision module 4 sets basal metabolic capacity coefficients for different metabolizers. The values ​​are as follows: 1.8 for rapid metabolizers, 1.0 for normal metabolizers, 0.5 for intermediate metabolizers, and 0.2 for slow metabolizers.

[0063] Besides the CYP450 enzyme system, drug transporters such as P-glycoprotein (P-gp, encoded by the ABCB1 gene) and organic anion transporters (OATP, encoded by the SLCO family genes) play important roles in drug absorption and excretion. The individualized decision module 4 analyzes the variations in these transporter genes and corrects the basal metabolic capacity coefficient. The correction formula is:

[0064] ,

[0065] in, This represents the final metabolic capacity coefficient. To determine the number of transporter genes examined. For the first The variation status indicator variable of each transporter gene (value -1 when there is a loss-of-function variation, value 1 when there is a enhancement-of-function variation, and value 0 when there is no significant variation). For the first The influence coefficient of a transporter gene variation is determined based on the degree of influence of the variation on drug clearance rate reported in the literature, and is usually in the range of 0.1-0.3.

[0066] Based on metabolic capacity coefficient The individualized decision-making module 4, combined with the pharmacokinetic interaction risk level given by the hypergraph convolution prediction module 2 and the pharmacodynamic interaction type given by the pharmacodynamic synergy assessment module 3, calculates the individualized medication risk score. :

[0067] ,

[0068] in, and The weighting coefficients for pharmacokinetic and pharmacodynamic interactions are preferably 0.65 and 0.35, respectively. This is a numerical representation of the risk level of pharmacokinetic interactions (3 for high risk, 2 for medium risk, and 1 for low risk). Let pharmacokinetic risk function be defined as follows: This indicates that the lower the metabolic capacity, the higher the risk. Let be the pharmacodynamic effect function, defined as follows: The stronger the synergistic or antagonistic effect, the more cautious the assessment needs to be. The individualized medication risk score ranges from 0 to 10. When the score exceeds 6.0 (the second preset threshold), the individualized decision-making module 4 determines that medication intervention is necessary.

[0069] When the intervention condition is triggered, the individualized decision-making module 4 automatically generates a dosage adjustment plan. For Western medicine dosages, the adjustment multiple is determined. The determination is based on a combination of the metabolic capacity coefficient and the magnitude of changes in blood drug concentration caused by pharmacokinetic interactions:

[0070] ,

[0071] For example, if the patient is a CYP2C9 slow metabolizer ( ), and the hypergraph convolution prediction module 2 predicted a 50% increase in warfarin blood concentration ( Then the dosage adjustment factor for Western medicine is... It is recommended to reduce the warfarin dose to 20% of the original dose and strengthen INR monitoring.

[0072] For traditional Chinese medicine (TCM) dosage adjustments, the strategy is primarily based on the type of pharmacodynamic interaction. When the pharmacodynamic synergy assessment module 3 determines it to be a synergistic effect, the individualized decision-making module 4 suggests appropriately reducing the TCM dosage, generally by 20%–40% of the original dosage, to avoid excessive synergistic effects leading to adverse reactions. When it is determined to be an antagonistic effect, the module suggests adjusting the compatibility of the TCM compound, removing or replacing specific TCM herbs that antagonize the Western medicine, or adjusting the administration time to ensure that the interval between the TCM and Western medicine is at least 2 hours, reducing the possibility of simultaneous action on the same target.

[0073] The dosage adjustment plan generated by the individualized decision-making module 4 is output in a structured form, including specific medication recommendations, monitoring indicators, follow-up periods, and other information, which is convenient for clinicians to refer to and implement.

[0074] To achieve continuous system optimization, the individualized decision-making module 4 established an adverse reaction data feedback mechanism. During medication use, if an adverse reaction occurs, clinicians enter detailed information about the adverse reaction through the system interface, including basic patient information (age, gender, weight, etc.), medication regimen (name and dosage of traditional Chinese medicine, name and dosage of Western medicine), type of adverse reaction (e.g., bleeding, abnormal liver function, drug rash), severity (mild, moderate, severe), and time of occurrence. These adverse reaction case data are labeled and added to the training dataset.

[0075] The individualized decision-making module 4 feeds back newly added adverse reaction case data to the hypergraph convolutional prediction module 2, triggering incremental model learning. Incremental learning employs an online gradient descent algorithm, updating the model parameters every 10 new cases. Let the current model parameters be... The newly added case dataset is The loss function is cross-entropy loss. The parameter update formula is:

[0076] ,

[0077] in, The learning rate is set to 0.001 to ensure stable model updates. This represents the gradient of the loss function with respect to the model parameters. Through incremental learning, the hypergraph convolutional prediction module 2 can continuously absorb real clinical data and optimize prediction accuracy. In particular, for combinations of traditional Chinese medicine and Western medicine that were not fully covered in the initial training data, incremental learning can quickly improve the model's predictive ability for these combinations.

[0078] In a specific application example of this invention, patient Zhang, male, 65 years old, was taking aspirin 100mg once daily and simvastatin 20mg once daily for coronary heart disease, and additionally taking Liuwei Dihuang Pills 6g twice daily for insomnia. The patient's gene testing results showed that his CYP2C9 genotype was 1 / 3 (intermediate metabolite). The ABCB1 gene C3435T site is TT homozygous (P-gp function is reduced). , ), calculated The hypergraph convolution prediction module 2 predicts that paeonol in Liuwei Dihuang Wan (a traditional Chinese medicine formula) competitively inhibits CYP3A4, and the predicted increase in simvastatin blood concentration is 40%, with a risk level of high risk. No significant pharmacodynamic interactions were found in Module 3 of the pharmacodynamic synergy assessment. ) Personalized Decision Module 4 calculates personalized medication risk scores. Although the intervention threshold of 6.0 was not reached, considering the patient's advanced age and the presence of two gene mutations, the system gave a cautious suggestion, recommending that the simvastatin dose be adjusted to 60% of the original dose, i.e., 12mg once daily, and that creatine kinase and liver function indicators be tested at 1 week and 4 weeks after medication.

[0079] The closed-loop feedback synergy mechanism of this invention is embodied in the bidirectional information flow between the hypergraph convolution prediction module 2 and the pharmacodynamic synergy evaluation module 3. The hypergraph convolution prediction module 2 outputs the pharmacokinetic interaction risk level, while the pharmacodynamic synergy evaluation module 3 outputs the pharmacodynamic interaction type and synergy score. When a high risk or strong antagonism is detected, the pharmacodynamic synergy evaluation module 3 calculates a feedback adjustment factor, adjusting the hyperedge weights of the hypergraph convolution prediction module 2 in reverse to enhance information propagation from relevant nodes. This closed-loop mechanism ensures the synergistic optimization of pharmacokinetic prediction and pharmacodynamic evaluation, avoiding the one-sidedness of single-level analysis. Simultaneously, the individualized decision-making module 4 feeds back adverse reaction data to the hypergraph convolution prediction module 2 for incremental learning, forming a second closed loop of prediction-application-feedback-optimization, ensuring continuous improvement in system performance.

[0080] This invention also provides an intelligent method for assessing the safety of combined use of traditional Chinese medicine and Western medicine, referring to... Figure 4 This includes the following steps:

[0081] Step S1: Receive information on traditional Chinese medicine compound prescriptions through the traditional Chinese medicine component analysis module, break down the traditional Chinese medicine compound prescriptions into individual chemical components based on the traditional Chinese medicine chemical component database, identify the molecular structure and content of each individual chemical component through mass spectrometry analysis technology, and generate a feature vector of traditional Chinese medicine components.

[0082] Specifically, users can input the name and dosage of a traditional Chinese medicine (TCM) compound through the system interface, or upload a sample of the TCM compound for LC-MS analysis. The TCM component analysis module retrieves known chemical component information of the TCM compound from the component database, combines it with the mass spectrometry identification results, determines the actual monomeric components and their content in the batch of TCM compound, and generates a feature vector.

[0083] Step S2: Construct a quaternary hypergraph network of traditional Chinese medicine components, Western medicine, drug-metabolizing enzymes, and targets using a hypergraph convolution prediction module. The hyperedges of the hypergraph network connect nodes of traditional Chinese medicine monomer components, nodes of Western medicine molecules, nodes of drug-metabolizing enzymes, and nodes of targets. Perform high-order semantic dependency learning on the hypergraph network using a hypergraph convolutional neural network. Calculate the competitive inhibition intensity of traditional Chinese medicine components and Western medicine molecules on drug-metabolizing enzymes based on the graph attention mechanism, and predict the risk level of changes in Western medicine blood concentration caused by pharmacokinetic interactions.

[0084] Specifically, the user inputs the name and dosage of the Western medicine to be used in combination. The hypergraph convolutional prediction module obtains the metabolic enzyme information of the Western medicine from the drug database and combines it with the substrate information of the metabolic enzymes of the traditional Chinese medicine component to construct a quaternary hypergraph network. For each traditional Chinese medicine component-Western medicine-metabolic enzyme triplet in the network, the module calculates the binding energy through molecular docking, learns the node embedding representation through the hypergraph convolutional neural network, and comprehensively calculates the competitive inhibition strength. Based on the inhibition strength and pharmacokinetic model, the module predicts the magnitude of change in the blood concentration of the Western medicine and determines the risk level based on the magnitude of the change.

[0085] Step S3: Obtain the target information of traditional Chinese medicine components and Western medicine molecules through the pharmacodynamic synergy assessment module, and construct a target-pathway dual-layer network. The first layer of the target-pathway dual-layer network is the target interaction network, and the second layer is the signal pathway network. Calculate the target overlap and pathway crosstalk intensity of traditional Chinese medicine components and Western medicine molecules through network topology analysis. Based on the overlap, determine the type of pharmacodynamic interaction that is synergistic or antagonistic. When the pharmacokinetic interaction risk level is higher than the first preset threshold or the pharmacodynamic interaction type is antagonistic, the pharmacodynamic synergy assessment module generates a risk warning signal and adjusts the prediction parameters of the hypergraph convolution prediction module in reverse.

[0086] Specifically, the pharmacodynamic synergy assessment module retrieves a list of targets for traditional Chinese medicine components and Western medicine molecules from a target database and calculates the target overlap. For overlapping targets, it analyzes the consistency of action directions and calculates a synergy score. Simultaneously, the module analyzes the signaling pathways activated by the traditional Chinese medicine components and Western medicine molecules and calculates pathway crosstalk intensity. Combining the analysis results at both the target and pathway levels, the module determines the type of pharmacodynamic interaction. If determined to be high-risk or strongly antagonistic, the module calculates a feedback regulation factor and updates the hyperedge weights of the hypergraph convolution prediction module, achieving closed-loop feedback optimization.

[0087] Step S4: The individualized decision-making module receives patient genotype data, including CYP450 enzyme system gene polymorphism information and drug transporter gene variation information. Based on the precision medicine algorithm, the module predicts the patient's metabolic capacity coefficients for traditional Chinese medicine components and Western medicine molecules. Based on the metabolic capacity coefficients and pharmacodynamic interaction types, the module calculates an individualized medication risk score. When the individualized medication risk score exceeds a second preset threshold, a dosage adjustment plan is automatically generated. The dosage adjustment plan includes suggestions for adjusting the dosage of traditional Chinese medicine and Western medicine. The individualized decision-making module feeds back the actual adverse reaction data to the hypergraph convolution prediction module for incremental model learning.

[0088] Specifically, users upload the patient's genetic testing report or input genotype information for key gene loci. The personalized decision-making module determines the metabolotype classification based on the CYP450 enzyme system genotype, sets the basal metabolic capacity coefficient, and corrects it based on transporter gene variations to obtain the final metabolic capacity coefficient. Combining the outputs of the first two modules, a personalized medication risk score is calculated. When the score exceeds a threshold, the module automatically generates a dosage adjustment plan based on the metabolic capacity coefficient, the magnitude of changes in blood drug concentration, and the type of pharmacodynamic interaction. The plan is presented in a visual report, including the adjusted dosage, medication schedule, indicators to be monitored, and follow-up plan. If adverse reactions occur during patient medication, clinicians enter relevant information, and the system feeds it back to the hypergraph convolutional prediction module, triggering incremental learning to update model parameters.

[0089] The advantages of this invention are as follows: First, by using a quaternary hypergraph network for modeling, it can simultaneously consider the complex relationships between multiple components of traditional Chinese medicine, chemical drugs, metabolic enzymes, and targets, significantly improving the comprehensiveness and accuracy of predictions compared to traditional pairwise relationship analysis. Second, it establishes a closed-loop feedback and synergistic mechanism between pharmacokinetics and pharmacodynamics, enabling the two levels of analysis to promote each other and dynamically optimize, avoiding the limitations of isolated analysis. Third, it integrates patient genotype information to achieve truly individualized decision-making, enabling more precise risk assessment and dosage adjustment compared to group-based medication recommendations. Fourth, through incremental learning from adverse reaction data, the system performance continuously improves, ensuring that optimal predictive capabilities are always maintained.

[0090] To verify the effectiveness of this invention, the research team collected clinical data from 500 patients using a combination of traditional Chinese medicine and Western medicine. Of these, 300 cases were used for model training, and 200 cases were used for independent validation. The validation results showed that the hypergraph convolution prediction module achieved an accuracy of 89.5% in predicting pharmacokinetic interactions, a 13.2 percentage point improvement compared to the traditional graph neural network model (76.3%). The pharmacodynamic synergy assessment module achieved an accuracy of 85.2% in determining the type of pharmacodynamic interaction. After integrating the individualized decision-making module, the system's sensitivity for predicting adverse reactions was 91.7%, its specificity was 87.3%, and its AUC value was 0.923, significantly better than the prediction model that did not consider genotype information (AUC value 0.835).

[0091] In clinical application validation, the research team piloted the system in the cardiology, endocrinology, and oncology departments of three tertiary hospitals. During the pilot period, medication safety assessments and individualized recommendations were provided to 1235 patients using combined traditional Chinese medicine and Western medicine. The system identified 267 high-risk combination regimens, with an adoption rate of 94.0% after confirmation by clinicians. Under the guidance of the system, the incidence of adverse reactions from combined traditional Chinese medicine and Western medicine in the pilot departments decreased from 12.8% to 3.6%, a reduction of 71.9%. Patient satisfaction surveys showed that 89.3% of patients believed that the individualized medication recommendations provided by the system helped improve medication safety, and 93.7% of clinicians believed that the system's predictive results had important reference value for clinical decision-making.

[0092] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart safety assessment system for combined use of traditional Chinese medicine and Western medicine, characterized in that, include: The traditional Chinese medicine component analysis module is used to receive information on traditional Chinese medicine compound prescriptions, break down the traditional Chinese medicine compound prescriptions into individual chemical components based on the traditional Chinese medicine chemical component database, identify the molecular structure and content of each individual chemical component through mass spectrometry analysis technology, and generate a feature vector of traditional Chinese medicine components. The hypergraph convolution prediction module, connected to the traditional Chinese medicine component analysis module, is used to construct a four-element hypergraph network of traditional Chinese medicine component-Western medicine-metabolic enzyme-target. The hyperedges of the hypergraph network connect traditional Chinese medicine monomer component nodes, Western medicine molecule nodes, drug-metabolic enzyme nodes, and target nodes. The hypergraph network is subjected to high-order semantic dependency learning through a hypergraph convolutional neural network. Based on the graph attention mechanism, the competitive inhibition intensity of traditional Chinese medicine component and Western medicine molecule on drug-metabolic enzyme is calculated, and the risk level of changes in Western medicine blood concentration caused by pharmacokinetic interaction is predicted. The pharmacodynamic synergy assessment module, connected to the hypergraph convolution prediction module, is used to acquire the target information of the traditional Chinese medicine components and the Western medicine molecules, and construct a target-pathway dual-layer network. The first layer of the target-pathway dual-layer network is a target interaction network, and the second layer is a signal pathway network. The overlap of the targets and the crosstalk intensity of the pathways between the traditional Chinese medicine components and the Western medicine molecules are calculated through network topology analysis. Based on the overlap, the type of pharmacodynamic interaction of synergistic or antagonistic effects is determined. When the risk level of the pharmacokinetic interaction is higher than a first preset threshold or the type of pharmacodynamic interaction is antagonistic, the pharmacodynamic synergy assessment module generates a risk warning signal and adjusts the prediction parameters of the hypergraph convolution prediction module in reverse. The individualized decision-making module, connected to the pharmacodynamic synergistic evaluation module, receives patient genotype data, including CYP450 enzyme system gene polymorphism information and drug transporter gene variation information. Based on precision medicine algorithms, it predicts the patient's metabolic capacity coefficients for the traditional Chinese medicine components and the Western medicine molecules. Based on the metabolic capacity coefficients and the pharmacodynamic interaction types, it calculates an individualized medication risk score. When the individualized medication risk score exceeds a second preset threshold, it automatically generates a dosage adjustment plan, including traditional Chinese medicine dosage adjustment suggestions and Western medicine dosage adjustment suggestions. The individualized decision-making module feeds back actual adverse reaction data to the hypergraph convolution prediction module for incremental model learning.

2. The intelligent safety assessment system for combined use of traditional Chinese medicine and Western medicine according to claim 1, characterized in that, The traditional Chinese medicine component analysis module includes: The component database unit is used to store chemical component information of traditional Chinese medicine compound prescriptions. The chemical component information includes the name, molecular formula, molecular weight and pharmacological effects of monomer components. The mass spectrometry identification unit is used to perform liquid chromatography-mass spectrometry analysis on the traditional Chinese medicine compound to obtain the mass-to-charge ratio data and retention time data of each monomer chemical component; The feature extraction unit is used to match the mass-to-charge ratio data and the retention time data with the standard data in the component database unit to determine the types and contents of each monomeric chemical component in the traditional Chinese medicine compound and generate the feature vector of the traditional Chinese medicine component.

3. The intelligent safety assessment system for combined use of traditional Chinese medicine and Western medicine according to claim 1, characterized in that, The hypergraph convolution prediction module calculates the competitive suppression strength in the following manner: The molecular docking binding energy between the traditional Chinese medicine monomer component and the drug-metabolizing enzyme was obtained, and the molecular docking binding energy between the Western medicine molecule and the drug-metabolizing enzyme was obtained. The competitive inhibitory intensity of the traditional Chinese medicine monomer component on the drug-metabolizing enzyme by the Western medicine molecule is calculated based on the molecular docking binding energy. The feature information of the traditional Chinese medicine monomer component nodes, the Western medicine molecule nodes, the drug metabolism enzyme nodes and the target nodes are aggregated by hypergraph convolution operation, and the embedding representation of each node is updated. The change in the blood concentration of the Western medicine is predicted based on the updated node embedding representation.

4. The intelligent safety assessment system for combined use of traditional Chinese medicine and Western medicine according to claim 1, characterized in that, The pharmacodynamic synergy assessment module determines the type of pharmacodynamic interaction in the following ways: The intersection ratio of the set of targets of the traditional Chinese medicine components and the set of targets of the Western medicine molecules is calculated to obtain the target overlap. When the overlap of the target points is greater than a third preset threshold, the direction of the agonistic or antagonistic effect of the traditional Chinese medicine components and the Western medicine molecules on the common target is analyzed. If the directions of action are consistent, it is determined to be a synergistic effect type; if the directions of action are opposite, it is determined to be a mutually antagonistic type. The crosstalk intensity between the signaling pathway activated by the traditional Chinese medicine component and the signaling pathway activated by the Western medicine molecule is calculated, and the determination result of the pharmacodynamic interaction type is corrected when the crosstalk intensity exceeds a fourth preset threshold.

5. The intelligent safety assessment system for combined use of traditional Chinese medicine and Western medicine according to claim 1, characterized in that, The individualized decision-making module calculates the metabolic capacity coefficient in the following manner: Genotyping of the CYP450 enzyme system gene in the patient was performed to determine whether the patient was a rapid metabolizer, intermediate metabolizer, or slow metabolizer. Based on the genotyping results, a basal metabolic rate coefficient was set: 1.5-2.0 for rapid metabolizers, 0.8-1.2 for intermediate metabolizers, and 0.3-0.7 for slow metabolizers. The basal metabolic capacity coefficient is corrected based on the drug transporter gene variation information to obtain the final metabolic capacity coefficient.

6. The intelligent safety assessment system for combined use of traditional Chinese medicine and Western medicine according to claim 1, characterized in that, The dosage adjustment scheme is generated as follows: The dose adjustment factor of the drug molecule is calculated based on the metabolic capacity coefficient, and the dose adjustment factor is inversely proportional to the metabolic capacity coefficient. The direction of dosage adjustment of the traditional Chinese medicine is determined according to the type of pharmacodynamic interaction. When the type of pharmacodynamic interaction is synergistic, the dosage of the traditional Chinese medicine is reduced. When the type of pharmacodynamic interaction is antagonistic, the compatibility of the traditional Chinese medicine compound is adjusted. The dosage adjustment plan is generated by combining the pharmacokinetic interaction risk level, the pharmacodynamic interaction type, and the individualized medication risk score.

7. The intelligent safety assessment system for combined use of traditional Chinese medicine and Western medicine according to claim 1, characterized in that, The hypergraph convolutional prediction module uses a graph attention mechanism to calculate the attention weights between nodes. The attention weights represent the contribution of different nodes to the prediction results. The hypergraph convolutional prediction module performs weighted aggregation of node features based on the attention weights.

8. The intelligent safety assessment system for combined use of traditional Chinese medicine and Western medicine according to claim 1, characterized in that, Also includes: The real-time monitoring module is used to collect the patient's physiological indicators and clinical performance data in real time during medication. The physiological indicators include blood pressure, heart rate, and liver and kidney function indicators. When abnormal physiological indicators or adverse reactions are detected, an early warning message is immediately sent to the individualized decision-making module. The individualized decision-making module responds to the early warning information, reassesses the individualized medication risk score, and generates emergency treatment suggestions.

9. The intelligent safety assessment system for combined use of traditional Chinese medicine and Western medicine according to claim 1, characterized in that, The incremental learning of the model is achieved in the following way: Collect adverse reaction case data that occur in actual clinical applications, including patient information, medication regimen, adverse reaction type and severity; The adverse reaction case data are added as new training samples to the training dataset of the hypergraph convolution prediction module; The model parameters of the hypergraph convolutional neural network are updated using an online learning algorithm to improve the prediction accuracy of the hypergraph convolutional prediction module for similar combinations of traditional Chinese medicine and Western medicine.

10. A method for intelligently assessing the safety of combined use of traditional Chinese medicine and Western medicine, employing the intelligent safety assessment system for combined use of traditional Chinese medicine and Western medicine as described in any one of claims 1-9, characterized in that, Includes the following steps: The TCM component analysis module receives TCM compound information, and based on the TCM chemical component database, it breaks down the TCM compound into individual chemical components. Mass spectrometry analysis technology is used to identify the molecular structure and content of each individual chemical component, and a TCM component feature vector is generated. A quaternary hypergraph network consisting of traditional Chinese medicine components, Western medicine, drug-metabolizing enzymes, and targets is constructed using a hypergraph convolution prediction module. The hyperedges of the hypergraph network connect nodes of traditional Chinese medicine monomer components, nodes of Western medicine molecules, nodes of drug-metabolizing enzymes, and nodes of targets. The hypergraph network is subjected to high-order semantic dependency learning through a hypergraph convolutional neural network. Based on the graph attention mechanism, the competitive inhibition intensity of traditional Chinese medicine components and Western medicine molecules on drug-metabolizing enzymes is calculated, and the risk level of changes in Western medicine blood concentration caused by pharmacokinetic interactions is predicted. The pharmacodynamic synergy assessment module obtains the target information of the traditional Chinese medicine components and the Western medicine molecules, and constructs a target-pathway dual-layer network. The first layer of the target-pathway dual-layer network is a target interaction network, and the second layer is a signal pathway network. The target overlap and pathway crosstalk intensity of the traditional Chinese medicine components and Western medicine molecules are calculated through network topology analysis. Based on the overlap, the type of pharmacodynamic interaction of synergistic or antagonistic effects is determined. When the risk level of the pharmacokinetic interaction is higher than a first preset threshold or the type of pharmacodynamic interaction is antagonistic, the pharmacodynamic synergy assessment module generates a risk warning signal and adjusts the prediction parameters of the hypergraph convolution prediction module in reverse. The personalized decision-making module receives patient genotype data, including CYP450 enzyme system gene polymorphism information and drug transporter gene variation information. Based on precision medicine algorithms, it predicts the patient's metabolic capacity coefficients for the traditional Chinese medicine components and the Western medicine molecules. Based on the metabolic capacity coefficients and the pharmacodynamic interaction types, it calculates a personalized medication risk score. When the personalized medication risk score exceeds a second preset threshold, it automatically generates a dosage adjustment plan, which includes traditional Chinese medicine dosage adjustment suggestions and Western medicine dosage adjustment suggestions. The personalized decision-making module feeds back actual adverse reaction data to the hypergraph convolution prediction module for incremental model learning.

Images