Traditional Chinese medicine formula screening method and system based on target point structure distribution
Through the traditional Chinese medicine prescription screening method based on the target structure distribution, an ultra-graph structure was generated and prescription screening was carried out, and the problems of inefficient screening of traditional Chinese medicine prescriptions in the prior art were solved, and efficient and interpretable Chinese medicine prescription screening was achieved.
Patent Information
- Application Number
- CN202510054989.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-14
- Publication Date
- 2025-05-16
AI Technical Summary
The prior art is difficult to effectively explain the effectiveness and safety of traditional Chinese medicine prescriptions at the target level, especially when dealing with multiple traditional Chinese medicine combinations, and cannot adapt to the challenges of complexity of traditional Chinese medicine.
The traditional Chinese medicine prescription screening method based on the target structure distribution is used to generate hypergraph structures by obtaining relevant targets for specific diseases, map the prescription to be screened into the hypergraph structure, and use the trained prescription screening model for screening.
It significantly improves the speed and interpretability of traditional Chinese medicine prescription screening, can effectively complete the prescription screening task within a few minutes, enhances the interpretability of the traditional Chinese medicine treatment process, and has good scalability.
Smart Images
Figure CN120015244A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent optimization of traditional Chinese medicine prescriptions, and more specifically to a method and system for screening traditional Chinese medicine prescriptions based on target structure distribution. Background Art
[0002] At present, TCM experts rely on traditional diagnostic methods such as observation, smell, questioning, and palpation combined with patients' medical records for comprehensive evaluation, and select the best TCM prescription based on long-term accumulated TCM practice experience. However, this classic method is mainly limited to the study of disease symptoms, and it is difficult to deeply explore the complexity of TCM ingredients and their mechanisms of action, resulting in limitations in explaining the effectiveness and safety of TCM prescriptions at the target level.
[0003] However, with the development of protein-protein interaction (PPI) networks and the successful application of drug efficacy evaluation methods based on this network distance in the field of Western medicine, new ideas have been provided for the study of traditional Chinese medicine. Nevertheless, since traditional Chinese medicine usually involves a large number of effective targets, far more than Western medicine, it is extremely difficult to directly apply existing PPI network distance-based methods to the screening of traditional Chinese medicine prescriptions. In addition, these methods are inefficient in dealing with multiple combinations of traditional Chinese medicines because they rely too much on the topological features between the drug and disease target sets, limiting their adaptability to the complexity of traditional Chinese medicines.
[0004] Therefore, developing a method that can overcome the above challenges and is suitable for large-scale screening of TCM prescriptions, especially a method that can optimize TCM prescriptions based on target structure distribution and improve their interpretability at the target level, is an issue that technical personnel in this field urgently need to solve. Summary of the invention
[0005] In view of this, the present invention provides a method and system for screening Chinese medicine prescriptions based on target structure distribution, which overcomes the above-mentioned defects.
[0006] In order to achieve the above object, the present invention adopts the following technical solution:
[0007] A method for screening Chinese medicine prescriptions based on target structure distribution, comprising the following specific steps:
[0008] Acquire relevant targets of a specific disease, and generate a corresponding hypergraph structure based on the relevant targets of the specific disease;
[0009] Acquire a formula to be screened, map the formula to be screened into the hypergraph structure, and obtain a mapped formula to be screened;
[0010] The mapped formula to be screened is input into the trained formula screening model to obtain a screening formula.
[0011] Optionally, the step of acquiring the hypergraph structure is:
[0012] Analyze the relevant targets of a specific disease by functional enrichment, and obtain a target adjacency matrix based on the relevant targets;
[0013] A plurality of target point communities are generated based on the target point adjacency matrix according to the community partition model, and the plurality of target point communities are optimized to generate a hypergraph structure.
[0014] Optionally, the specific steps for optimizing the target community are:
[0015] Utilizing Gaussian distribution and an extended density clustering algorithm to simplify multiple groups of target communities and generate a simplified hypergraph structure;
[0016] The simplified hypergraph structure is filtered to obtain a hypergraph structure.
[0017] Optionally, a model based on the difference between dynamic and static representations may be used to filter the simplified hypergraph structure.
[0018] Optionally, the training steps of the prescription screening model are:
[0019] Constructing a prescription data set including effective Chinese medicine prescriptions and random Chinese medicine prescriptions, and mapping the prescription data to the hypergraph structure to obtain the mapped data set;
[0020] A neural network model is constructed, and the neural network model is trained using the mapping data set to obtain the prescription screening model.
[0021] Optionally, in the prescription data set, the effective Chinese medicine prescription is used as a positive sample, and the random Chinese medicine prescription is used as a negative sample.
[0022] Optionally, when screening the prescription, it is also necessary to judge whether the prescription is effective based on a preset threshold.
[0023] A Chinese medicine formula screening system based on target structure distribution, comprising:
[0024] A hypergraph structure generation module, used to obtain relevant targets of a specific disease and generate a corresponding hypergraph structure based on the relevant targets of the specific disease;
[0025] A mapping module, used for obtaining a formula to be screened, mapping the formula to be screened into the hypergraph structure, and obtaining a mapped formula to be screened;
[0026] The formula screening module inputs the mapped formula to be screened into the trained formula screening model to obtain a screened formula.
[0027] It can be seen from the above technical solutions that the present invention provides a method and system for screening Chinese medicine prescriptions based on target structure distribution, which has the following beneficial effects compared with the prior art:
[0028] Greatly improved screening speed: The present invention solves the problem of computational bottlenecks and can efficiently complete the task of formula screening within a few minutes, greatly accelerating the screening process of traditional Chinese medicine formulas.
[0029] Enhanced interpretability and expansibility: The present invention combines the macroscopic and microscopic perspectives of the TCM treatment process, which not only links the target structure at the microscopic level with the drug efficacy at the macroscopic level, but also provides good interpretability for the mechanism of drug treatment of diseases. The present invention allows the results to be analyzed according to the degree of influence of various factors in the treatment process of different diseases, and even in the case of insufficient positive sample prescription data, it can screen out potential effective prescriptions, providing a basis for subsequent more detailed medical experimental verification, thereby further enhancing the model's ability to interpret more prescriptions. In addition, the method has good expansibility and can adapt to the growing demand for TCM prescriptions and new scientific research results. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0031] Figure 1 The present invention provides a flow chart of the method. DETAILED DESCRIPTION
[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 creative work are within the scope of protection of the present invention.
[0033] On the one hand, an embodiment of the present invention discloses a method for screening Chinese medicine prescriptions based on target structure distribution, which divides the overall generation model into two parts: generating a streamlined and reliable hypergraph structure and using the hypergraph structure and prescription efficacy to train a prescription evaluation network. First, the community relationship divided by the graph neural network model is used to preliminarily express the target micro information, the Gaussian distribution is used to eliminate redundant information, the extended density clustering algorithm is used to obtain the streamlined hypergraph structure, and the neural network model based on the representation difference is used to obtain the reliable hypergraph structure; the prescription data set generated by the random combination of existing prescriptions and commonly used Chinese medicines for the disease is constructed, and then the prescription data set is mapped to the hypergraph structure, and a prescription evaluation network is trained to evaluate the quality of different prescriptions. Finally, and are used in combination to evaluate the efficacy of other prescriptions on the disease; when applied, it is necessary to collect protein target information of a specific disease in advance. Currently, there is target information of existing prescriptions and target information of commonly used Chinese medicines, such as Figure 1 As shown, the specific steps are:
[0034] Step 1: Obtain relevant targets of a specific disease, and generate a corresponding hypergraph structure based on the relevant targets of the specific disease; the relevant targets include protein target information of the specific disease and target information of commonly used traditional Chinese medicines for the disease;
[0035] Step 2: Obtain the formula to be screened, map the formula to be screened into the hypergraph structure, and obtain the mapped formula to be screened;
[0036] Step 3: Input the mapped prescription to be screened into the trained prescription screening model to obtain the screening prescription.
[0037] In one embodiment, the steps of obtaining the hypergraph structure are:
[0038] Step 11, analyzing the relevant targets of a specific disease through functional enrichment, and obtaining a target adjacency matrix based on the relevant targets;
[0039] Step 12: Generate multiple groups of target communities based on the target adjacency moments according to the community partition model, and optimize the multiple groups of target communities to generate a hypergraph structure.
[0040] Furthermore, the targets significantly expressed in the corresponding specific diseases are collected through the functional enrichment method; and the target adjacency matrix is obtained from the collected targets; then the community partition model is constructed using the graph neural network (GCN), and the community partition model is used to generate K groups of target communities; where the input of the graph neural network is the connection of different subgraphs of the original protein interaction network, and the output is the partition of the target K groups of communities in the protein interaction network. The graph neural network is an algorithm that extends the convolution algorithm to the field of ordinary graphs, and obtains a new representation of nodes by convolving node neighborhood information.
[0041] In one embodiment, the specific steps of optimizing the target community are:
[0042] Step 121, using Gaussian distribution and extended density clustering algorithm to simplify multiple groups of target communities and generate a simplified hypergraph structure;
[0043] Step 122: Filter the simplified hypergraph structure based on the difference to obtain the hypergraph structure.
[0044] In one embodiment, a model based on the difference between dynamic and static representations may be used to filter the simplified hypergraph structure.
[0045] Furthermore, firstly, the redundant community information is simplified by using Gaussian distribution, and the extended density clustering algorithm f 2 Generate a simplified hypergraph structure H; then use the neural network model f based on representation difference 3 Filter and simplify the hypergraph structure H to obtain a credible hypergraph structure H′; the neural network model based on representation difference uses a multi-head attention mechanism. When the dynamic representation and static representation are very different, the corresponding score S i It will be higher, retaining some hyperedges with higher scores while retaining as many targets as possible.
[0046] In one embodiment, the steps of training the prescription screening model are:
[0047] Step 31, constructing a prescription data set including effective Chinese medicine prescriptions and random Chinese medicine prescriptions, and mapping the prescription data to a hypergraph structure to obtain a mapping data set;
[0048] Step 32: construct a neural network model, and use the mapping data set to train the neural network model to obtain a prescription screening model.
[0049] In one embodiment, in the prescription data set, effective Chinese medicine prescriptions are used as positive samples, and random Chinese medicine prescriptions are used as negative samples.
[0050] In one embodiment, when screening the prescription, it is also necessary to judge whether the prescription is effective based on a preset threshold.
[0051] Furthermore, we first use existing reliable prescriptions or prescriptions recommended by experts (i.e., effective Chinese medicine prescriptions) as positive samples, and random combinations of Chinese medicines commonly used in the disease as negative samples to construct a prescription dataset; and then map these Chinese medicine prescriptions to a concise and reliable hypergraph structure; then build a neural network model f 4 , trained with the mapped data set, is used to evaluate the quality of Chinese medicine prescriptions. The output of the model is an evaluation score of the prescription. A threshold τ is set through the validation set. The judgment basis is that if the model output value is higher than the threshold, the prescription is effective, and if it is lower than the threshold, the prescription is invalid.
[0052] In the above embodiment, a hypergraph is an information-intensive representation method. An edge of a normal graph connects two vertices, while an edge of a hypergraph can connect any number of vertices. In the process of Chinese medicine prescriptions acting on disease targets, the targets can participate in a variety of different functions, and the same target plays different roles in different functions. Introducing a hypergraph structure to model the Chinese medicine-disease process is more in line with the actual situation of Chinese medicine.
[0053] In this embodiment, the disclosed method has the following specific steps: first, functional enrichment of disease targets and existing prescription targets is performed to obtain functions significantly expressed in the disease process, targets related to the task are selected and their adjacency matrix A is obtained. Then, the adjacency matrix A is used as input and the GCN network model f is used to construct the adjacency matrix A. 1 Obtain the community relationship C of these targets. Take the target community relationship C as input, and use Gaussian distribution and extended density clustering algorithm f 2 Obtain a simplified hypergraph structure H, and then use the representation difference-based model f 3 Verify and obtain a reliable hypergraph structure H'. Finally, train a prescription evaluation network model f 4 To learn the advantages of existing Chinese medicine prescriptions, in this process, the network is constrained by streamlining the credible hypergraph structure H' to prevent overfitting. The trained network can be used to quickly screen a large number of potential Chinese medicine prescriptions, and several Chinese medicine prescriptions with higher scores can be retained for further testing.
[0054] The prior art has obvious deficiencies in distinguishing between random combinations of commonly used Chinese medicines and existing effective prescriptions. For diseases that have not been widely studied, taking Parkinson's disease as an example, the accuracy rate can usually only reach 50%-60%. For example, the method based on PPI network distance, the method based on K-means clustering, and the method based on spectral clustering have accuracy rates of 50%, 50%, and 61.1%, respectively; while the accuracy rate of the method in this embodiment reaches 88.9%; for diseases that have been deeply studied, taking stroke as an example, the method based on PPI network distance, the method based on K-means clustering, and the method based on spectral clustering have accuracy rates of 55.6%, 66.7%, and 72.2%, respectively, while the method disclosed in this embodiment reaches an accuracy rate of 100%. It can be seen that the method in this embodiment shows excellent performance in testing the effectiveness of prescriptions and significantly improves the accuracy rate.
[0055] Based on the hypergraph structure H′ generated based on the structural characteristics of the protein target, the present invention maps the formula to the hypergraph structure to connect the macroscopic and microscopic characteristics of the drug-disease process, and uses the neural network model f 4 Learn some relevant weights to achieve efficient and accurate evaluation of the efficacy of the prescription.
[0056] On the other hand, this embodiment also discloses a Chinese medicine formula screening system based on target structure distribution, including:
[0057] A hypergraph structure generation module, used to obtain relevant targets of a specific disease and generate a corresponding hypergraph structure based on the relevant targets of the specific disease;
[0058] A mapping module, used to obtain the formula to be screened, map the formula to be screened into a hypergraph structure, and obtain the mapped formula to be screened;
[0059] The formula screening module inputs the mapped formula to be screened into the trained formula screening model to obtain the screened formula.
[0060] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0061] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for screening Chinese medicine prescriptions based on target structure distribution, characterized in that: The specific steps are: Acquire relevant targets of a specific disease, and generate a corresponding hypergraph structure based on the relevant targets of the specific disease; Acquire a formula to be screened, map the formula to be screened into the hypergraph structure, and obtain a mapped formula to be screened; The mapped formula to be screened is input into the trained formula screening model to obtain a screening formula.
2. A method for screening Chinese medicine prescriptions based on target structure distribution according to claim 1, characterized in that: The steps for obtaining the hypergraph structure are: Analyze the relevant targets of a specific disease by functional enrichment, and obtain a target adjacency matrix based on the relevant targets; A plurality of target point communities are generated based on the target point adjacency matrix according to the community partition model, and the plurality of target point communities are optimized to generate a hypergraph structure.
3. A method for screening Chinese medicine prescriptions based on target structure distribution according to claim 2, characterized in that: The specific steps for optimizing the target community are: Utilizing Gaussian distribution and an extended density clustering algorithm to simplify multiple groups of target communities and generate a simplified hypergraph structure; The simplified hypergraph structure is filtered to obtain a hypergraph structure.
4. A method for screening Chinese medicine prescriptions based on target structure distribution according to claim 3, characterized in that: The simplified hypergraph structure may be filtered using a model based on the difference between dynamic and static representations.
5. A method for screening Chinese medicine prescriptions based on target structure distribution according to claim 1, characterized in that: The training steps of the prescription screening model are: Constructing a prescription data set including effective Chinese medicine prescriptions and random Chinese medicine prescriptions, and mapping the prescription data to the hypergraph structure to obtain the mapped data set; A neural network model is constructed, and the neural network model is trained using the mapping data set to obtain the prescription screening model.
6. A method for screening Chinese medicine prescriptions based on target structure distribution according to claim 5, characterized in that: In the prescription data set, the effective Chinese medicine prescription is used as a positive sample, and the random Chinese medicine prescription is used as a negative sample.
7. A method for screening Chinese medicine prescriptions based on target structure distribution according to claim 1, characterized in that: When screening prescriptions, it is also necessary to judge whether the prescription is effective based on the preset threshold.
8. A Chinese medicine formula screening system based on target structure distribution, characterized in that: include: A hypergraph structure generation module, used to obtain relevant targets of a specific disease and generate a corresponding hypergraph structure based on the relevant targets of the specific disease; A mapping module, used for obtaining a formula to be screened, mapping the formula to be screened into the hypergraph structure, and obtaining a mapped formula to be screened; The formula screening module inputs the mapped formula to be screened into the trained formula screening model to obtain a screened formula.