CN-122025030-A - Method, equipment, medium and product for generating traditional Chinese medicine prescription

CN122025030ACN 122025030 ACN122025030 ACN 122025030ACN-122025030-A

Abstract

The application discloses a traditional Chinese medicine prescription generation method, equipment, medium and product, and relates to the field of artificial intelligence; constructing a traditional Chinese medicine prescription generation network based on an attention mechanism, a graph convolution network and reinforcement learning according to the multi-source heterogeneous data; the traditional Chinese medicine prescription generation network is utilized to generate traditional Chinese medicine prescriptions for the acquired syndromes and symptoms. The application can realize the intelligent and personalized generation of the traditional Chinese medicine prescription and improve the accuracy and rationality of the traditional Chinese medicine diagnosis and treatment.

Inventors

Niu Qijie
ZHANG HUAMIN
LI BING
ZONG WENJING
WANG JINGAI
TONG LIN
ZENG ZILING
TIAN SIWEI

Assignees

中国中医科学院中医基础理论研究所

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (10)

1. A method for generating a traditional Chinese medicine prescription is characterized by comprising the following steps: The method comprises the steps of obtaining multi-source heterogeneous data, wherein the multi-source heterogeneous data comprises disease target point data, targets corresponding to symptoms and symptoms, traditional Chinese medicine and action targets, protein interaction networks and traditional Chinese medicine co-occurrence information in historical formulas; The traditional Chinese medicine prescription generation network comprises an attention mechanism module, a graph convolution network module and a reinforcement learning module, wherein the attention mechanism module is used for determining importance scores of each traditional Chinese medicine based on attention mechanism according to target points, traditional Chinese medicines, action targets and protein interaction networks corresponding to disease target point data, symptoms and symptoms, based on the attention mechanism, to obtain traditional Chinese medicine importance probability distribution, the graph convolution network module is used for constructing a traditional Chinese medicine-traditional Chinese medicine co-occurrence network by utilizing traditional Chinese medicine co-occurrence information in the historical prescription based on the graph convolution network, and calculating a compatibility intensity score between any two traditional Chinese medicines, and the reinforcement learning module is used for selecting basis based on a reinforcement learning framework by taking the traditional Chinese medicine importance probability distribution as an initial state, taking the compatibility intensity score as a reward signal, making a sequence decision in a traditional Chinese medicine space by an intelligent body, comprehensively evaluating and sequencing the candidates, and outputting an optimal traditional Chinese medicine prescription; the traditional Chinese medicine prescription generation network is utilized to generate traditional Chinese medicine prescriptions for the acquired syndromes and symptoms.
2. The method of claim 1, wherein the obtaining multi-source heterogeneous data further comprises: and cleaning and standardizing the multi-source heterogeneous data.
3. The method for generating a Chinese medicinal prescription according to claim 1, wherein the processing procedure of the attention mechanism module specifically comprises: mapping the disease target data into a disease attention weight vector by adopting a target space mapping technology; adopting a target space mapping technology to encode syndrome and symptom target data into symptom association vectors; Converting the topological center index of the traditional Chinese medicine and the action target point in the protein interaction network into a topological importance vector; Carrying out multidimensional interactive calculation on the disease attention weight vector, the symptom association vector and the topological importance vector by utilizing a multi-head attention mechanism to obtain a characteristic interaction result; And carrying out target point quantity normalization processing on the characteristic interaction result, and carrying out Softmax function normalization to obtain the importance probability distribution of the traditional Chinese medicine.
4. The method for generating a Chinese medicinal prescription according to claim 3, wherein the normalizing the target number of the feature interaction result and normalizing the target number by a Softmax function to obtain a probability distribution of importance of the Chinese medicinal prescription comprises: using the formula Determining an importance score H of each traditional Chinese medicine; Wherein V is a disease attention weight vector, Q is a symptom association vector, K is a topological importance vector, Is the total number of target points of the traditional Chinese medicine.
5. The method for generating a Chinese medicinal formulation according to claim 3, wherein the step of converting the topological center index of the Chinese medicinal formulation and the action target in the protein interaction network into the topological importance vector comprises the following steps: according to the protein interaction network, a PageRank algorithm is adopted to determine the topological centrality index.
6. The method for generating a Chinese medicinal prescription according to claim 1, wherein the Chinese medicinal composition-Chinese medicinal composition co-occurrence network is an undirected weighted graph, wherein nodes in the undirected weighted graph represent Chinese medicaments, and the side weights represent total times of co-occurrence of two Chinese medicaments and Chinese medicaments in the historical prescription.
7. The method of claim 1, wherein the reinforcement learning framework uses Q-learning algorithm, and wherein the reinforcement learning is performed using epsilon-greedy strategy.
8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the method of generating a chinese medicinal prescription according to any one of claims 1-7.
9. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed by a processor, implements the method for generating a chinese medicinal prescription according to any one of claims 1-7.
10. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements the method of generating a chinese medicinal prescription according to any one of claims 1-7.

Description

Method, equipment, medium and product for generating traditional Chinese medicine prescription Technical Field The application relates to the field of artificial intelligence, in particular to a method, equipment, medium and product for generating a traditional Chinese medicine prescription. Background Traditional Chinese medicines are accumulated in long-term medical practice of people and are important components of ancient excellent cultural heritage. Through long-term medical practice, a plurality of medicines are matched and decocted to prepare soup, so that a traditional Chinese medicine prescription is formed. In the traditional cases, the treatment effect of the patient depends on the experience of the middle doctor, and more reasonable traditional Chinese medicine formulas can be used by the patient in many cases, but the middle doctor cannot want due to the experience limitation, and a system for providing a plurality of traditional Chinese medicine formulas for the middle doctor to select is not available. In recent years, with the rapid development of artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) technology, its application in the field of traditional Chinese medicine has gradually become a research hotspot. Particularly in the aspect of generating and optimizing traditional Chinese medicine prescriptions, researchers try to introduce methods such as machine learning, deep learning and the like into the dialectical treatment process of traditional Chinese medicine so as to realize intelligent and personalized prescription recommendation. However, the prior art solutions still have significant shortcomings in terms of theoretical compliance, model interpretability, clinical practicality, etc. Firstly, most of the existing AI auxiliary prescription generation methods only take "diseases" as centers, and the prescription recommendation (such as a method based on network pharmacology or knowledge graph) is performed by constructing a mapping relation of "diseases-targets-traditional Chinese medicines". Such methods ignore the differences between syndromes and symptoms. Secondly, the existing model generally lacks the deep modeling capability of the compatibility rule of the traditional Chinese medicines, cannot effectively capture the complex nonlinear synergistic relationship among the traditional Chinese medicines, and cannot distinguish the drug pairs of high-frequency general drugs (such as liquorice) and drugs with specific synergistic effects, so that the accuracy and the innovation of the formula are affected. Furthermore, currently mainstream AI models (e.g., deep neural networks) generally suffer from "black box" problems, and the decision process lacks interpretability. In the field of traditional Chinese medicine, which is highly dependent on experience and logic reasoning, doctors and patients often need to know "why a certain drug is selected" or "why the drug pair is recommended". If the model can not provide transparent and traceable scoring basis, clinical trust is difficult to obtain, and the basic principle of Chinese medicine 'principle and method medicine' consistency is also violated. In addition, data noise and multi-source heterogeneity further exacerbate model training instability. The traditional Chinese medicine has wide data sources, including ancient books, modern clinical records and databases (such as TCMSP, ETCM, symMap, etc.), and the terms are expressed unevenly, target points are annotated unevenly, and the labeling quality of samples is uneven. If systematic cleaning, standardization and structuring are not performed on the data, the direct input of the data into the model will cause characteristic deviations, thereby affecting the reliability of prescription generation. Aiming at the problems, an intelligent prescription generation method which can deeply integrate the syndrome differentiation logic of the traditional Chinese medicine, effectively model the traditional Chinese medicine compatibility synergistic mechanism and has good interpretability and anti-noise capability is needed. Disclosure of Invention The application aims to provide a method, equipment, medium and product for generating a traditional Chinese medicine prescription, which can realize intelligent and personalized generation of the traditional Chinese medicine prescription and improve the accuracy and rationality of traditional Chinese medicine diagnosis and treatment. In order to achieve the above object, the present application provides the following solutions: In a first aspect, the present application provides a method for producing a traditional Chinese medicine formulation, the method comprising: The method comprises the steps of obtaining multi-source heterogeneous data, wherein the multi-source heterogeneous data comprises disease target point data, targets corresponding to symptoms and symptoms, traditional Chinese medicine and action targets, protein interaction networks and