Quantitative study on medicinal properties of traditional Chinese medicine based on BP neural network
10.7501/j.issn.0253-2670.2020.16.023
- VernacularTitle: 基于多层前馈神经网络的中药药性量化研究
- Author:
Le DENG
1
Author Information
1. School of Informatics and Engineering, Hunan University of Chinese Medicine
- Publication Type:Journal Article
- Keywords:
BP neural network;
Drug vector;
Efficacy;
Medicinal property;
Quantification of herbal medicine attributes
- From:
Chinese Traditional and Herbal Drugs
2020;51(16):4277-4283
- CountryChina
- Language:Chinese
-
Abstract:
Objective: It is difficult to accurately grasp the essential characteristics of medicinal properties of traditional Chinese medicine due to the abstraction and vagueness. This paper proposes a Quantitative Model of Traditional Chinese Medicine's Properties based on BP Neural Network (QM-BP Model) to train and realize quantitative representations of Chinese herbal medicine (CHM). Methods: Data for analysis were obtained and organized by conceptual analysis. Sample pairs of the associations were obtained based on the relationships of CHM and their efficacy. Then a QM-BP model with three-tier structure in form of CHM-drug vector-efficacy was constructed, initialized and trained according to prior organized CHM data. Finally, rules of correlation of CHM and their efficacy was obtained by training dataset with drug vectors representing quantitative attributes of CHM. Results: Based on the training of QM-BP model, 474 TCM and 528 effects included in the textbook of TCM were trained and combined based on the training of QM-BP model. It was found that the BP drug vectors representing drug properties after training reflected the attribute characteristics of CHM better than the initial quantitative values. In addition, as BP drug vector and word vector have similar properties, the BP drug vectors for CHM with similar efficacy was relatively close in Euclidean distance while the CHM with different efficacies were relatively far in Euclidean distance. Conclusion: In this paper, a BP neural network was adopted to construct a medicine vector training model. Based on the correlation between the medicinal properties and efficacy of TCM, the quantified values of the medicinal properties were modified to represent medicinal properties more accurately. In future work, the QM-BP model can be applied to the analysis of herb pairs and prescriptions to analyze the rules of combination related to medicinal properties and the compatibility within prescriptions.
