Construction and Application of a Traditional Chinese Medicine Component Database Based on Deep Learning
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
To address the challenges of integrating heterogeneous, multi-source data and low component identification accuracy for traditional Chinese medicine (TCM) components, this paper constructs a multi-source fusion TCM component database and proposes a TCM Component Feature Adaptive Fusion Algorithm (TCFEA). The database integrates authoritative data from the Chinese Pharmacopoeia and other sources with laboratory HPLC-MS and XRD data, totaling 12,000 records (500 TCMs) and partitioned into training/validation/test sets with an 8:1:1 ratio. The TCFEA algorithm extracts multi-source features using BERT-BiLSTM, dynamic convolution, and GCN. It then combines multi-head attention with residual connections for adaptive fusion, and introduces a modified Focal Loss to address data imbalance. Experiments show that TCFEA achieves a test set accuracy of 96.8%, a 5.2 percentage point improvement over Attention-Fusion, with an F1-Score of 95.7% and an AUC of 0.982. It also achieves an 88.7% accuracy for identifying trace components less than 0.1 mg/g, an 18.5 percentage point improvement over XGBoost. In terms of efficiency, training time is 8.2 hours (1.8 hours shorter than CNN-LSTM), and the inference speed is 280 entries per second. The database prototype system averages 0.5 seconds per sample for drug testing, achieving an unknown component prediction accuracy of 92.1%. This research provides efficient data and algorithmic support for the analysis of traditional Chinese medicine components, promoting the modernization of traditional Chinese medicine research.
Fields of Study & Subjects
Traditional Chinese Medicine Analysis
Computational Drug Discovery Methods
Advanced Neural Network Applications
Publication details
- Journal / source
- IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
- Pages
- 178–183
- Publication date
- 28 January 2026
- Language
- EN