基于人工神经网络模型的黄芩提取物性状-成分与药效相关性研究

目的探索将性状指标(颜色与味道)引入到黄芩提取物质量评价体系中,建立全面科学的黄芩提取物质量评价方法。方法参考《中国药典》2015年版一部方法制备各种黄芩提取物样品,分别测定样品的性状(颜色和味道)、成分(4种黄酮类成分黄芩苷、汉黄芩苷、黄芩素、汉黄芩素)的含量及药效(体外抑菌率),采用人工神经网络(ANN)分别建立以下模型:颜色-抑菌率ANN模型、味道-抑菌率ANN模型、成分-抑菌率ANN模型、颜色/味道-抑菌率ANN模型、颜色/成分-抑菌率ANN模型、味道/成分-抑菌率ANN模型、颜色/味道/成分-抑菌率ANN模型,通过比较以上7种模型的预测能力,探索黄芩提取物整体质量评价方法。结果采用颜色、味道与成分含量3种指标相结合与体外抑菌率所建立的ANN模型训练和预测能力最好,其中r2最高(r2=0.92),均方根误差(RMSE)最低(RMSE=3.54),说明颜色、味道与成分含量3种指标之间存在信息互补,3种指标相结合能快速准确地预测体外抑菌率。结论本研究证实将性状指标引入到现有的黄芩提取物质量评价体系中,性状结合成分含量指标能全面快速评价黄芩提取物的整体质量。

Correlation between features-composition and pharmacodynamics of Scutellaria baicalensis extracts based on ANN model

Objective To establish a comprehensive scientific evaluation method for the quality of Scutellaria baicalensis by introducing the feature indicators (color and taste) into the quality evaluation system of S. baicalensis extracts (SBE) in this study. Methods Various samples of SBE were prepared according to Chinese Pharmacopoeia (2015 version) process, the features (color and taste), components (the content of four flavonoids:baicalin, wogonoside, baicalein and wogonin), and efficacy (inhibition rate in vitro) of the samples were determined. The following models were established by Artificial Neural Network (ANN) model:color-inhibition rate model, taste-inhibition rate model, composition-inhibition rate model, color/taste-inhibition rate model, color/component-inhibition rate model, taste/component-inhibition rate model and color/taste/component-inhibition rate model. The overall quality assessment for SBE was explored by comparing the predictive power of the above seven models. Results The color/taste/component-inhibition rate ANN model (r²=0.92, RMSE=3.96) showed the best training and prediction ability. The value of r² was the highest and RMSE was the lowest, which indicated there was information complementation among the three indicators and the combination of three indicators can quickly and accurately predict the inhibition rate in vitro for SBE. Conclusion This study confirmed that feature indicators combined with component indicator can comprehensively and quickly evaluate the overall quality of SBE.