随机森林算法在中药指纹图谱中的应用:以不同品牌夏桑菊颗粒指纹图谱分析为例

该研究旨在建立随机森林算法鉴别和分类不同品牌夏桑菊颗粒,为多指标的复杂指纹图谱的鉴别提供有效的参考。采用高效液相法采集83批不同品牌的夏桑菊颗粒指纹图谱,比较主成分分析、偏最小二乘法-判别分析、随机森林等方法在处理不同分类样品复杂数据时的不同。结果表明本研究成功建立了83批不同品牌夏桑菊颗粒的指纹图谱;经过不同模式识别方法比较可得,主成分分析分析只能解释56.52%的方差贡献率,同时不能完全将样品分类;偏最小二乘法-判别分析优于主成分分析的结果,能达到一定的分离,解释总体方差贡献率63.43%;而随机森林法能够很好的将样品分为3类,且3类样本的10折交互验证准确率达到96.5%。因此,随机森林算法联合HPLC指纹图谱能够有效构建中药质量控制和分析体系。

Application of random forest algorithm in fingerprint of Chinese medicine:different brands of Xiasangju granules as example

To establish a random forest algorithm for identifying and classifying different brands of Xiasangju granules, and provide effective reference for identifying multi-index complex fingerprint. HPLC method was used to collect the fingerprint of 83 batches of Xiasangju granules from different manufacturers. The classification of Xiasangju granules samples based on chromatographic fingerprints was identified by chemometric methods including principal component analysis (PCA), partial least squares discriminate analysis (PLS-DA) and random forest analysis (RF). The superiority of the above three chemometric methods was compared. The results showed that the fingerprints of 83 batches of Xiasangju granules were established in this study. PCA could only explicate 56.52% variance contribution rate and could not completely classify the samples; PLS-DA analysis was superior to PCA, explicating 63.43% variance contribution rate and could obtain certain separation; RF could well classify the samples into 3 types, and the predication accuracy of the proposed method was 96.5%. Therefore, The results indicate that RF combined with HPLC fingerprint could effectively construct traditional Chinese medicine quality control and analysis system.