中国蜂胶薄层色谱分析结果的机器学习判别研究

Research on Discrimination of Thin-layer Chromatographic Results of Chinese Propolis Using Machine Learning

关键词:  中国蜂胶  薄层色谱法  机器学习  判别分析

Key words:  Chinese propolis  thin-layer chromatogram  machine learning  discriminant analysis

摘要:

目的 对中国蜂胶和杨树胶的薄层色谱分析结果进行机器学习判别研究,探讨可行的策略。方法 样品用甲醇提取后,点样于硅胶60薄层板,以正己烷-乙酸乙酯-甲酸(10:4:0.5)为展开剂,用浓氨水和双氧水熏至显色清晰,然后用扫描仪扫描获取薄层色谱图。图像用机器学习作判别分析。用图像分析处理提取光密度色谱图数据后再分别用主成分分析、偏最小二乘判别、线性判别分析进行判别。将样品色谱道图像按照展开路径截取,分别用7种图像特征描述符提取图像特征后,用支持向量机对数据建模、判别。结果 主成分分析、偏最小二乘判别的结果显示中国蜂胶和杨树胶的色谱数据存在明显的分类界限。用线性判别分析进行判别,正确率达到100%。用图像特征描述符提取特征结合支持向量机的方法,以ColorLayout、Tamura图像特征描述符进行判别的效果最好,正确率为100%。结论 在中国蜂胶薄层色谱鉴别中应用机器学习判别分析是可行的,有助于提高薄层色谱图判别的客观性。用ColorLayout、Tamura图像特征描述符提取色谱图特征再结合支持向量机的方法简捷易用,适合用于大量薄层色谱图的机器学习分析。

Abstract:

OBJECTIVE To study discrimination of thin-layer chromatograms of Chinese propolis and poplar tree gum using machine learning and to investigate a feasible strategy. METHODS Samples were extracted with methanol and the solutions were applied on Silica 60 plate. A mixture of n-hexane-ethyl acetate-formic acid(10:4:0.5) was used as the mobile phase. The plate was treated with the ammonia and hydrogen peroxide vapours until the colour of the spots on the plate was clear. Then the image of the thin-layer chromatogram was acquired by office scanner. Images of chromatograms were discriminated using machine learning. Densitometric chromatogram data were acquired from images by image analysis and discriminated using principal component analysis, partial least squares discrimination, linear discriminant analysis, separately. And the lane images of samples were separately cropped according to their migration paths, features of lane images were extracted by seven image feature descriptors separately, and data were used for training and discrimination using support vector machine. RESULTS The results of principal component analysis and partial least squares discrimination showed that there was obvious classification boundary between chromatographic data of Chinese propolis samples and those of poplar tree gum samples. And the correct rate of machine learning using linear discriminant analysis was 100%. For the methods combining features extraction using image feature descriptors and support vector machine, discrimination according to ColorLayout and Tamura descriptor obtained better results, and their correction rates were both 100%. CONCLUSION It is feasible to use machine learning in discriminant analysis of thin-layer chromatographic identification of Chinese propolis, and it would be beneficial to the discriminant objectivity of thin-layer chromatographic identification. The method combining features extraction using ColorLayout, Tamura image descriptor and support vector machine were convenient to use and feasible to be used in analysis based on machine learning for a great number of thin-layer chromatograms.