基于数据预处理的板蓝根注射液指纹图谱模式识别

中药色谱指纹图谱具有成分复杂、图谱基线和成分保留时间易受外界因素干扰等特点.本文对获得的色谱指纹图谱进行基线校正和色谱峰保留时间校正的数据预处理,不需色谱峰的积分,直接用各色谱信号的全谱数据评价相似度、主成分投影分析和谱系聚类分析,将各厂家的注射液样本正确分类并检出其中的异常样本,为处理和评价中药色谱指纹图谱数据,提供客观、便捷的方法.

Pattern recognition of chromatographic fingerprints of Banlangen injections based on data preprocessing

The instability of baseline and retention time of the chromatogram of traditional Chinese medicine often happen because of the column changes,the instrument drift and interaction function within analytes.Two data preprocessing algorithms,asymmetric least- squares baseline estimation and correlation optimization warping(COW),were used to correct the baseline and retention time shifts of Banlangen injection chromatographic fingerprints,respectively.Then the entire chromatographic profiles were directly used for similari- ty evaluation,principle component analysis and hierarchical clustering without peak integral.Eighteen Banlangen injections from three companies were divided into three categories and the outlier was detected.The methods of data preprocessing and pattern recognition in this paper were subjective and convenient,and it can be applied to evaluate the chromatographic fingerprint.