基于核主成分分析的流式细胞数据分群方法研究

Cell data clustering method in flow cytometry based on kernel principal component analysis

针对多参数流式细胞数据分析过程复杂、自动化程度不高、要求操作者具有一定专业背景等问题,本文提出了一种基于核主成分分析算法(KPCA)进行多参数流式细胞数据分群的方法。利用KPCA对多参数流式细胞数据进行非线性变换,降低数据的维度,得到主成分特征变量下的散点图分群结果,并使用改进的Kmeans聚类算法实现不同亚群的自动设门。以人体外周血淋巴细胞样本检测结果为实验数据,分别对其进行传统分群、主成分分析(PCA)分群、KPCA分群处理,并对特征参数的选取进行了探索。结果表明,KPCA方法能够较好地应用于多参数流式细胞数据分析中,与传统细胞分群方法相比,该方法无需操作者具备专业知识,即可实现快速准确的自动分群,能够提高流式细胞仪临床诊断分析的效率。

The process of multi-parametric flow cytometry data analysis is complicate and time-consuming, which requires well-trained professionals to operate on. To overcome this limitation, a method for multi-parameter flow cytometry data processing based on kernel principal component analysis （KPCA） was proposed in this paper. The dimensionality of the data was reduced by nonlinear transform. After the new characteristic variables were obtained, automatical clustering can be achieved using improved K-means algorithm. Experimental data of peripheral blood lymphocyte were processed using the principal component analysis （PCA）-based method and KPCA-based method and then the influence of different feature parameter selections was explored. The results indicate that the KPCA can be successfully applied in the multi-parameter flow cytometry data analysis for efficient and accurate cell clustering, which can improve the efficiency of flow cytometry in clinical diagnosis analysis.