Skip to Main Content
This work concerns a program (EFlow) that has been developed to assist experts in clustering flow cytometry multivariate data. EFlow aims to reduce subjectivity but keeping the expert's control over the analysis. In the present paper we will focus on the use of the iterative optimization algorithm. This algorithm is a hill-climbing unsupervised procedure that in general guarantees good local but not global optimization. A comparison between clustering processes achieved by an expert in flow cytometry and by EFlow was performed in a data related to bone marrow cells stained for the surface expression of molecules (B220, IgM and CD23) that characterize the B cell lineage. This data set is considered to present a medium difficult and had 63780 cells with 7 channels or dimensions. Two types of clustering processes were compared, one with four clusters and another with five clusters. The results obtained by EFlow were considered satisfactory by the expert as they were significantly in agreement with his classification. In some clusters, EFlow classified correctly up to 100% of the cells. At the present time few commercial cytometry software offer equivalent clustering analysis.