A rapid, non-parametric clustering scheme for flow cytometric data

摘要

A rapid, non-parametric program to identify clusters in multidimensional space is presented. The local event density, based on a low resolution histogram count, is stored on hashcoded, sorted, linked lists. Regions of monotonically decreasing density around a given mode are grouped into a cluster. Nearby clusters may be merged.The algorithm can locate populations of only 1%, even in diffuse data, and gives percentages and locations of populations in real flow cytometric data in good agreement with results of conventional hand analysis. This is done in less than half a minute on a 68000 based microcomputer.