Advanced real-time classification methods for flow cytometry data analysis and cell sorting
James F. Leary, Lisa M. Reece, James A. Hokanson, Judah I. Rosenblatt · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2002
While many flow cytometric data analysis and 'discovery' methods have been developed, few of these have been applied to the problem of separating out purified cell subpopulations by cell sorting. The fundamental problem is that the data analysis techniques have been performed using relatively slow computational methods that take far more time than is allowed by the sort decision on a cell sorter (typically less than a millisecond). Thus cell sorting, which is really a form of 'real-time data classification,' is usually done with few, if any, multivariate statistical tools used either in the sort decision or in the evaluation of the correctness of the classification. We have developed new multivariate data analysis and 'data discovery' methods that can be implemented for real-time data classification for cell sorting using linked lookup tables. One multivariate 'data discovery' method, 'subtractive clustering,' has been used to find which clusters of cells are different between two or more files (cell samples) and to help guide analysis or sort boundaries for these cell subpopulations. Multivariate statistical methods (e.g. principal component analysis or discriminant function analysis) were implemented in linked lookup tables to establish analysis/sort boundaries that include 'costs (or penalties) of misclassification. Costs of misclassification provided a measure of the quality of the analysis/sort boundary and were expressed in simple terms that describe the tradeoff between yield and purity.