Scott R. McLaughlin

High-speed cell classification systems for real-time data classification and cell sorting

When conducting flow cytometric data analyses or cell sorting which results in classification of cells into two or more subpopulations, it is difficult to know which method is providing the best method and further classified using other methods.In this work we describe high-speed cell classification methods suitable for real-time data classification or cell sorting. Real-time generation of statistical classifiers and methods for comparing different classification methods by ROC analysis are also discussed. Multiparameter data mixtures of human MCF-7 breast cancer cells and human bone marrow wee analyzed by several cell classification systems including cluster analyses, principal components and discriminant function analysis. True classifier tags, implemented as additional correlated listmode parameters not used for these analyses, were used to uniquely identify each cell type and to compare classifier results. The performance of classifier systems was also assessed using ROC analysis. Preliminary results are discussed in terms of the advantages/disadvantages and problems/pitfalls of each method.

Real-time multivariate statistical classification of cells for flow cytometry and cell sorting: a data mining application for stem cell isolation and tumor purging

Multivariate statistics can be used for visualization of cell subpopulations in multidimensional data space and for classification of cells within that data space. New data mining techniques we have developed, such as subtractive clustering, can be used to find the differences between test and control multiparameter flow cytometric data, e.g. in the problem of human stem cell isolation with tumor purging. They also can provide training data for subsequent multivariate statistical classification techniques such as discriminant function or logistic regression analyses. Using lookup tables, these multivariate statistical calculations can be performed in real-time, and can even include probabilities of misclassification. Thus, the only distinction between off-line classification of cells in data analysis and real-time statistical decision-making for cell sorting is the time limit in which a classification decision must be made. For real-time cell sorting we presently are able to perform these classifications in less than 625 microseconds, corresponding to the time that it takes the cell to travel from the laser intersection point to the sort decision point in a flow cytometer/cell sorter. Statistical decision making and the ability to include the costs of misclassification into that decision process will become important as flow cytometry/cell sorting moves from diagnostics to therapeutics.

Isolation of Rare Cells by High-Speed Flow Cytometry and High-Resolution Cell Sorting for Subsequent Molecular Characterization - Applications in Prenatal Diagnosis, Breast Cancer and Autologous Bone Marrow Transplantation

As we have discussed previously (1), analysis and isolation of rare cell subpopulations have been of interest to researchers and clinicians in many areas of biology and medicine including: a) detection of somatic cell mutations in mutagenized cells (2), b) detection of human fetal cells in maternal blood for prenatal diagnosis of birth defects (3), c) detection of CALLA+ cells (4), d) detection of minimal residual diseases (5,6), e) detection of stem cells (7), and f) detection of rare HIV-infected cells in peripheral blood (8). Unfortunately, conventional flow cytometer/cell sorters operating at rates below 10,000 cells/sec require many hours to analyze and/or isolate cell subpopulations of low frequencies (e.g. 10-4 – 10-7) making them impractical to use for routine analysis and sorting of such cell subpopulations. One simple method for processing cells at higher speeds on conventional flow cytometers has been described (9). This method triggers the data acquisition or sort signal on a rare fluorescence signal.