HIGH-RESOLUTION MULTIPARAMETER CELL ANALYSIS AND SORTING

点击翻译按钮获取中文摘要

Our proposed multivariate statistical analysis and sorting methods are based on what we perceive to be some of the more fundamental problems in multiparameter flow cytometry and cell sorting. The ability of experimenters in both the basic research and clinical laboratories to acquire complex multiparameter flow cytometric data has far out-stripped most currently available data analysis methods which deal with data at the histogram or bivariate display level. Listmode reprocessing and use of color to "map" the data through various histogram or bivariate displays is useful but inadequate for at least three reasons. First, the selection of regions is usually done by arbitrarily chosen rectilinear or bit-map boundaries. Second, even if all possible bivariate displays are viewed one does not see all of the data since these projections are at fixed orientations to the actual multidimensional data. Expecting to see all of the multidimensional data with bivariate displays is analogous to trying to see 2 dimensional data with the separate histograms. Third, non-flow cytometric (FCM) data is usually not included in the overall analysis (and never directly in the sorting process) of flow cytometry data in terms of statistical analyses of the combined FCM and non-FCM data. To address these problems we first attempt to view this complex data in as high a dimensionality as possible through use of projection pursuit analyses (e.g. principal components, Friedman-Tukey) and our own "local multivariate dimensionality" analyses and the use of our unique autostereo scopic display and 3D mouse. Second, we attempt to address the question of proper selection of analysis and sort boundaries based on rigorous multivariate statistical techniques rather than arbitrarily constructed rectilinear or bit-map boundaries. Third, we attempt to directly combine other non-flow cytometric information about the cells of interest (e.g. patient information, physicians' diagnoses) with flow cytometric variables using recursive partitioning techniques. In addition, we attempt to address the problem of high-resolution cell sorting beyond rectilinear and bit-map sorting by using the preceding statistical methods to allow real-time sorting of cells based on sophisticated statistics and expert systems criteria using our special "flexible" sorting system. Lastly, we propose implementing the preceding techniques in hardware, software and optics that will allow for direct sorting and subsequent analyses of cells to confirm the utility of these multivariate techniques for a variety of important biological and clinical applications.