While cell sorting as a technology has been in existence for over 30 years, only recently has its importance for basic research, clinical and commercial applications been fully appreciated. For basic research, single-cell PCR exploits the superb analytical power of multiparameter flow cytometry (FCM) to select small cell subpopulations from complex mixtures, ushering in a new age of single-cell molecular biology. For clinical research, applications such as stem cell isolation (with perhaps simultaneous tumor purging and gene therapy) are on the horizon. For commercial applications, high-speed sorting of bacterial clones with inserted human gene sequences may provide new capabilities for drug and vaccine design by the pharmaceutical industry. These are but a few examples of a so-called """"""""mature"""""""" technology undergoing a re-birth due to other technological advances and important new applications. Yet despite the increased importance of cell sorting, the technology fails to take advantage of technological and basic research advances particularly in the area of statistical classification of cells essential to an intelligent sort decision. In this proposal we address four problem areas that need serious work to make the technology reach its full potential. First, there is a need to better identify the number and location of cell subpopulations. Human pattern recognition can only be useful if complex multidimensional data can be viewed in a more useful fashion using data dimensionality reduction. New data mining techniques we have been developing such as """"""""subtractive clustering"""""""" can assist human pattern recognition in complex multidimensional data spaces and help discover the important differences between two or more cell samples for training sets for subsequent statistical classification techniques. (Specific Aim 1). Second, additional FCM parameters should be added only after determining through the use of logistic regression and stepwise discriminant function analysis methods what, if any, discriminating power they add to the classification of cell types of interest and sort decision boundaries should be less arbitrary and more based on statistical decision making (Specific Aim 2). Third, very high-speed cell sorting needs to be done as a multi-step rather than single-step classification decision (Specific Aim 3). Flow cytometry has largely avoided the consequences of such mistakes in classifying cells because it is used as only one, of several sources of information. Cell sorting, as real-time cell classification combining data analysis and decision-making, may soon be used in the clinical arena, e.g. for re-infusing cancer patients with their own autologous transplants. Mistakes in cell classification could lead to serious consequences for patients receiving misclassified (e.g. tumor) cells. Sort decisions need to make intelligent tradeoffs between yield and purity, and they should include costs of misclassification (Specific Aim 4).
| Szaniszlo, Peter; Wang, Nan; Sinha, Mala et al. (2004) Getting the right cells to the array: Gene expression microarray analysis of cell mixtures and sorted cells. Cytometry A 59:191-202 |
