Classification of cell subpopulations using multiple cellular parameters from high-content imaging studies

Cell samples from in vitro studies are almost never uniform; individual cells of a population may differ with respect to their developmental stage, phase in the cell cycle, state of transfection, or by natural variability. Therefore, a sample of cells is usually composed of several subcategories that depend on the experiment setting and research interest. A critical interest of much molecular research is to deconvolute the mixture population and quantitatively assess the subcategories by their proportions in the cell population. In this research, subpopulation classification analysis using the simultaneous analysis of multiple cellular parameters is proposed. In particular, an iterative K-means clustering algorithm is used to establish subpopulation fingerprints, which are then used to classify unknown cell samples by support vector machines. The analyses demonstrate that the multivariate subpopulation analysis is statistically powerful and biologically relevant.