Drexel University Home Pagewww.drexel.edu DREXEL UNIVERSITY LIBRARIES HOMEPAGE >>

iDEA: Drexel E-repository and Archives > Drexel Theses and Dissertations > Drexel Theses and Dissertations > Identifying regions of difference in flow cytometric data

Please use this identifier to cite or link to this item: http://hdl.handle.net/1860/859

Title: Identifying regions of difference in flow cytometric data
Authors: Achuthanandam, Ramasubramaniam
Keywords: Electrical and computer engineering;Flow cytometry;Pattern perception
Issue Date: 27-Jul-2006
Abstract: We propose a pattern recognition framework for the identification of biologically interpretable regions of difference between groups of flow cytometric data samples. The identification of regions of difference is reduced to a problem of feature selection. We propose the use of a filter-based feature selection method using the two sample t-test to identify relevant features. We compare the performance of our method to an existing alternative, Frequency Difference Gating (FDG), and to other commonly used feature selection methods including filter-based methods such as a non-parametric Mann-Whitney test, a non-parametric resampling based Welch t-test, the Pearson’s square correlation coefficient and a wrapper method using support vector machine based recursive feature elimination. We evaluate the performance of the various methods using commonly used feature selection metrics, feature selection accuracy and classification accuracy. We also present the use of a metric from signal detection theory, the area under the Receiver Operating Characteristic curve (AUC), to evaluate the feature selection performance. Using these metrics, we demonstrate that the pattern recognition framework using the filter based tests outperform FDG and the wrapper based method. We show the performance using synthetically generated data and real biological data obtained from treating mice with an agonistic anti-CD3 antibody to determine activation levels in CD4+ and CD8+ T-cells. We find that our method is able to identify all the regions of differences between the two groups of samples. The pattern recognition framework is also used to analyze regions of differences in data collected from mice treated with an agonist (drug), erythropoietin (EPO). The regions of difference detected represent cells known to be affected by EPO. Using a possibilistic clustering algorithm, we identify the presence of four cluster of cells within our regions of difference, that respond differentially to EPO. The identity of the cells was determined by physically separating the cells of interest from the mixture and studying their morphology. In this experiment the technique provided us with information that EPO acts by stimulating and expanding basophilic erythroblasts in mice bone marrow.
URI: http://hdl.handle.net/1860/859
Appears in Collections:Drexel Theses and Dissertations

Files in This Item:

File Description SizeFormat
Achuthanandam_Ramasubramaniam.pdf13.93 MBAdobe PDFView/Open
View Statistics

Items in iDEA are protected by copyright, with all rights reserved, unless otherwise indicated.


Valid XHTML 1.0! iDEA Software Copyright © 2002-2010  Duraspace - Feedback