Feature Selection and Binary Classification Using Microarray Data
DNA microarray technology has greatly influenced the realms of
biomedical research, with the hopes of significantly impacting the
diagnosis and treatment of diseases. Microarrays have the ability
to measure the expression levels of thousands of genes
simultaneously. This research entails a rigorous and systematic
comparison of statistical classifiers composed of supervised
learning methods and either univariate or genetic algorithm-based
multivariate feature subset selection (FSS) techniques, all applied
to six published two-class microarray datasets.
These analyses should provide insights into univariate vs.
multivariate FSS and how to obtain realistic and honest
misclassification error rates via cross-validation (CV).
Ultimately, this research puts to test the more traditional
implementations of CV and FSS and provides a solid foundation on
which these topics can and should be further investigated when
performing limited-sample classification using high-dimensional
gene expression data.
The analyses presented should be especially useful to researchers
in the fields of bioinformatics, supervised learning, and variable
selection, as well as anyone interested in microarray analysis.
Michael L. Lecocke, Ph.D.: Studied microarray gene expression analysis at Rice University and M.D. Anderson Cancer Center, Houston, TX. Currently Assistant Professor in Dept of Mathematics at St. Mary's University and Adjunct Assistant Professor in Dept of Epidemiology and Biostatistics at University of Texas Health Science Center, San Antonio, TX.
