Identifying Differentially Expressed Genes Using Classification Error

• Activity Number: 61
• Type: Contributed
• Date/Time: Sunday, August 3, 2003 : 4:00 PM to 5:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Xiaohua Zhang*+ and Dhammika Amaratunga and Kathryn M. Roeder
• Companies: Merck Research Laboratories and RWJ Pharmaceutical Research Institute and Carnegie Mellon University
• Keywords: microarrays ; differentially expressed genes ; classification error ; Type I error ; class prediction

Abstract

In current literature, people focus on achieving a preset type I error in identifying differentially expressed genes. However, it is very hard to set up a reasonable significance level due to the low ratio of sample size to variable number and huge number of individual tests with one test for one gene in microarray experiments. Recently, the use of microarrays to predict tumor classes and to predict drug safety and efficacy has attracted intensive interest. Biomedical scientists are often more interested in searching a modest number of genes to discriminate samples with low classification error than type I error. Driven by this interest, we develop our method by implementing both criteria, differential expression and low misclassification rates, in the process of selecting differentially expressed genes, without the need to set up a per gene significance level. Our analyses show that it is feasible to select a modest number of differentially expressed genes to achieve a low misclassification rate. Our simulation study shows that the classification error method chooses a very reasonable number of differentially expressed genes for this purpose.