Gene-Class Testing for Multigroup Comparisons

• Activity Number: 487
• Type: Contributed
• Date/Time: Thursday, August 7, 2008 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Ching-Wei Chang*+ and James J. Chen+
• Companies: National Center for Toxicological Research and National Center for Toxicological Research
• Keywords: Gene expression ; generalized linear model ; random effect

Abstract

Gene-class testing (GCT) has been proposed for gene expression analysis to understand biological functions. In GCT, the hypothesis is that if there are treatment effects in the gene class. Some statistical methods for gene-class testing are only for one-sided test, that is, the changes of individual gene expressions in a gene class are all in one direction: either up or down. The one-side test might not close to the reality. Another problem is that current methods only focus on the two-group comparison. However, many studies involve comparisons of more than two groups. In this study, we use the generalized linear model by assuming random effects of gene expressions for a two-sided test in multi-group comparisons. The generalized linear model approach can take the clinical covariates into consideration. We propose computing techniques to deal with the large number of parameters.