Statistical Modeling of Microarray Gene Expression Data

• Activity Number: 294
• Type: Invited
• Date/Time: Wednesday, August 14, 2002 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section*
• Author(s): Ernst Wit*+
• Affiliation(s): University of Glasgow
• Keywords: gene expression ; ANOVA ; MCMC algorithm ; data mining ; differential expression ; microarrays

Abstract

A key challenge for statisticians dealing with microarray experiments is to develop statistical models and procedures that are both honest reflections of the many sources of variability present in such experiments, as well as fast, algorithmic approaches to be able to give answers in real time. We argue that this requires a keen understanding of biological and physical processes, as well as a pragmatic mix of statistical procedures. Quick MLE image correction methods are developed based on the joint distribution of the mean and median to correct for intensity truncation effects. An ANOVA procedure is proposed to normalize raw gene expression data. These normalized data can then be piped into various inference tools, depending on the question of interest. Differentially expressed genes can be detected by empirical Bayes methods (Efron et al. 2001), and new confidence intervals for posterior probabilities are derived. Clustering algorithms of time-course experiments to detect similarly expressed genes typically ignore the time structure of the data. Correlation methods and dynamic MCMC algorithms could introduce more realistic models to answer more meaningful biological questions.