Finding Differentially Expressed Genes in Microarray Experiments Augmented by Gene Annotation Data

• Activity Number: 360
• Type: Invited
• Date/Time: Wednesday, August 11, 2004 : 2:00 PM to 3:50 PM
• Sponsor: SSC
• Author(s): Rafal Kustra*+
• Companies: University of Toronto
• Keywords: multivariate analysis ; regularization ; microarray ; differential expression ; high-dimensional data

Abstract

High-throughput genomics (HPG), including microarray techniques, is one of the biggest stories in biology and life sciences, and will likely map one of the key future directions for biostatistics and applied statistics. The HPG data are uniquely challenging to statisticians, and classical methods of our field are rarely applicable directly. One of the defining elements of HPG datasets is a huge number of correlated measurements done on small number of independent samples. Typically in microarray studies, tens of thousands of genes have their expression levels probed, but the number of samples are between a dozen and few hundred. A number of ways was proposed to deal with this huge-dimensionality problem, either as a pre-analytic step (clustering genes, PCA) or incorporated in the analysis. We will describe our approach that incorporates auxiliary sources of information, such as geneontology classification, to build effective regularizers into our models. We concentrate on the problem of finding differentially expressed genes and use crude measures of similarity based on hierarchical, GO annotation to augment the covariance matrix in multivariate models.