Bayesian Methods for Microarray Data Analysis

Thursday, October 2, 2003 - 1:30pm - 2:20pm
Keller 3-180
Marco Ramoni (Harvard Medical School)
Data produced by microarray experiments - measuring thousands of genes with limited replicates - present unparallel opportunities to understand the global behavior of the genome and unprecedented analytical challenges. This talk will introduce a general Bayesian framework able provide coherent solutions to some critical problems of microarray data analysis and open new, unexplored avenues of discovery. The talk will start by describing a Bayesian approach to the analysis of comparative experiments, able to deliver high sensitivity and superior reproducibility. It will then describe a Bayesian solution to clustering gene expression data and it will introduce a principled probabilistic criterion to automatically identify the optimal number of clusters underlying a set of microarray experiments. It will also show how this clustering method can be naturally extended to profile the temporal behavior of gene expression dynamics. Finally, the talk will take this Bayesian framework one step forward, and show how it can be used to dissect the regulatory mechanisms of gene expression using a new class of Bayesian networks, called Generalized Gamma Networks, specifically designed to handle the peculiar distributional nature of microarray data and the non-linearity of gene expression control.