Title: Hierarchical mixture priors for the analysis of gene expression data in cancer studies
Speaker: Guido Consonni, University of Pavia (Italy)
Date: Monday, December 8, 2003, 4:30 pm
Location: DIMACS Center, CoRE Bldg, Room 431, Rutgers University, Busch Campus, Piscataway, NJ
Technologies for the collection of genetic data, such as those based on microarrays for gene expression, are developing at a very fast rate. Cancer research is among the most important application areas for gene expression investigations, one reason being that cancer classification is still rather coarse and mainly based on morphological features, only occasionally supplemented with genetic-based information. Since cancer is a heterogeneous disease from a genetic standpoint, one would hope that an approach based on genetic information should provide a better basis for classification and ultimately for individualised prognosis and therapy.
Using probabilistic ideas to define the notion of differential expression for each gene, we assume that the distribution for the gene expression data is a mixture of three components (underexpression, normal expression, overexpression). Next we place an hierarchical mixture prior, with a random number of components, on some model parameters, in order to achieve a Bayesian clustering of the genes.
This research is still in progress, and accordingly the presentation will focus on issues rather than results.
This is joint work with Leonardo Bottolo, University of Pavia.