EVENTS

Seminar: Clustering Posterior Distributions - Application to Somatic Embryogenesis in Maize (Alicia Carriquirry)

Seminar

Department of Statistics

Thursday, 17 September 2009 at 3:30PM

Sidney Smith Hall, Room 1086

Clustering Posterior Distributions - Application to Somatic Embryogenesis in Maize

Professor Alicia Carriquirry

Iowa State

In multiple treatment microarray experiments, we are often interested in locating groups of genes coregulated by the same biological pathways. The statistical problem is then clustering genes based on their expression values over multiple treatments.

We do not observe true gene expression, however. Instead, we observe a single or perhaps a replicated noisy estimate of gene expression under each treatment condition. There has been extensive work on methods for clustering genes based on gene expression reported in the literature, but in most cases, the clustering algorithm was implemented using the estimated expression levels without accounting for the uncertainty about the true levels of gene expression. In an earlier manuscript (Love and Carriquiry, JASA, 2008) we proposed a hierarchical modeling approach to estimate the posterior distribution of gene expression when multiple readings (or scans) are available for each gene under each treatment condition. The approach explicitly recognizes that true expression is unobservable.

We propose a clustering algorithm to nd groups of co-regulated genes in time experiments when we have an estimated posterior expression distribution for each gene. The distance be- tween the posterior distributions of expression ratios is measured using a symmetric divergence proposed by Jereys (1946) which equals the sum of the two Kullback-Leibler divergences. We show, via a simulation experiment, that the clustering approach described here outperforms two standard methods that are also based on the posterior distribution of expression. When implemented on data obtained from a maize embryogenesis experiment, the resulting groupings are justiable from a biological viewpoint.

Cookies and beverages will be served at 3:10 p.m.