Functional Data Analysis of Microarray Data 


"A Robust Bayes Analysis of a Comparative Microarray 
Experiment", Dhammika Amaratunga, Johnson & Johnson 
Pharmaceutical Research & Development, and Javier Cabrera, 
Rutgers University.  

Abstract

One of the many challenges of microarray data analysis is 
to produce results that are unaffected by the numerous 
outliers usually present in microarray data. We will 
present an example of an experiment, in which microarrays 
consisting of genes from a virus were exposed to a solution 
of fluorescently labeled cDNAs prepared from either mock or 
true infected human fibroblast cells. The statement levels 
of the various viral genes were recorded with the objective 
of detecting which viral genes are expressed to a 
significantly higher degree when exposed to the true 
infection as compared to the mock infection. To analyze 
this data, we fit a Bayesian variance components model and 
computed the Bayes estimator and the Bayes posterior 
distribution of the average gene statement level. Then the 
posterior distributions obtained from this fit were used as 
input to a hierarchical model to establish gene 
upregulation. Since the data contained many outliers, we 
developed a robust version of the first stage of the 
procedure that produced Bayes estimators that were highly 
efficient and that, at the same time, were highly resistant 
to outliers. The overall procedure appears to be one that 
would be useful in general for analyzing data from 
comparative microarray experiments.