Stein Shrinkage & Bayesian Inference for Cluster Analysis

In a pioneering paper on the analysis of gene-expression data derived
from microarray, Eisen et al. ("Cluster analysis and display of
genome-wide expression patterns," PNAS, 98), describe a "similarity
metric" for genes and argue that their proposed clustering method
"uses standard statistical algorithms to arrange genes according to
similarity in pattern of gene expression". We examine the underlying
statistical assumptions and suggest a broad class of similarity
metrics. In particular, we suggest that the methods based on "Stein
shrinkage" are more informative than the others. 

(Joint work with V. Cherepinsky.)