Abstract
Annals of Applied Statistics 2012, Vol. 6, No. 3, 1327-1347 In many fields, researchers are interested in large and complex biological
processes. Two important examples are gene expression and DNA methylation in
genetics. One key problem is to identify aberrant patterns of these processes
and discover biologically distinct groups. In this article we develop a
model-based method for clustering such data. The basis of our method involves
the construction of a likelihood for any given partition of the subjects. We
introduce cluster specific latent indicators that, along with some standard
assumptions, impose a specific mixture distribution on each cluster. Estimation
is carried out using the EM algorithm. The methods extend naturally to multiple
data types of a similar nature, which leads to an integrated analysis over
multiple data platforms, resulting in higher discriminating power.
