Bayesian biclustering with the plaid model

Biclustering is an active and promising research topic in unsupervised learning. With the aim of uncovering condition-specific similarities between objects, it may be applied in areas such as collaborative filtering and bioinformatics. The plaid model is amongst the most flexible biclustering models. However, its potential has not yet been fully explored. In this paper we extend the plaid model with a Bayesian framework and a collapsed Gibbs sampler. We show that the new method is useful in a gene expression study both in finding gene-specific associations between microarrays and condition-specific associations between genes.