Predicting Binding of Transcriptional Regulators with a Two-Way Latent Grouping Model

Binding of transcription factors to the promoter regions of genes can be measured genome-wide to reveal regulatory networks. The measurements are expensive, however, and methods for predicting bindings from earlier data would reduce the cost. At best, genome-wide studies could be targeted based on a few test samples. We model the binding patterns using recent ideas from collaborative filtering and biclustering. A main difference from biclustering is that we compute a Bayesian prediction using all possible clusterings.