Figure 1

Schematic overview of the Pro Bic model and the conditional probability distributions of the attributes. Gene, Array and Expression represent the three Pro Bic classes of the PRM model. For each class, a set of specific gene, array and expression objects exists (denoted by the lowercase letters g, a and e respectively). The complete set of genes, array and expression objects that belong to a certain class are indicated by uppercase letters G, A and E. For the Gene (Array) class, a Boolean attribute B _b indicates whether a gene (array) belongs to a bicluster b or not. For each gene (array) object, the gene-bicluster labels g.B _b (over all biclusters b) and the array-bicluster labels a.B _b are the hidden variables of the model. Each object e of the Expression class has one single numeric attribute e.level that contains the expression level for each specific gene and array combination. The array class has an additional attribute ID that uniquely identifies each individual array object a.

The conditional probability distribution P(e.level|e.gene.B,e.array.B,e.array.ID) is modeled as a set of Normal distributions, one for each array-bicluster combination. A number of marginal distributions P(a.B _b ), P(g.B _b ) and P(g.B) allow expert knowledge to be introduced in the model, as explained in the main text.