We consider two Bayesian network models (Fig. 1 ). The first model consists of a continuous node (circle) and a discrete node (square). The continuous node has a vector of gene expression ratios which is assumed to follow the binary mixture Gaussian distribution ( Fig. 1a ) (e.g. one for cancer, and the other for normal). The dicrete node defines proportions of these binary states. The second model additionally includes relationships between continuous nodes (Fig. 1b) . In these models, a

more »

... te node represents extrinsic factors (e.g. mutation, environment factor, or other genes which are not included in this experiment), and our purpose is to estimate the effect of such factors in the network context. The inter-gene relations in the network context are determined as follows. network context. The inter-gene relations in the network context are determined as follows.