Inferring the structure of gene regulatory networks from gene expression data has attracted a growing interest during the last years. Several machine learning related methods, such as Bayesian networks, have been proposed to deal with this challenging problem. However, in many cases, network reconstructions purely based on gene expression data do not lead to satisfactory results when comparing the obtained topology against a validation network.

Therefore, in this paper we propose an "inverse" approach: Starting from a priori specified network topologies, we identify those parts of the network which are relevant for the gene expression data at hand. For this purpose, we predict the expression level of a given gene from its relevant regulators.