Prior beliefs about the structure of a probabilistic network

The exploitation of prior information is often advocated to face the curse of dimensionality which affects multivariate statistical models in biomedical research and bioinformatics. In a recent contribution dealing with score-and-search learning of Bayesian Networks, the elicitation of beliefs in large spaces of structures has been con- sidered in a general setup based on elicited network features. In this contribution we build on the above cited work and provide a simpler formalization dealing with incomplete prior information in the form of DAG of network features. A case study illustrates the proposed approach.