Biological systems involve a large number of different entities, each functionning in a coordinated manner with the others. Their understanding is crucial and can be approached at different scales, from the molecular to the systemic one. The observation of all these entities in different contexts and at different scales generates a “tsunami of data", posing complex and interesting computational problems. My work focuses on the development of methods for knowledge generation and knowledge extraction from these massive data. The manuscript is organized in three axes. The first axis deals with methods to identify interpretable, robust and replicable signatures in high dimensional unimodal data. The second axis proposes the development of a new approach to integrate multimodal data (miRNA + MRI), and to identify disease progression scores. Finally, the third axis also deals with the integration of heterogeneous data, but with a systemic approach, i.e. taking into account the known relationships between entities. The work presented illustrates the complexity of extracting information from existing databases, despite the constant efforts of the bioinformatics community to structure and unify the available information.