Data understanding means dealing with highly heterogeneous data, in different schemes (text, relational, graphs, streams), from various domains (life sciences, astronomy, culture) and associated with knowledge encoded through many formalisms (graphs, ontologies, linked open data). Challenges to be tackled by Digicosme thus belong to data integration, knowledge representation and acquisition, data and knowledge reconciliation, data quality assessment. Our originality will lie in combining techniques from NLP, data mining, machine learning, knowledge representation (based on logics, ontologies, …), reasoning over such representations and using algorithms from graph and database theories. Collaborations with Scilex (through formal methods) and COMEX (through information theory) axis are key aspects of our approaches. Targeted applications include e-Health, e-Sciences, IoT, Open sciences and FAIR data.