A Model-Based Mediator System for Scientific Data Management

A database mediator system combines information from multiple existing source databases and creates a new virtual, mediated database that comprises the integrated entities and their relationships. When mediating scientific data, the technically challenging problem of mediator query processing is further complicated by the complexity of the source data and the relationships between them. In particular , one is often confronted with complex multiple-world scenarios in which the semantics of individual sources, as well as the knowledge to link them, require a deeper modeling than is offered by current database mediator systems. Based on experiences with federation of brain data, this chapter presents an extension called model-based mediation (MBM). In MBM, data sources export not only raw data and schema information but also conceptual models (CMs), including domain semantics, to the mediator, effectively lifting data sources to knowledge sources. This allows a mediation engineer to define integrated views based on (1) the local CMs of registered sources and (2) auxiliary domain knowledge sources called domain maps (DMs) and process maps (PMs), respectively, which act as sources of glue knowledge. For complex scientific data sources, semantically rich CMs are necessary to represent and reason with scientific rationale for linking a wide variety of heterogeneous experimental assumptions, observations, and conclusions that together constitute an experimental study. This chapter illustrates the challenges using real-world examples from a complex neuroscience integration problem and presents the methodology and some tools, in particular the knowledge-based integration of neuroscience data (KIND) mediator prototype for model-based mediation of scientific data.