Incremental ontology-based integration for translational medical research

Translational medical research is an emerging concept that aims at transforming discoveries from basic sciences into diagnostic and therapeutic applications. In the opposite direction, clinical data are needed for feedback and as stimuli for the generation of new research hypotheses. This process is highly data-intensive and centered around the idea of integrating data from basic biomedical sciences, clinical sciences and patient care. Therefore collaboration and information exchange is needed between previously separated domains, many of which are themselves fragmented. The complexity and heterogeneity of the involved data is constantly growing with increasing scientific progress and related biomedical structures and processes are subject to rapid change. For this reason, structured domain knowledge, e.g., from knowledge bases, is often required in order to adequately understand and interpret results. Furthermore, integration solutions have to be robust and flexible enough to handle changes in data and metadata. Security and privacy aspects are highly relevant and require the incorporation of complex access control mechanisms as well as concepts for data anonymization and pseudonymization. In this thesis, first an ontology-based methodology for integrating heterogeneous biomedical datasets in a distributed environment is proposed. It advocates an incremental approach that builds upon data coexistence and aims at carrying out semantic integration in a demand oriented and flexible manner. The importance of structured domain knowledge is addressed by blurring the boundaries between primary data and metadata. Data federation allows researchers to maintain control over their local datasets and is also utilized to model a fine-grained access control mechanism. Robustness is achieved by designing the system as a set of loosely coupled components, which can be added, altered and removed independently of each other. Second, an implementation based on a large distributed graph of uniquely identifiable nodes is presented. The groundwork is laid by novel techniques for mapping biomedical data sources into the graph. As all further components require an integrated access to the global data, several compile-time and and run-time optimization techniques for the efficient distributed execution of queries are presented. Manual semantic integration is supported by concepts for browsing, annotating and mapping data items. Automated semantic integration and data transformation is supported via a flexible workflow engine, which builds upon the querying interface. Here, result sets can be post-processed with a scripting language that provides domain-specific operators, such as semantic reasoning and data anonymization. The resulting transformed data can be re-integrated into the graph. Finally, a prototypical implementation is presented which integrates the individual components into a comprehensive data integration solution. It provides a querying interface for applications and allows to administer the data space via a unified graphical user interface for data integrators.