Advances in information technology contributed powerful tools for the development of scientific applications. Today scientists routinely exploit data mining and analysis tools, visualization tools, and increasingly leverage database and workflow management systems. Some of these tools run out-of-the-box, others require customization. But when it comes to metadata management, there is very little reusable technology that scientists can draw upon. Manipulating metadata, such as database schemas, taxonomies, interface definitions, and mappings between them, requires tedious object-at-a-time programming and in-depth database expertise. This problem is symptomatic across other data-intensive domains, including enterprise information management, healthcare, and financial applications.