Attributes and Semantic Types

Though the object hierarchy provides a useful structure for organizing data in OME, it provides no means for storing or representing the data itself. This is because the objects in the object hierarchy carry very little state themselves, usually nothing more than a name and an owner. Rather, each piece of data is a logically distinct object, called on attribute, which refers to an object in the object hierarchy (Not just any object, though; attributes can refer to datasets, images, and features, but not projects).

Attributes can also be global, signifying that they represent entities outside of the object hierarchy. A gene, for instance, could be represented as a global attribute. Information about that gene, calculated in the context of a specific image, would be image attributes that contain references to the gene's global attribute.

If the data itself is represented by attributes in OME, the descriptions of the data are represented by semantic types. In the OME semantic type model, each attribute is an instance of a single semantic type. The semantic type defines several fields, or semantic elements, which specify the particular simple values which make up the data of that type. The type also has a granularity, which specifies which kind of object in the hierarchy (dataset, image, feature, or global) that instances of that type refer to.

Note that every piece of data in OME follows this pattern. OME differs from many imaging systems in that the pixels of an image have no special status. They are represented as semantically typed attributes just like any other piece of data. One benefit of this is that it allows an image to have multiple sets of pixels — for instance, the raw pixels directly from the microscope and the pixels that were post-processed via a deconvolution algorithm.

All attributes in OME carry one additional piece of information: Provenance. The provenance of an attribute is its derivation or history, i.e. where it came from. Since OME was initially conceived as an analysis system as well as an image repository, the source of an attribute refers to the algorithm or process that generated it. Algorithms (e.g. segmentation) or processes (e.g. import) are defined as Modules which have inputs and outputs. In this way, the entire provenance of a given attribute can be traced through module outputs and inputs back to its unqualified origin, which includes things like user inputs and original data files. Thus, all data in OME always carries three qualifiers: Type (what it means), Context (what it pertains to, or granularity), and Provenance (where it came from).

OME predefines a core set of semantic types which are useful for describing the kinds of data and metadata currently common in image biology. This includes descriptions of the state of the microscope which acquired an image, information about the experimental conditions of the image sample, and user-defined annotations. However, the type model in OME is completely extensible; new types can be defined and specified easily, allowing new kinds of data to be incorporated into OME with a minimum of fuss.