PharmaDM's concept of a relational database incorporates a scientific data cube. The data cube is a common term in the context of data warehousing. In data warehousing, a data cube is used to store numerical data that can be viewed at all possible levels of detail. A typical example in marketing is the amount of sales of a product in a certain store at a certain date. The product gamma, different stores and timeline are then the dimensions of the data cube. Sales can be viewed in summaries (e.g. from the sales of a product in one store to the summarised sales in all stores of a certain geographical area) or in detail. The operations performed on the data cube to view these different levels of abstraction are roll-up and drill-down, respectively. The operation to view selections of data (like the sales data for drinks only) is called slicing (i.e. selection of a sub-cube). Each dimension of the cube is thus defined by its rigid concept hierarchy through which the user can navigate using the described set of operations.

The data cube model is an ideal means for the life science researcher to navigate through the data in an orderly way, and to view the data at all desired levels of detail. Biological, chemical and clinical information can then each represent a different dimension of the cube




The data cube model is an ideal means for the life science researcher to navigate through the data in an orderly way, and to view the data at all desired levels of detail. Biological, chemical and clinical information can then each represent a different dimension of the cube. A data point in the cube could for example be the expression level of a certain gene in a certain tissue (biological dimension), under the influence of a certain drug (chemical dimension) for a certain patient (clinical dimension). However, in order to make the data cube a practical and functional tool in life science applications, PharmaDM has extended the data cube functionalities drastically:

• In addition to numerical data, also symbolic data can be included in the cube.
• Concept hierarchies in the different dimensions need no longer be rigid, but can be user-defined, or can be generated automatically, like a decision tree built in a data mining experiment.