Principles of model development: Separation of Taxonomies

GALEN aims to produce an application-independent model. It has identified two features of many existing ontologies which reduce application independence and has designed means for avoiding them which are summarised under the principle of separating taxonomies [Rector 1995; Rector 1997; Rector and Horrocks 1997]. The fundamental principle is that why something has been classified in the way it has been should always be explicit in the model. If a concept has been classified along several axes, then the information should be sufficient for an application to choose which axes are relevant to that application.

This is achieved by following two principles:

• The subsumption (kind-of) relation is carefully distinguished from other transitive relations, such as partitive (part-of) or causitive (caused-by). A special mechanism, refinement is used to manage the interaction between subsumption and other transitive relations.
• The taxonomy of elementary concepts is analysed into pure hierarchies in which each elementary concept has only a single elementary parent. The principle for specialisation for this primary taxonomy of elementary concepts for each part of this taxonomy is carefully chosen and uniform. All other classification is performed automatically on the basis of the descriptions and definitions of the concepts. For example the asserted taxonomy of elementary concepts for chemicals is based on their structure. There is a second, pure hierarchy of chemical functions which exists separately from the hierarchy of the chemicals themselves. Testosterone is asserted into the chemical hierarchy as being, structurally, a steroid but is then additionally described as having the function of a male sex hormone. It is therefore classified under both headings - chemical by structure and chemical by function - but the two hierarchies remain distinct and can be used for different purposes as required. The choice of principle for the taxonomy of elementary entities is necessarily arbitrary and is often made on the basis of convenience and efficiency. There are a few necessary exceptions to the rule that elementary concepts have only a single elementary parent, but they account for less than 5% of cases and occur for specific reasons.