Mungall paper

See link to Mungall draft paper on temporalized relations here.

Executive Summary

Since their inception, biological ontologies have made use of a core set of relations such as "part of" and "participates in". These relations were standardized and given a formal interpretation with the publication of the "Relations in Biomedical Ontologies paper" by Smith et al in 2005. Recently there has been a proposal to end support for these simple relations and replace them with more complex "temporalized relations", as part of the BFO2 OWL release. Ontology developers would need to replace existing relations with temporalized ones if they are to be conformant to BFO. The temporalized forms of "part of" in the current release are "part of at some times", "part of at all times", "part of at all times that whole exists". This set may even be extended to more forms in future releases.

This would be a radical change impacting all ontology developers and users, requiring complex re-engineering. Normally, such an unprecedented major change would have to be justified by a major flaw in the current system of simple non-temporalized relations. We might expect the flaw to be a major hindrance to ontology development or usage, and the new solution to work better. We might also expect the flaw to be described in a peer-reviewed publication.

However, this is not the case. There has been no demand from ontology developers or their users to make this transition. In fact most ontology developers have not been consulted. The sole justification for this radical change lies in a formal technicality, concerning the difference between the intepretation of relations in the Web Ontology Language (OWL) and the temporal interpretation provided in the Relations Ontology paper. Most experienced OWL modelers have been aware of this misalignment for a number of years, and accept this as a trade-off that comes with the benefits of using OWL. This trade-off has yet to cause difficulties for practical ontology development and use. Ongoing work suggests it may even be possible to find some satisfactory solution to the technicality whilst preserving existing relations.

Yet despite this lack of mandate, BFO2 does not include simple relations and provides temporalized relations in their place.

A number of experienced ontology modelers (including myself) have voiced serious concerns about the temporalized relations proposal. In addition I observed that less experienced ontology developers and users are highly confused by the BFO2 OWL proposal - they did not know how to migrate to or how to use the new relations, and they did not understand the underlying motivatation for introducing them in the first place. It was suggested to me that I collect some my objections in a document, which is what I have done in "A critique of BFO2 OWL temporalized relations".

The document provides an overview of the temporalization proposal, focusing on the part-of relation (although it should be understood that this is for the sake of brevity, and the proposal affects many more relations). I describe some of the features of BFO2 OWL that are widely accepted: the meaning of the BFO2 OWL relations are different from the interpretations in the 2005 Relations paper; important OWL characteristics such as transitivity and inverses differ from the reference form; temporalization leads to an increased number of relations, and these do not always connect together in intuitive ways. These are not new findings, but bear highlighting, as many potential BFO2 adopters were not previously aware of them.

I then provide an evaluation of BFO2 OWL, using examples from ontologies I work on, including the GO. Again, for reasons of space, I focus on "part of". I show that the current version of BFO2 OWL cannot accurately represent the parthood relationship between a nucleus and a cell without giving up on transitivity, a characteristic that is obviously essential. I also show that temporalized relations hinder our abilities to use OWL reasoning to check for certain kinds of mistakes in ontologies. Finally I demonstrate how the rigidity constraint imposed by BFO2 OWL means we cannot include certain classes, such as a pre-migratory neural crest cell, in our ontologies.

I consider this to be sufficient to demonstrate that not only does the current version of BFO2 OWL force overwhelming complexity onto ontologies, it simply will not work.

Recent discussions suggest workarounds for some of the problems I have highlighted. These workarounds involve the introduction of even more temporalized forms of relations, and the introduction of the concept of a history. Because these are neither formalized nor in the BFO2 OWL document, I am unable to give a thorough analysis demonstrating my concerns with these new proposed extension. However, I include a brief section describing these new relations.

My conclusions are that the temporalized relations in BFO2 OWL should not be used. They are far, far too complex, and even if this complexity could be overcome with significant tool support, they do not work in the currently published form. My recommendations are that ontology developers continue to use the simple non-temporalized relations they are accustomed to using, and that these relations should be supported so long as they are required by the community. This approach may not be perfect from a formal theoretical point of view, but it is essentially the approach that has been used for the last decade, and has not been documented to cause problems.