Information science for biology

Do ontologies for biology make the grade?

Soldatova, L. N. and King, R. D. Are the current ontologies in biology good ontologies? Nature Biotechnology, 23(9), pp. 1095-1098

dx doi 10.1038/nbt0905-1095

How can I resolve my different ideas and readings into one topic? When I consider what I have read over the last six weeks, several areas are prominent:

• the biomedical science as an information domain
• biomedical researchers as actors with information needs
• information retrieval in the biomedical sciences
• information management in the biomedical sciences
• knowledge discovery in the biomedical sciences
• innovation in the biomedical sciences
• the philosophy of science

As I was reading the paper from Soldatova and King, which I will come to in a moment, I felt that really what I have been thinking about is knowledge representation in the biomedical sciences. How are entities named? How are the related to one another? In research, how are observations and theories connected, stored and retrieved? Where do new theories originate from?

Perhaps the term ‘knowledge representation’ best describes what I am thinking about. This captures the high-order conceptual and theoretical relationships that are my primary interest, rather than the data description and retrieval tasks which are bread-and-butter for bioinformaticians. ‘Representation’ appeals to me because I am interested in how names symbolise entities in biology. Names have meaning for biologists, but that meaning might not be as tight and logical as ontologies might want them to be, or as biologists might really use them.

This paper from the Computational Biology research group at the University of Wales, Aberystwyth, highlights conceptual inconsistencies in the design of the MGED Ontology for describing microarray experiments and results. My Masters dissertation investigated the sharing and citation of microarray data through the Arrayexpress data repository, therefore it is interesting to read these criticisms of the community-developed ontology standard which describes this data-type.

Standards have been developed for creating ontologies, and these place careful constraints on ontology construction, based on formal logic. Intrinsic properties, class definitions and ‘Is-a’ relationships are strictly controlled. Failure to observe the standards will create logical inconsistencies both within the ontology, and in interoperability between ontologies.

The MGED Ontology breaks numerous rules and creates confusion for the paper authors with mixed classes, odd properties and errors in the ontology structure. Have these criticisms been addressed? I’m not sure, but the MGED Ontology is but one example of myriad ontologies in development to support biomedical data annotation and interpretation. Take a look at the Open Biomedical Ontologies and National Center for Biomedical Ontology’s BioPortal websites for many, many more.

Do all these ontologies meet the standards for logic for consistent application in machine processing or learning? I wonder if biologists themselves recognise the logical classes and properties built into these knowledge representations. More importantly, can they or would they apply these ontologies during the normal research process? Would these ontologies be built into instruments? Would annotation take place post-data acquisition, and who would do this? It strikes me that there would be an enormous investment of time and effort to describe even a simple cell biology experiment.

With all this effort going in to develop ontologies, how will they be used?