Positional Paper on a Semantic Web for Life Sciences

Our research primarily involves the application of natural language processing technology to biomedical literature in support of such applications as semi-automated functional annotation of proteins and genes, and gene name normalization for improved search and retrieval of text information. We have performed studies in the use of existing database resources in these efforts (Morgan, Hirschman et al. 2003) and together with CNB/CSIC-Madrid, we have organized and administered a challenge evaluation, BioCreAtIvE (Valencia, Blaschke et al. 2004), for text mining systems applied to biomedical literature. Our primary experience with ontologies is with GO (The Gene Ontology Consortium 2000), and with some of the specific hierarchical controlled vocabularies. These include the FlyBase Controlled Vocabulary (The FlyBase Consortium 1993) and the TVFac Hierarchy. Our focus has been on automating the association of small excerpts of text and the underlying entities described (mentioned) in the text with concepts in the ontologies. We focus here on how existing ontologies and related resources can be augmented to aid text-mining and how text mining evaluation techniques can contribute to ontology evolution, by viewing ontologies as annotation guidelines when constructing/populating them.