GO-WORDS: An Entropic Approach to Semantic Decomposition of Gene Ontology Terms

ABSTRACT The Gene Ontology (GO) has a large and growing number of terms that constitute its vocabulary. An entropy-based ap-proach is presented to automate the characterization of the compositional semantics of GO terms. The motivation is to extend the machine-readability of GO and to offer insights for the continued maintenance and growth of GO. A proto-type implementation illustrates the benefits of the approach. 1 INTRODUCTION The underlying motivation of the work described in this paper is to map annotations based on the Gene Ontology (GO) (Ashburner, et al., 2000) to a semantic representation that exposes the internal semantics of GO terms to computer programs. The Gene Ontology (GO) views each gene prod-uct as being a structural component of a biological entity, being involved in a biological process, and as having a mo-lecular function. These three dimensions of component (C), process (P) and function (F) are hierarchically refined into several thousand subconcepts or GO terms for a fine-grained description of gene products, and ultimately a repre-sentation of collective biological knowledge. The machine-readability of GO is based on explicit IS-A or PART-OF relations between different GO terms (Fig. 1). The represen-tation of each GO term in terms of a phrase in English is primarily meant for human readability, and not machine-readability (Wroe, et al., 2003) (Fig. 1). For example, while both humans and computer programs can understand that ‘Folic Acid Transporter Activity’ is one kind of ‘Vitamin Transporter Activity,” only a human reader can appreciate that proteins annotated with ‘Folic Acid Transporter Activ-ity’ actually