Disjunctive shared information between ontology concepts: application to Gene Ontology

Francisco M Couto,Mário J Silva

doi:10.1186/2041-1480-2-5

Abstract

BackgroundThe large-scale effort in developing, maintaining and making biomedical ontologies available motivates the application of similarity measures to compare ontology concepts or, by extension, the entities described therein. A common approach, known as semantic similarity, compares ontology concepts through the information content they share in the ontology. However, different disjunctive ancestors in the ontology are frequently neglected, or not properly explored, by semantic similarity measures.ResultsThis paper proposes a novel method, dubbed DiShIn, that effectively exploits the multiple inheritance relationships present in many biomedical ontologies. DiShIn calculates the shared information content of two ontology concepts, based on the information content of the disjunctive common ancestors of the concepts being compared. DiShIn identifies these disjunctive ancestors through the number of distinct paths from the concepts to their common ancestors.ConclusionsDiShIn was applied to Gene Ontology and its performance was evaluated against state-of-the-art measures using CESSM, a publicly available evaluation platform of protein similarity measures. By modifying the way traditional semantic similarity measures calculate the shared information content, DiShIn was able to obtain a statistically significant higher correlation between semantic and sequence similarity. Moreover, the incorporation of DiShIn in existing applications that exploit multiple inheritance would reduce their execution time.

Highlights

The large-scale effort in developing, maintaining and making biomedical ontologies available motivates the application of similarity measures to compare ontology concepts or, by extension, the entities described therein
For each subontology of Gene Ontology (GO), Simresnik:dishin provides the highest correlation coefficients and Simlin and Distjc provide the lowest correlation coefficients. These results show that in this study a more accurate calculation of the shared information content is more relevant than including the Information Content (IC) of the concepts being compared
Using Fisher’s transformation and a one-sample z test, Table 1 presents the p-values for the correlation coefficients of Simresnik:grasm and Simresnik:dishin, considering the null hypothesis as that these coefficients being equal to the coefficients of Simresnik and Simresnik:grasm, respectively [[28], eq 11.22]

Summary

Introduction

The large-scale effort in developing, maintaining and making biomedical ontologies available motivates the application of similarity measures to compare ontology concepts or, by extension, the entities described therein. A common approach, known as semantic similarity, compares ontology concepts through the information content they share in the ontology. The most straightforward comparison methods are sequence-based. They only require information on their internal structure (the sequence itself), but limit the analysis to proteins sharing a similar structure, independently of their biological role. This ignores ontological knowledge about the properties and relationships among proteins. In opposition or as a complement to structural similarity, we should attempt to compare proteins based on the relationships between them [1]

Methods

Results

Conclusion