The impact of Gene Ontology evolution on GO-Term Information Content

TL;DR

This study investigates how the evolution of the Gene Ontology from 2005 to 2014 significantly impacts the information content of GO terms, affecting semantic similarity measures in bioinformatics.

Contribution

It provides a comprehensive analysis of GO evolution's effect on information content calculations, highlighting biases that influence semantic similarity assessments.

Findings

Significant differences in IC calculations across GO versions.

Statistically significant impact of GO evolution on semantic similarity measures.

Biases in GO evolution affecting bioinformatics analyses.

Abstract

The Gene Ontology (GO) is a major bioinformatics ontology that provides structured controlled vocabularies to classify gene and proteins function and role. The GO and its annotations to gene products are now an integral part of functional analysis. Recently, the evaluation of similarity among gene products starting from their annotations (also referred to as semantic similarities) has become an increasing area in bioinformatics. While many research on updates to the structure of GO and on the annotation corpora have been made, the impact of GO evolution on semantic similarities is quite unobserved. Here we extensively analyze how GO changes that should be carefully considered by all users of semantic similarities. GO changes in particular have a big impact on information content (IC) of GO terms. Since many semantic similarities rely on calculation of IC it is obvious that the study of these changes should be deeply investigated. Here we consider GO versions from 2005 to 2014 and we calculate IC of all GO Terms considering five different formulation. Then we compare these results. Analysis confirm that there exists a statistically significant difference among different calculation on the same version of the ontology (and this is quite obvious) and there exists a statistically difference among the results obtained with different GO version on the same IC formula. Results evidence there exist a remarkable bias due to the GO evolution that has not been considered so far. Possible future works should keep into account this consideration.