Affiliations: Biometris, Wageningen University and Research Centre,
6700 AC Wageningen, The Netherlands | Applied Bioinformatics, Plant Research International,
6708 PB Wageningen, The Netherlands
Abstract: The Gene Ontology (GO) is a widely used controlled vocabulary for
the description of gene function. In this study we quantify the usage of
multiple and hierarchically independent GO terms in the curated genome
annotations of seven well-studied species. In most genomes, significant
proportions (6–60%) of genes have been annotated with multiple and
hierarchically independent terms. This may be necessary to attain adequate
specificity of description. One noticeable exception is Arabidopsis
thaliana, in which genes are much less frequently annotated with multiple
terms (6–14%). In contrast, an analysis of the occurrence of InterPro hits
in the proteomes of the seven species, followed by a mapping of the hits to GO
terms, did not reveal an aberrant pattern for the A. thaliana genome.
This study shows the widespread usage of multiple hierarchically independent GO
terms in the functional annotation of genes. By consequence, probabilistic
methods that aim to predict gene function automatically through integration of
diverse genomic datasets, and that employ the GO, must be able to predict such
multiple terms. We attribute the low frequency with which multiple GO terms are
used in Arabidopsis to deviating practices in the genome annotation and
curation process between communities of annotators. This may bias genome-scale
comparisons of gene function between different species. GO term assignment
should therefore be performed according to strictly similar rules and
standards.
Keywords: Gene Ontology, genome annotation, annotation strategies, protein function, gene function prediction, multi-label classification, Arabidopsis genome
