Use and misuse of the gene ontology annotations

Abstract

The Gene Ontology (GO) has a structure that allows powerful comparisons and inferences about gene functions, but its structure is often misunderstood or ignored in practice. Evidence codes, annotations for unknown functions and annotation qualifiers are vital aspects of GO annotations, but these crucial features of GO annotation are often overlooked. Functional profiling using GO annotations is often performed in an incorrect or inappropriate way. Important issues related to this include a tendency to perform only enrichment testing, using an incorrect reference set, lack of or an inappropriate correction for multiple comparisons, indiscriminate propagation of annotations through the hierarchy, and ignoring the correlations between GO terms. Any analysis using GO annotations should cite data sources, including the version of ontology, date of annotation files, numbers and types of annotations used, and the versions and parameters of software, to ensure that results are fully reproducible. Pie charts are not appropriate for displaying GO functional categorization because of the GO structure and annotation practices. Functional characterization studies should indicate the number of genes that are not mapped to any slim term, are mapped directly to the root node, or are unannotated.