Abstract
BackgroundOntologies are widely used as metadata in biological and biomedicaldatasets. Measures of semantic similarity utilizeontologies to determine how similar two entitiesannotated with classes from ontologies are, andsemantic similarity is increasingly applied inapplications ranging from diagnosis of disease toinvestigation in gene networks and functions of geneproducts.ResultsHere, we analyze a large number of semantic similarity measures andthe sensitivity of similarity values to the numberof annotations of entities, difference in annotationsize and to the depth or specificity of annotationclasses. We find that most similarity measures aresensitive to the number of annotations of entities,difference in annotation size as well as to thedepth of annotation classes; well-studied and richlyannotated entities will usually show highersimilarity than entities with only few annotationseven in the absence of any biological relation.ConclusionsOur findings may have significant impact on the interpretation ofresults that rely on measures of semanticsimilarity, and we demonstrate how the sensitivityto annotation size can lead to a bias when usingsemantic similarity to predict protein-proteininteractions.
