Me and my friends: gene mention normalization with background knowledge

“Tell me who your friends are, and I will tell you who you are” – this proverb best illustrates our approach to the normalization of gene names. In this approach, we rely on background knowledge that describes various aspects of a gene: it is localized on a chromosomal band, it belongs to an operon structure, it is a member of a gene family, its products take part in biological processes, they fulfil molecular functions, they occur at dedicated cellular locations, mutations of the gene ultimately cause diseases, its proteins contain domains and form secondary, tertiary and quaternary structures. Whenever a gene (or one of its products) is discussed, some of these aspects –the gene’s friends, that is, semantically related information– will be mentioned as well. The paradigm we follow with this approach demands not only the presence of a gene’s name, but also of some of its friends. We see every set of information available (see Methods section) for each gene as this gene’s description, or the context this gene typically “lives” in. Whenever we encounter ambiguities regarding proper identification of a gene, we assess each potential candidate gene by comparing its typical context against the predicted one (in this case, a PubMed abstract.) The descriptions for genes originate from various curated resources: EntrezGene provides organisms, summaries, chromosomal loci, Gene Ontology (GO) terms, and encoded proteins; UniProt provides functional descriptions, protein domains, interaction partners, keywords, and GO terms; more GO annotations are provided by GOA. Consider the example of the oncogene p54 (reflected in Figure 1.) Having resolved the issue of potential organisms, there are still human genes from EntrezGene that share the same name. Indeed, they refer to completely different genes with disjoint annotations. Based on the name alone, this problem could not be solved. Only comparing each of the gene’s contexts to the text reveals that one of the potential candidates is a RNA helicase, and the text indeed mentions “RNA helicase.” The text also mentions the exact chromosomal location of the correct gene.