Tagging gene and protein names in full text articles

Tagging Gene and Protein Names in Full Text Articles Lorraine Tanabe and W. John Wilbur National Center for Biotechnology Information NLM, NIH Bethesda, Maryland 20894 Abstract Current information extraction efforts in the biomedical domain tend to focus on finding entities and facts in structured databases or MEDLINE abstracts. We apply a gene and protein name tagger trained on Medline abstracts (ABGene) to a randomly selected set of full text journal articles in the biomedical domain. We show the effect of adaptations made in response to the greater heterogeneity of full text. 1 Introduction The application of large-scale genomics and proteomics technologies towards a wide variety of biological questions has resulted in a continuous stream of information regarding thousands of genes and gene products into the Medline database of biomedical abstracts. This repository has been recognized as a rich knowledge source for biological information retrieval, information extraction and text mining. However, abbreviated scientific abstracts cannot contain the same volume of information as the full text articles that they represent. It was recently shown that only 30% of protein interactions contained in the Dictionary of Interacting Proteins (DIP) (Xenarios et al., 2000) could be found in Medline sentences containing DIP protein pairs (Blaschke et al., 2000). This finding suggests that current information extraction efforts being applied to biomedical abstracts should be extended to full text databases. The basic task of identifying gene and protein names is a necessary first step towards making full use of the information encoded in biomedical text. This remains a challenging task due to the irregularities and ambiguities in gene and protein nomenclature. The irregularities are mainly the result of a lack of naming conventions, as well as the widespread practice of using many synonyms for one gene or protein. A glance at the Nomenclature section of the Nature Genetics website (http://www.nature.com/ng/web_specials/nomen /) shows the scope of the problem, as well as ideas for addressing it. The nomenclature guidelines implore authors to consult relevant nomenclature committees before announcing new genes, and to provide synonyms for genes in abstracts. Additional rules specify that: 4. Gene symbols are always italicised and never contain hyphens, greek letters, roman numerals, subscripts or superscripts. 5. All letters in human genes are upper- case…all letters in mouse genes are lower-case… Unfortunately, we are currently at a stage where these types of rules are not consistently applied to most biomedical abstracts, let alone to full text documents. Until the biomedical community adheres uniformly to nomenclature guidelines, ambiguities regarding gene/protein names will continue to be an obstacle for natural language processing of biomedical text. These ambiguities become apparent at the morphological, syntactic and semantic levels. For example, caco-2 refers to a cell line, but pai-1 is a gene name. Gene and protein names can contain verbs and other parts of speech that are hard to distinguish from the surrounding text, as in deleted in azoospermia-like, son of sevenless, ran, man, young arrest and never in mitosis. Genes can be transfected into cells, or combined with chemicals, …

Content truncated to 500 words