A Hybrid Approach to Finding Phenotype Candidates in Genetic Texts

Proceedings of COLING 2012: Technical Papers, pages 647–662, COLING 2012, Mumbai, December 2012. A hybrid approach to finding phenotype candidates in genetic texts Nigel Collier1,2 Mai − Vu Tran1,3 Hoang − Quynh Le1,3 Anika Oellrich4 Ai Kawazoe1 Martin Hall − May5 Dietrich Rebholz − Schuhmann2,6 (1) National Institute of Informatics, Tokyo, Japan (2) European Bioinformatics Institute, Cambridge, UK (3) University of Engineering and Technology - VNU, Hanoi, Vietnam (4) Wellcome Trust Sanger Institute, Cambridge, UK (5) University of Southampton, Southampton, UK (6) The University of Zurich, Zurich, Switzerland collier@nii.ac.jp,vutm@vnu.edu.vn,lhquynh@vnu.edu.vn ao5@sanger.ac.uk,zoeai@nii.ac.jp mhm@ecs.soton.ac.uk,rebholz@ebi.ac.uk ABSTRACT Named entity recognition (NER) has been extensively studied for the names of genes and gene products but there are few proposed solutions for phenotypes. Phenotype terms are expected to play a key role in inferring gene function in complex heritable diseases but are intrinsi- cally difficult to analyse due to their complex semantics and scale. In contrast to previous approaches we evaluate state-of-the-art techniques involving the fusion of machine learning on a rich feature set with evidence from extant domain knowledge-sources. The techniques are validated on two gold standard collections including a novel annotated collection of 112 ab- stracts derived from a systematic search of the Online Mendelian Inheritance of Man database for auto-immune diseases. Encouragingly the hybrid model outperforms a HMM, a CRF and a pure knowledge-based method to achieve an F1 of 77.07. Disagreement analysis points to further improvements on this emerging NE task. The annotated corpus and guidelines are available on request. KEYWORDS: conditional random fields, biomedicine, machine learning, genetic disorders, text mining. 647 1 Introduction Biomedical named entity recognition (NER) is a computational technique used to identify and classify strings of text (mentions) that designate important concepts in biomedicine. Over the last fourteen years there has been considerable interest in this problem with a variety of generic and entity-specific algorithms applied to extract the names of genes, gene products, cells, chem- ical compounds and diseases (Fukuda et al., 1998; Rindflesch et al., 1999; Collier et al., 2000; Kazama et al., 2002; Zhou et al., 2003; Settles, 2004; Kim et al., 2004; Leaman and Gonzalez, 2008). As the first stage in the integrated semantic linking of knowledge between literature and structured databases it is critically important to maximise the effectiveness of this step. Despite significant progress in NER there is still no one size fits all solution. Barriers arise because of ambiguity in the text and coding schema. Ambiguity in the text comes in vari- ous forms according to the semantic type of the entity but can be caused by a lack of stan- dard nomenclatures, extensive and growing nomenclatures for proteins/genes across multi- ple organisms or the widespread use of abbreviations and descriptive names. For example, (Krauthammer and Nenadic, 2004) illustrate uncontrolled naming in genes with bridge of sev- enless (boss) (FlyBase ID FBgn0000206) and Hunter and Bretonnel Cohen (2006) discuss term class ambiguity (e.g. is group a chemical entity or an assemblage of organisms?). Such chal- lenges have led to a variety of …

Content truncated to 500 words