Utilizing text mining results: The Pasta Web System

Utilizing Text Mining Results: The PastaWeb System G. Demetriou and R. Gaizauskas Department of Computer Science University of Sheffield Western Bank Sheffield S10 2TU UK � g.demetriou, r.gaizauskas � @dcs.shef.ac.uk Abstract Information Extraction (IE), defined as the activity to extract structured knowl- edge from unstructured text sources, of- fers new opportunities for the exploitation of biological information contained in the vast amounts of scientific literature. But while IE technology has received increas- ing attention in the area of molecular bi- ology, there have not been many exam- ples of IE systems successfully deployed in end-user applications. We describe the development of PASTAWeb, a WWW- based interface to the extraction output of PASTA, an IE system that extracts protein structure information from MEDLINE ab- stracts. Key characteristics of PASTAWeb are the seamless integration of the PASTA extraction results (templates) with WWW- based technology, the dynamic generation of WWW content from ‘static’ data and the fusion of information extracted from multiple documents. 1 Introduction The rapidly growing volume of scientific literature, a by-product of intensive research in molecular bi- ology and bioinformatics, necessitates efficient and effective information access to the published text sources. Information retrieval (IR) techniques em- ployed in WWW interfaces such as PubMed and Entrez are very useful in browsing bibliographic databases and in facilitating the linking between protein or genome sequences and related refer- ences. But the jobs of automatically locating and extracting specific information within the texts re- quire more specialised Natural Language Processing (NLP) techniques and have been the object of work in information extraction (IE) or text mining ((Cowie and Lehnert, 1996)). While current work on IE in biology has concen- trated by and large on the refinement of IE tech- niques and improving their accuracy, the incorpo- ration of an IE system’s output into effective inter- faces that genuinely assist the biological researcher in his/her work is equally important, and has been neglected to date. Of course improving IE tech- niques, their accuracy and cross-domain portability are important research objectives for language tech- nology researchers. But given that the techniques will remain imperfect for the foreseeable future, we must also ask how biologists can benefit today from the limited capabilities of existing IE technology. In this paper we describe an approach to pro- viding effective access to the results of the Protein Active Site Template Acquisition (PASTA) system (Humphreys et al., 2000; Demetriou et al., 2002), an IE system that extracts information about amino acid residues in protein structures and their roles in protein active sites directly from the published lit- erature. To experiment with a mechanism for de- livering PASTA results to biologist end-users, we have developed the PASTAWeb interface, a WWW- based interface that offers search and browsing facil- ities to the extracted protein structure information, as well as to the original text sources (MEDLINE Association for Computational Linguistics. the Biomedical Domain, Philadelphia, July 2002, pp. 77-84. Proceedings of the Workshop on Natural Language Processing in abstracts). PASTAWeb provides transparent text ac- cess …

Content truncated to 500 words