Conventional search engines (e.g., PubMed) can identify many facets of information contained in Medline records and give an indication of the current state of knowledge on a given topic. However, such queries are insufficient for investigators operating at the frontiers of scientific discovery, who need to assess the possible significance of new findings or hypothesized relationships that have not yet been experimentally tested. """"""""Arrowsmith"""""""" is a computer-assisted strategy designed to address this problem by facilitating the identification of information that is present implicitly within or across databases. It does this by identifying title words that are shared across a pair of disjoint literatures, filtering out words that are likely to be non-informative, and justaposing titles of papers in each literature that share a given title word, allowing the user to assess whether the two sets of papers are suggestive of a biologically meaningful inference when considered together. To test the feasibility of Arrowsmith for use by the scientific community, a set of field testers will be trained to evaluate the feasibility and utility of Arrowsmith analyses when used in the context of multi-disciplinary neuroscience groups that have both significant laboratory and informatics components. The utility of linking papers via shared terms taken from other fields of Medline records, e.g., abstract words or MeSH subject headings, will also be evaluated. Access to Arrowsmith will be provided via a Web site that will be upgraded and expanded to accommodate the larger public as well. Finally, it will be assessed whether Arrowsmith analyses can help facilitate new interdisciplinary research by identifying and alerting complementary groups of investigators who independently are tackling problems of joint interest, that may be best addressed by working together. Because of the generality of the Arrowsmith approach, it may be applied to a wide range of biomedical problems and, indeed, to conduct searches within or across any databases that contain textual material.
Showing the most recent 10 out of 22 publications
