USING ARROWSMITH - A COMPUTER-ASSISTED APPROACH TO FORMULATING AND ASSESSING SCIENTIFIC HYPOTHESES

Conventional computer searches of the biomedical literature (e.g. MEDL INE) allow investigators to retrieve much of the information that has already been published on a given topic. However, these searches are o f limited utility at the frontier of scientific discovery, when one wi shes to identify and assess new, untested scientific hypotheses, or to uncover biologically significant relations between two previously dis parate fields of inquiry. We have designed a set of interactive softwa re and database search strategies, collectively called ARROWSMITH, tha t facilitate the discovery of plausible hypotheses linking findings ac ross specialties (Artif. Intell. 91 (1997) 183-203). In the simplest i mplementation of ARROWSMITH, the user begins with an experimental find ing or hypothesis that two items A and C are related in some way. The titles of papers indexed in MEDLINE which contain the word 'A' (or syn onyms) are downloaded into a file A, and similarly a file C is created . The software constructs a list of words and phrases B common to file s A and C; automatic and manual editing are used to filter out uninter esting B-terms. For each B-term, the software generates an AB file of titles containing both 'A' and 'B', and a BC file of titles containing both 'B' and 'C'; these titles are juxtaposed to facilitate the user judging whether there is likely to be a biologically significant relat ion among A, B and C. ARROWSMITH has been employed to analyze research problems relating to oxidative stress, brain damage, Alzheimer's dise ase and schizophrenia. Applications of ARROWSMITH include: anticipatin g adverse drug reactions, identifying mechanisms by which agents modul ate cellular or organismal responses, suggesting new therapeutic appro aches, identifying possible risk factors for diseases, and identifying potential animal models for human conditions. A simplified experiment al version of ARROWSMITH is now freely accessible on the World Wide We b (http://kiwi.uchicago.edu). (C) 1998 Elsevier Science Ireland Ltd. A ll rights reserved.