Challenges in adapting text mining for full text articles to assist pathway curation

Abstract

Annotation of biological pathway databases is largely driven by manual effort with little assistance from text mining. It is a great challenge to the pathway curators to keep up with the pace of ever-growing literature. There have been recent efforts to fill this gap through text mining by identifying the relevant papers and the textual evidence pertaining to pathway information. In the current work, we evaluated the performance of a text mining system that extracts events describing molecular pathways from full text articles and its potential role in assisting manual curation of pathway databases. We specifically investigated the merits of mining full text articles for extracting pathway events by comparing the performance of our system on both full text articles and biomedical abstracts. From the preliminary results, we observed that by processing full text articles the performance of the system improved by nearly 22% against a small drop of 5% in the precision in comparison against the extractions from PubMed abstracts. Preliminary analysis of the text mining results for selected pathways from PharmGKB suggest that the pathway curators do use their biological knowledge to infer new information that go beyond what is often expressed in either the full text articles or abstracts. This study is an attempt to identify the magnitude of gaps that exist between the text mining deliverables and the demands of pathway curation.