Northeastern University is awarded a grant to develop a system that analyzes figures in biological literature. The Biological Knowledge available online goes well beyond the data that is currently collected to populate databases. Much of the critical knowledge contained in the figures and text of published papers deals with complex biological systems such as cells and organisms and with aspects of experiments and results such as methods, observations, judgments, explanations, etc. Approximately 50% of the content of a typical Biology research paper is comprised of figures and discussions of figure content. The system developed first analyzes the figures and the text to build indexes of their content and figure-text interrelations. Visually based interactive systems for users are built that allow them to drill-down to specific information in individual papers, starting with high-level exemplars. The project uses 10,000 or more papers from the Open Access publisher, BioMed Central. This allows all the resulting knowledge bases and systems to be freely distributed, unimpeded by copyright issues. The methods for doing this are extensions of the PI's successful Diagram Understanding System amended to use clustering and modern machine learning techniques such as boosting. The approach to text is equally novel. The text patterns are discovered by automated information-theory-based techniques combined with computational linguistics. The patterns are collected into a Framework Bank, much in the tradition of WordNet, FrameNet, the Proposition Bank, and the Gene Ontology. Users of the systems can range from students to researchers and may include the general public, with its growing interest in biomedical research results. The heavy involvement of undergraduate students in Computer Science and Biology, including women and minorities, will continue to inspire them about careers in research.
