Literature-Based Knowledge Discovery Using Link Analysis

Authors: Ning Shang

Primary Advisor: Jorge Herskovic, MD, PhD (co-author)

Committee Members: Manueal Wahle, MS (co-author)

Masters thesis, The University of Texas School of Biomedical Informatics at Houston.

Abstract:

Hypothesis evaluation is time intensive and demanding of expert knowledge, and there is a need for automating the process. Closed Literature-Based Discovery (LBD) can support hypothesis evaluation. In this paper, we leverage a predicate graph built from semantic predications extracted from MEDLINE. Link analysis of this predicate graph using PageRank uncovers potentially valuable hypothesis-supporting concepts. We develop a framework for closed literature discovery to support the manual review process. In this case study, we show the usefulness and feasibility of our method by investigating a classical example of LBD, Swanson's discovery of the link between fish oil and Raynaud’s in 1986. Our method was able to replicate Swanson's findings. The conclusion is that the PageRank link analysis algorithm can be used for hypothesis evaluation in closed LBD.