Poster Sessions

BIOINFO-19

Daniel Russ

D. Russ, F. Austin, C. Roth, C. Johnson

Preliminary Assessment of High-risk, High-reward Research at NIH Using Context-sensitive Text Mining

The NIH Reform Act of 2006 authorized the NIH to support High-Risk, High-Reward (HRHR) Demonstration projects and evaluate these activities. In response, we developed a context-sensitive text mining tool to identify HRHR research from NIH peer-review summary statements. We present the results of an analysis of 1,896 summary statements from investigator-initiated \"unsolicited\" grant applications submitted during the Oct. 2007 and Jan. 2008 NHLBI Council Rounds. Internal \"reviewers\" classified a sample consisting of 200 summary statements previously labeled by the text mining tool. The sample contained 50 HRHR and 50 non-HRHR summary statements from each of the council rounds. We show that the text mining tool can identify HRHR research with high recall (~0.94) and adequate precision (~0.47). In addition, preliminary results show that HRHR research fares better than non-HRHR during the peer review process.