Automating Literature Surveillance
Today, if researchers want to study complex relationships among genes, diseases and drugs, they have to hope that human curators have read the scientific literature, extracted the relevant information, and put it in a database. “It would be a lot more efficient if computers could perform that surveillance of the literature for us,” says Beth Percha, a graduate student working with Russ Altman, PhD, at Stanford University.
In recent work, Percha and Altman made steps toward that goal, effectively extracting drug-gene relationships from the literature and clustering them in ways that proved meaningful (see dendrogram caption). Percha is also applying the same method to other situations such as gene-disease and disease-drug relationships. Ultimately, she’d like to be able predict drug-drug interactions based on drug-gene relationships automatically extracted from the literature.
“The dendrogram is pretty and it’s a good sanity check because it reproduces knowledge we already have,” Percha says. “But what’s exciting is to be able to discover new relationships from the literature quickly, cheaply and without a ton of human effort.”