Machine Learning Techniques for Establishing the Provenance of Biological Interactions

A substantial amount of knowledge about biomolecular interactions resides in the biomedical literature and has been increasingly made available as structured, curated database entries. Although recent research in biological assertion extraction has mostly focused on automatically detecting interacting entities, automated generation of provenance for interacting pairs remains of equal importance. We show that a number of machine learning algorithms can be used to directly establish sentence-level support for given entity-entity interactions in biological databases. We specifically focus on finding support for specific interaction entries in database assertions about proteinprotein interactions.