Overview of the BioCreative VI chemical-protein interaction Track

Despite the considerable number of available systems that recognize automatically mentions of genes/proteins and chemicals in text, only a limited number of attempts were made so far to extract interactions between them. Most biomedical relation extraction systems focus on the extraction of protein-protein or gene/chemical-disease relations. The detection of interactions between drugs and proteins/genes is of key relevance for pharmacological and clinical research, playing an important role for drug discovery, understanding of molecular mechanism of adverse drug reactions, describing drug metabolism or drawing regulatory networks of importance for systems pharmacology. The BioCreative VI ChemProt track represents the first attempt to promote the development of systems for extracting chemical-protein interactions (CPIs), of relevance for precision medicine as well as for drug discovery and basic biomedical research. The novel ChemProt corpus consists of text exhaustively annotated by hand with mentions of chemical compounds/drugs and genes/proteins, as well as 22 different types of compound-protein relations. To focus on a subset of important relations, 5 relation classes were used for evaluation purposes, including agonist, antagonist, inhibitor, activator and substrate/product relations. A total of 13 participating teams returned 45 runs for this track. Despite the biological complexity of the considered relation types, top-scoring teams could obtain an F-measure across relation classes of 64.10%. Performance varied depending on the relation class: for the antagonist relation class the best team obtained an F-measure of 72.56% (precision of 80.75%, recall of 65.87%) while for inhibition/down-regulation the best value was of 71.48% (with a precision of 76.51% and a recall of 67.07%). Keywords—text mining; chemical compound; drug; protein; drug target; agonist; antagonist, inhibitor; activator; gene regulation; chemical-protein relation