How Complex are Complex Protein-protein Interactions?

The extraction of protein-protein interactions (PPI) from text requires a formal PPI representation. We use the BioInfer and GENIA corpora to study two such representations: a “binary” interaction model consisting of pairs of proteins and a “complex” model where interactions are defined as a network of proteins and their relations. As both of these formats can be seen as graphs, we contrast them with syntactic dependency graphs, a common tool for PPI extraction. We find that unlike binary interactions, complex interactions closely resemble dependency parses, especially those in the Stanford scheme. We therefore argue that despite appearances, complex interactions might be easier to extract. We also notice the similarity between the independently developed BioInfer and GENIA interaction representations and the Stanford dependency scheme. This suggests an emerging consensus on the representation for complex PPI, supporting the value of these tools and resources for PPI extraction.