Health and life sciences' research fields make an intensive use of graphical information to represent complex relations between biological entities, called molecular pathways. Interpretation of molecular pathway diagrams requires domain-specific knowledge to remove ambiguity. We propose a fully automatic method that detects entities and extracts their relations from diagrams. It uses image analysis and a domain-specific cognitive model and automatically produces a structured textual version of the content. Results on an annotated dataset show precision of 0.99 and recall of up to 0.85 for the detection of molecular entities and interactions.