Scientific figures, captions and accompanying text provide a valuable resource that comprise the evidence generated by a published scientific study. Extracting information pertaining to that evidence requires a pipeline made up of several intermediate steps. We describe machine reading analysis applied to papers that had been curated into the European Bioinformatics Institute's INTACT database describing molecular interactions. We unpack multiple steps in an extraction pipeline that ultimately attempts to identify the type of experiments being performed automatically. We apply machine vision and natural language processing to classify figures and their associated text based on the type of methods used in the experiment to a level of accuracy that can likely support future biocuration tasks.