ChemWARD : extracting chemical structure from printed diagrams

Abstract

Over the years, a vast amount of literature in the field of chemistry has accumulated, and searching for documents about specific molecules is a formidable task. To the extent that the literature is textual, services like Google enable relatively easy search. While search indexes like Google are very good at finding such things, its difficult to describe molecules completely using text because text can't easily indicate molecular structure, and molecular structure defines chemical properties. ChemWARD is a system that extracts the molecular structure from the printed diagrams that are ubiquitous in chemistry literature and converts them to a machine readable format in order to allow chemists to search the literature by drawing a molecular structure instead of typing a chemical formula. We describe the architecture of the system and report on its performance, demonstrating its ability to achieve an overall accuracy rate of 85.5% on printed diagrams extracted from published chemical literature.