PharmKi: A Retrieval System of Chemical Structural Formula Based on Graph Similarity

Abstract

Different from conventional media type, chemical structural formula (CSF) is a primary search target as a unique identifier for each compound in the research field of medical information retrieval. This paper introduces a graph-based CSF retrieval system, PharmKi, accepting the photos taken from smartphones and the sketches drawn on tablet PCs as inputs. To establish a compact yet efficient hypergraph representation for molecules, we propose a graph-isomorphism-based algorithm for evaluating the spatial similarity among graphical CSFs, as well as selecting dominant acyclic subgraphs on the basis of overlapping analysis. The results of comparative study demonstrate that the proposed method outperforms the existing methods with regard to accuracy and efficiency.