Image-based automated chemical database annotation with ensemble of machine-vision classifiers

This paper presents an image-based annotation strategy for automated annotation of chemical databases. The proposed strategy is based on the use of a machine vision-based classifier for extracting a 2D chemical structure diagram in research articles and converting them into standard chemical file formats, a virtual “Chemical Expert” system for screening the converted structures based on the level of estimated conversion accuracy, and a fragment-based measure for calculation intermolecular similarity. In particular, in order to overcome limited accuracies of individual machine-vision classifier, inspired by ensemble methods in machine learning, it is attempted to use of the ensemble of machine-vision classifiers. For annotation, calculated chemical similarity between the converted structures and entries in a virtual small molecule database is used to establish the links. Annotation test to link 121 journal articles to entries in PubChem database demonstrates that ensemble approach increases the coverage of annotation, while keeping the annotation quality (e.g., recall and precision rates) comparable to using a single machine-vision classifier.