The abundant R-group information available in chemical publications plays a crucial role in data-driven artificial intelligence (AI) research in the field of medicinal chemistry. In real-world publications, R-groups are expressed in various textual and graphical forms, thereby rendering their manual integration labor-intensive and inefficient. Although automated tools exist for R-group recognition, they remain underdeveloped, creating a clear requirement for precise and comprehensive automated parsing tools. This paper presents RGReco, a novel framework that combines deep learning and chemical rules to parse and integrate R-group information from images and text through a multistage pipeline. In addition, a new process for recognizing substituent structures and parsing-related text is proposed. To evaluate the performance of RGReco, a dataset containing common types of R-group images was constructed from real-world scientific literature. Using this dataset, RGReco achieved a precision of 86.4%, a recall of 79.7%, and an F1 score of 82.9%. These results demonstrate that RGReco effectively handles the diversity of R-group images in real-world scenarios, offering researchers a new technological tool for accelerating the extraction of chemical information.
XIANG et al. (Tue,) studied this question.