Significant amounts of information on chemical compounds, their structures, and their properties have been published in scientific articles. Only a fraction of this knowledge is available in open databases. Retrospectively curating open data from books and journals automatically or semi-automatically, therefore, is a timely challenge [1].<br/>Towards this end, tools for extracting chemical structure depictions and converting them into computer-readable formats are needed. Such an Optical Chemical Structure Recognition (OCSR) tool translates the image of a chemical structure into a machine-readable representation. With DECIMER (Deep lEarning for Chemical ImagE Recognition) [2] an open-source automated software solution has been developed to address the OCSR problem through deep learning for image segmentation and recognition. DECIMER includes a deep-learning-based segmentation algorithm (DECIMER-Segmentation) [3] for automated recognition and segmentation of chemical structures from the scientific literature, as well as an OCSR engine (DECIMER-Image transformer) [4] with a deep learning model based on CNN + Transformer networks which can predict SMILES with over 90% accuracy from depictions of chemical structures.<br/>DECIMER can be applied to older articles before vector images were introduced to PDFs since it uses bitmap images of journal pages. Segmented images can be directly fed into the DECIMER-Image transformer to be converted into SMILES strings. Benchmark results on the OCSR benchmark datasets show that the DECIMER - Image transformer outperforms all currently available open-source algorithms for OCSR. DECIMER’s code is open source and the trained models are also openly available.

The DECIMER (Deep Learning for Chemical Image Recognition) project

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.