Deep Chem: An AI-Powered Tool for Automated Recognition of Chemical Structures and Predicting their Impact on Health

Abstract: It has become nearly impossible to extract and retain chemical structure data from earlier papers that are only available in printed or scanned form as a result of the expanding availability of chemical structure data. Although systems for rule-based optical chemical structure recognition (OCSR) have been created to automate this procedure, they have drawbacks including being sluggish and prone to mistakes. The Deep Chem project has developed an application for recognizing chemical structures in order to address this problem and comparing the impact of these chemical structures in different foods on human health based on their molar mass. The app uses deep learning approaches to automate the recognition of chemical structures from printed or scanned articles. Traditional rule-based optical chemical structure recognition (OCSR) tools can be slow and prone to errors, but the Deep Chem app aims to provide a faster and more reliable solution. A collection of 50–100 million molecules were used to predict SMILES encodings of chemical structure renderings with over 96% accuracy for structures without stereochemical information and over 89% accuracy for structures with stereochemistry information. Additionally, the app provides a feature to compare the impact of different foods on health based on their molar mass. This work is entirely based on open-source software and open data and is available to the general public for any purpose.