The Open Reaction Database

The Open Reaction Database (ORD) is an initiative aimed at addressing the fragmentation of chemical reaction data. Currently, most reaction information is locked in unstructured formats like journal articles, patents, and electronic lab notebooks, which hinders the development of predictive chemistry models. The ORD provides a centralized, open-access repository and a standardized schema that allows researchers to record, share, and search for organic reaction data in a machine-readable format.

The ORD schema is built using Protocol Buffers, which offers a robust way to define structured data fields while allowing for flexibility. The schema is organized into nine logical sections, including reaction inputs, conditions, setup, outcomes, and provenance. By enforcing structured inputs—such as specific units for amounts and defined roles for chemical components—the schema ensures consistency across entries.

To lower the barrier to entry, the project provides an interactive web-based editor for those who prefer not to use programmatic submission methods (e.g., Python scripts). The database also includes a search interface that allows users to query reactions by chemical structure (SMILES/SMARTS), DOI, or other metadata, making it easier to retrieve high-quality data for training machine learning models.

Predictive synthesis, including retrosynthesis planning and reaction condition optimization, is currently limited by the lack of large, high-quality, and standardized datasets. By adopting the FAIR (Findable, Accessible, Interoperable, and Reusable) principles, the ORD aims to create a culture of data sharing that goes beyond traditional publication requirements. The authors argue that moving away from unstructured PDF-based reporting toward structured data will significantly accelerate the development of next-generation predictive chemistry tools and improve the reproducibility of chemical research.