About the Open Reaction Database#
We are building an open access chemical reaction database to support machine learning and related efforts in reaction prediction, chemical synthesis planning, and experiment design. Our initial meeting took place on 31 October 2019 and included experts from pharma, academia, and tech.
We expect that this database will be the starting point for the development of best-in-class tools and models for reaction prediction and synthesis planning. Additionally, we hope it will serve as a basis for experimental efforts in industry and academia (e.g., to reduce duplication or focus data generation on underrepresented areas).
The ORD is described in JACS and was highlighted in C&EN.
Goals#
Our overarching goal, stated above, is to “support machine learning and related efforts in reaction prediction, chemical synthesis planning, and experiment design”.
We aim to:
Provide a structured data format for chemical reaction data
Provide an interface for easy browsing and downloading of data
Make reaction data freely and publicly available for anyone to use
Encourage sharing of precompetitive proprietary data, especially HTE data
We conducted a survey in late 2019/early 2020 to help define the scope and use cases. With the help of 172 respondents, 93.5% of whom reported having a chemistry background and 20.8% a computer science background, we have established our focus on single-step organic reactions.
We aim to accommodate data relevant to medicinal chemistry, process chemistry, flow chemistry, photochemistry, and electrochemistry. Time-course data, unstructured analytical data, and metadata about how the reaction was performed will also be accepted. Reactive molecular dynamics simulations, gas-phase reaction kinetics, and electronic structure calculations for molecular featurization are being left for other initiatives.
Additionally, we want to be clear that since the database is in active development, some features not considered “must-haves” for the initial deployment may not be supported immediately. We welcome feedback about features that make or break your excitement in and use of the database.
Non-goals#
Our non-goals for this initiative, at least right now, are to:
Capture structured data in a manner designed for programmatic execution on automated synthesis hardware (i.e. treating reactions as action sequences)
Store processed analytical data (e.g. NMR peak assignments) other than summary statistics (e.g. conversion, yield, purity, selectivity); note that unprocessed data (e.g. an exported LCMS file) will be stored
Integrate model building or external datasets as part of the database
There are some practical consequences of these non-goals:
Because we are structuring reactions as single-step reactions, there are some more complex operations that cannot be captured in a structured format by the schema. For example, this OrgSyn example describes a procedure whereby three components are mixed in a chilled vessel on an ice bath, a fourth component is added and stirred for several minutes still on ice, and then the vessel is removed from the ice bath and allowed to warm up to room temperature over several hours. This temperature ramp will be captured in a free text field, but the structured temperature field must record either zero Celsius or room temperature.
Analytical data will not be in a unified format and would be difficult to learn from directly. This might represent a missed opportunity to train, e.g. structural elucidation models that predict a molecular structure on the basis of spectral data. Users will still be encouraged to record processed data in a text format. Unprocessed data will be stored in its original file format as exported by analytical instruments.
Automated synthesis efforts will require additional future work or a separate database, which could lead to fragmentation in the community. However, we expect that the reaction procedure captured in the ORD will be able to be converted to action sequences through simple translation scripts.
Leadership#
Governing Committee#
The ORD is governed by a Governing Committee with representatives from many industrial and academic institutions. This committee reviews all aspects of the database, from the underlying structured data representation to the public interface(s) to promotion and publicity. The current membership of the governing committee is:
Advisory Board#
We are forming an Advisory Board to include representatives from many institutions and industry segments. The primary role of the Advisory Board is to encourage community engagement with the database. The current membership of the Advisory Board is:
Alán Aspuru-Guzik (Toronto, MADNESS)
Timothy Cernak (Michigan, Entos)
Lucy Colwell (Cambridge, SynTech, Google)
Werngard Czechtizky (AstraZeneca)
JW Feng (Consultant)
Matthew Gaunt (Cambridge, SynTech)
Alex Godfrey (Consultant)
Mimi Hii (A*STAR, Imperial, ROAR)
Greg Landrum (ETH Zurich)
Fabio Lima (Novartis)
Christos Nicolaou (Novo Nordisk)
Sarah Reisman (Caltech, C-CAS)
Francesco Rianjongdee
Marwin Segler (Microsoft)
Matthew Sigman (Utah, C-CAS)
Jay Stevens (BMS)
Sarah Trice
Huimin Zhao (UIUC, MMLI)
Commitment to Open Access#
As the name of the initiative suggests, this will be an open database in every sense of the word. All data and code associated with the database will be made publicly available under commonly used licenses that protect open access.
When proprietary tools are used (see Interfaces and tools), they will only be used to provide “extra” functionality that is not part of the core data or the code responsible for data validation and processing. This extra functionality will be made publicly accessible on the web for anyone to access.
The database is purposely designed to avoid the control or influence of a single institution. This also ensures that the core data and functionality of the database will not be affected by any contributor choosing to cease their involvement in the initiative.
All data submitted to the database will be made available under the CC-BY-SA license, a well-known license for creative works. Additionally, the various software tools developed for the database will be made available under the Apache license; this is another well-known and OSI-approved license that is used by many organizations around the world.
How to help#
Additional technical help will also be required to aid in tasks such as processing submissions and implementing the various tools and interfaces to improve the user experience. We welcome any donation of time to improve using the schema, technical infrastructure, front-end work, etc.; or simply to provide feedback on the user experience.
If you are interested in receiving updates or participating in future meetings, please request to join the open-reaction-database mailing list.
Past issues of the ORD newsletter are on LinkedIn, where you can also subscribe.
How is the Open Reaction Database being used?#
We have a public Zotero Group and library where we keep track of Open Reaction Database citations, and also the publications which have reactions in the database.
Note
Have we missed something? Please email help@open-reaction-database.org if you find published work which employed the Open Reaction Database data, schema, or other tools, and is not already included in the library.
Note
The Datasets sub-collection does not include the individual USPTO patents (there is over 13k of them) for clarity.
Frequently Asked Questions#
How will my institution benefit from giving others access to our data?
The ORD is an open science initiative. Your institution will be able to access the data shared by all other contributors as well. We view this sharing as no different from how we currently share data when we publish scientific articles, except the data will be available in an accessible, standardized format.
What if <institution> decides to stop participating in the initiative?
The ORD is not reliant on the participation of any individual organization. However, hosting interfaces and tools to enhance its functionality requires computational resources; if an organization that provides them stops participating and the Governing Committee believes these tools to be important to the adoption and usability of the ORD, the committee will need to seek funding for web hosting.
Will my institution have a right to use any tool developed on our data?
No. The data itself will have a share-alike clause so that derivative work must be shared under identical terms. Anyone is free to develop software tools, algorithms, trained machine learning models, etc. under a different license as long as they follow the terms for the underlying data.
Can I simply fork your repositories and create my own version of the database?
Yes! As long as you conform to the terms of the licenses.