Overview
This document is designed to provide an overview of the ORD’s scope and goals, high-level design, technical progress, and plans for access and interfacing.
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.
Interface
Getting the data
Anyone can download their own copy of the data and associated code (with or without a GitHub account). We expect and encourage researchers to download copies of the entire repository. Snapshots of the repository will be backed up to Figshare at regular intervals. Anyone with a GitHub account can submit data or code to the database.
Although we have defined the schema using Protocol Buffers, each reaction can be defined in a human-readable JSON or pbtxt format. The GitHub repository containing the official version of the database may use the proto binary format for storage efficiency and speed. Archived snapshots of the repository will convert examples to a human-readable format so that the data are more immediately accessible.
GitHub
The official Open Reaction Database repository is located here.
Important
We use Git LFS to efficiently store the Dataset records that make up the ORD. To view these files locally, you’ll need to install Git LFS before cloning the repository.
If you are planning to make submissions to the database, you should start by creating a fork of the ORD repository on GitHub. Otherwise, you can simply clone the repository directly.
You may also want to use --depth to create a
shallow clone
instead of fetching the entire history of the repository:
$ git clone --depth=1 "https://github.com/${GITHUB_USERNAME}/${REPOSITORY}"
After cloning your fork, set the upstream remote to track the official
database repository:
$ cd "${REPOSITORY}"
$ git remote add upstream https://github.com/Open-Reaction-Database/ord-data.git
Contributing to the database
Submissions to the ORD will be handled primarily via GitHub, as pull requests to a public repository governed by the CC-BY-SA license (see below). All data merged into the official repository (under the open-reaction-database GitHub organization) will then be hosted on GitHub under this license.
Note
The submission workflow is described in detail here.
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 (Google)
Matthew Gaunt (Cambridge, SynTech)
Alex Godfrey (NCATS Consultant)
Mimi Hii (Imperial, ROAR)
Greg Landrum (T5 Informatics)
Fabio Lima (Novartis)
Christos Nicolaou (Recursion)
Sarah Reisman (Caltech, C-CAS)
Francesco Rianjongdee (GSK)
Marwin Segler (Microsoft)
Matthew Sigman (Utah, C-CAS)
Jay Stevens (BMS)
Sarah Trice (Xtalpi)
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.