UPDATE: This repository has been moved to the NeoChemSynthWave organization, and will be removed shortly.
The computer-assisted synthesis field grew to become one of the most active fields in chemoinformatics, with new and improved approaches centered around chemical reactions being published on a regular basis. To understand the gist of a chemical reaction, it is essential to know how the atoms of the participating compounds re-arrange during the chemical transformation. This process is widely known as atom mapping, and it has proven to be quite challenging. Taking all of this into account, the goal of this project is to systematically curate, categorize and facilitate the access to existing open-source chemical reaction atom mapping libraries in one place.
Author: Haris Hasić, M.Eng.
Affiliation:
Dr.Sci. Student @
Ishida Laboratory, Department of Computer Science, Tokyo Institute of Technology &&
Research Engineer @ Elix, Inc.
Current Version: 2023.1.0
To use the chemical_reaction_atom_mapping package, please ensure that the chytorch-rxnmap, epam.indigo, rxnmapper and tqdm libraries are available. A minimal execution environment can be set up using conda and pip as follows:
conda create -c conda-forge -n chemical-reaction-atom-mapping python=3.8 rdkit -y
conda activate chemical-reaction-atom-mapping
pip install chython chytorch chytorch-rxnmap epam.indigo rxnmapper tqdmThe scripts directory is primarily meant to illustrate how to utilize the chemical_reaction_atom_mapping
package to run the atom mapping procedures on chemical reaction data. The first line of each script is just a reminder
to add the path to the project root directory to the PYTHONPATH variable to make the package visible before running
the scripts.
Currently, the chemical_reaction_atom_mapping package supports the following open-source chemical reaction atom mapping libraries:
- The EPAM Indigo library incorporates a number of unique algorithms developed by EPAM, as well as some standard algorithms well-known in the cheminformatics world. [1]
- The RXNMapper library is constructed around a chemically agnostic attention-guided reaction mapper transformer model that shows a remarkable performance in terms of accuracy and speed, even for strongly imbalanced reactions. [2]
- The Chytorch RxnMap library is constructed around a transformer neural network adopted for the direct processing of molecular graphs as sets of atoms and bonds, as opposed to SMILES/SELFIES sequence-based approaches, in combination with the Bidirectional Encoder Representations from Transformers (BERT) network. [3]
This project is published under the MIT License. For more details on the license information of individual chemical reaction atom mapping libraries, please refer to the original publications.
If you are interested in contributing to this project by reporting bugs, submitting feedback or anything else that might be beneficial, please feel free to do so via GitHub issues or e-mail. Also, check out the latest Elix, Inc. research.
- EPAM Indigo Library: https://lifescience.opensource.epam.com/indigo/index.html. Accessed on: May 18th, 2023.
- Schwaller, P., Hoover, B., Reymond, J., Strobelt, H., and Laino, T. Extraction of Organic Chemistry Grammar from Unsupervised Learning of Chemical Reactions. Sci. Adv., 2021, 7, 15. DOI: https://doi.org/10.1126/sciadv.abe4166.
- Nugmanov, R., Dyubankova, N., Gedich, A., and Wegner, J.K. Bidirectional Graphormer for Reactivity Understanding: Neural Network Trained to Reaction Atom-to-Atom Mapping Task, J. Chem. Inf. Model., 2022, 62, 14, 3307–3315. DOI: https://doi.org/10.1021/acs.jcim.2c00344.