unixjunkie / aizynthtrain

Tools to train synthesis prediction models

aizynthtrain

aizynthtrain is a collection of routines, configurations and pipelines for training synthesis prediction models for the AiZynthFinder software.

Prerequisites

Before you begin, ensure you have met the following requirements:

• Linux, Windows or macOS platforms are supported - as long as the dependencies are supported on these platforms.

• You have installed anaconda or miniconda with python 3.9 - 3.10

The tool has been developed on a Linux platform.

Installation

First clone the repository using Git.

Then execute the following commands in the root of the repository

conda env create -f env-dev.yml
conda activate aizynthtrain
poetry install

the aizynthtrain package is now installed in editable mode.

Now, you need to create a Jupyter Kernel for this environment

conda activate aizynthtrain
ipython kernel install --name "aizynthtrain" --user

Usage

There is a number of example configurations and SLURM scripts in the configs/uspto folder that was used to retrain the USPTO-based expansion model for AiZynthFinder.

The example configurations can be used as-is to reproduce the modelling, whereas the SLURM scripts migt have to be adjusted to your system.

To use the given configurations, follow this procedure:

For the training and validation of the expansion models, you need a number of data files that can be downloaded from Zenodo. In addition the pipelines create a lof of artifcats and therefore, you will start with setting up a folder for your modelling

1. Create a new folder where the pipelines will be executed using

   python configs/uspto/setup_folder.py PATH_TO_YOUR_FOLDER

2. cd PATH_TO_YOUR_FOLDER

3. Perform the pipelines from the rxnutils package to download and prepare the USPTO datasets for modelling: https://molecularai.github.io/reaction_utils/uspto.html

The rxnutils package pipelines should have produced a file uspto_data_mapped.csv that will be used as the starting point of the template-extraction pipeline

4. Adapt and execute the configs/uspto/template_pipeline.sh SLURM script on your machine.

The template-extraction pipeline will produce a number of artifacts, the most important being

• reaction_selection_report.html that is the report of the reaction selection
• template_selection_report.html that is the report of the template selection
• uspto_template_library.csv and uspto_ringbreaker_template_library.csv that are the final reactions and reaction templates that will be used to train the expansion model.

5. Adapt and execute the configs/uspto/expansion_pipeline.sh SLURM script on your machine.

The expansion model pipeline will produce these important artifacts

• uspto_expansion_model_report.html that is the report of the model training and validation
• uspto_keras_model.hdf5 the trained Keras model
• uspto_unique_templates.csv.gz the template library for AiZynthFinder

You can also execute the configs/uspto/ringbreaker_pipeline.sh to train a RingBreaker model.

Testing

Tests uses the pytest package, and is installed by poetry

Run the tests using:

pytest -v

Contributing

We welcome contributions, in the form of issues or pull requests.

If you have a question or want to report a bug, please submit an issue.

To contribute with code to the project, follow these steps:

1. Fork this repository.
2. Create a branch: git checkout -b <branch_name>.
3. Make your changes and commit them: git commit -m '<commit_message>'
4. Push to the remote branch: git push
5. Create the pull request.

Please use black package for formatting, and follow pep8 style guide.

Contributors

• @SGenheden

The contributors have limited time for support questions, but please do not hesitate to submit an issue (see above).

License

The software is licensed under the Appache 2.0 license (see LICENSE file), and is free and provided as-is.

References

1. Genheden S, Norrby PO, Engkvist O (2022) AiZynthTrain: robust, reproducible, and extensible pipelines for training synthesis prediction models. ChemRxiv. Prerint. https://doi.org/10.26434/chemrxiv-2022-kls5q
2. Kannas C, Thakkar A, Bjerrum E, Genheden S (2022) rxnutils – A Cheminformatics Python Library for Manipulating Chemical Reaction Data. ChemRxiv. https://doi.org/10.26434/chemrxiv-2022-wt440-v2
3. Genheden S, Thakkar A, Chadimova V, et al (2020) AiZynthFinder: a fast, robust and flexible open-source software for retrosynthetic planning. J. Cheminf. https://jcheminf.biomedcentral.com/articles/10.1186/s13321-020-00472-1
4. Thakkar A, Kogej T, Reymond J-L, et al (2019) Datasets and their influence on the development of computer assisted synthesis planning tools in the pharmaceutical domain. Chem Sci. https://doi.org/10.1039/C9SC04944D