aizynthtrain is a collection of routines, configurations and pipelines for training synthesis prediction models for the AiZynthFinder software.
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.
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 installthe 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" --userThere 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
Create a new folder where the pipelines will be executed using
python configs/uspto/setup_folder.py PATH_TO_YOUR_FOLDER
cd PATH_TO_YOUR_FOLDER
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
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 selectiontemplate_selection_report.html that is the report of the template selectionuspto_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.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 validationuspto_keras_model.hdf5 the trained Keras modeluspto_unique_templates.csv.gz the template library for AiZynthFinderYou can also execute the configs/uspto/ringbreaker_pipeline.sh to train a RingBreaker model.
Tests uses the pytest package, and is installed by poetry
Run the tests using:
pytest -vWe 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:
git checkout -b <branch_name>.git commit -m '<commit_message>'git pushPlease use black package for formatting, and follow pep8 style guide.
The contributors have limited time for support questions, but please do not hesitate to submit an issue (see above).
The software is licensed under the Appache 2.0 license (see LICENSE file), and is free and provided as-is.