SynPlanner is an open-source tool for retrosynthetic planning,
designed to increase flexibility in training and developing
customized retrosynthetic planning solutions from raw chemical data.
It integrates Monte Carlo Tree Search (MCTS) with graph neural networks
to evaluate applicable reaction rules (policy network) and
the synthesizability of intermediate products (value network).
SynPlanner can be used for:
- ⚒️ Standardizing and filtering reaction data
- 📑 Extracting reaction rules (templates) with various options
- 🧠 Training policy and value networks using supervised and reinforcement learning
- 🔍 Performing retrosynthetic planning with different MCTS-based search strategies
- 🖼️ Visualising found synthetic paths and working with graphical user interface
SynPlanner can be installed using conda/mamba package managers.
For more information on conda installation please refer to the
official documentation.
To install SynPlanner, first clone the repository and move the package directory:
git clone https://github.com/Laboratoire-de-Chemoinformatique/SynPlanner.git
cd SynPlanner/Next, create SynPlanner environment with synplan_env_linux.yaml file:
conda env create -f conda/synplan_env_linux.yaml
conda activate synplan_env
pip install .After installation, SynPlanner can be added to Jupyter platform:
conda install ipykernel
python -m ipykernel install --user --name synplan_env --display-name "synplan"
Colab tutorials do not require the local installation of SynPlanner but their performance is limited by available computational resources in Google Colab
Currently, two tutorials are available:
- Retrosynthetic planning can be used for retrosynthetic planning of any target molecule with pre-trained retrosynthetic models and advanced analysis of the search tree.
- SynPlanner benchmarking can be used for retrosynthetic planning of many target molecules for benchmarking or comparison analysis.
Jupyter Tutorials requires the local installation of SynPlanner but can be executed with advanced computational resources on local servers
Currently, five tutorials are available:
Quick-start tutorials. These tutorials can be used for easy execution of the default SynPlanner pipeline:
- SynPlanner pipeline presents the full pipeline of SynPlanner starting from raw reaction data and resulting in ready-to-use retrosynthetic planning.
Advanced tutorials. These tutorials provide advanced explanations and options for each step in the SynPlanner pipeline:
- Step 1: Reaction data curation can be used for reaction standardization and reaction filtration.
- Step 2: Reaction rules extraction can be used for extracting reaction rules from curated reaction data.
- Step 3: Policy network training can be used for policy network training.
- Step 4: Retrosynthetic planning can be used for retrosynthetic planning.
SynPlanner pipeline can be accessed by neat command-line interface (CLI). For example, retrosynthetic planning of several target molecules with pre-trained models can performed with the following commands:
synplan download_all_data --save_to synplan_data
synplan planning --config configs/planning.yaml --targets synplan_data/benchmarks/sascore/targets_with_sascore_1.5_2.5.smi --reaction_rules synplan_data/uspto/uspto_reaction_rules.pickle --building_blocks synplan_data/building_blocks/building_blocks_em_sa_ln.smi --policy_network synplan_data/uspto/weights/ranking_policy_network.ckpt --results_dir planning_resultsMore details about CLI can be found in SynPlanner Documentaion
Contributions are welcome, in the form of issues or pull requests.
If you have a question or want to report a bug, please submit an issue.