Synto - SYNthesis planning TOol ======== Synto is a tool for chemical synthesis planning based on Monte-Carlo Tree Search (MCTS) with various implementations of policy and value functions.

Installation

Important: all versions require python from 3.8 and up to 3.10!

Linux distributions ^^^^^^^^^^^

Dev: Installation with conda-lock """"""

conda-lock is a tool used for creating deterministic environment specifications for conda environments. This is useful for ensuring consistent environments across different machines or at different times. To install conda-lock, follow these steps:

1. Install conda-lock

You need to have conda or mamba installed on your system to install conda-lock. If you have not installed conda yet, you can download it from Miniconda or Anaconda.

wget "https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh" bash Miniforge3-$(uname)-$(uname -m).sh

Once conda is installed, you can install conda-lock by running:

conda install -c conda-forge -n base conda-lock

or, if you are using `mamba`:

mamba install -c conda-forge -n base conda-lock

2. Install the environment using the conda-lock file

Once you have a .conda-lock file, you can create a conda environment that exactly matches the specifications in the lock file. To do this, use:

conda-lock install -n synto --file conda-linux64-GPU-lock.yml

This command will read the .conda-lock file and create an environment with the exact package versions specified in the file.

Note

Make sure that the .conda-lock file is in your current working directory or provide the path to the file when using the conda-lock install command.

Dev: Installation with poetry """"""

It requires only poetry 1.3.2. To install poetry, follow the example below, or the instructions on https://python-poetry.org/docs/#installation

For example, on Ubuntu we can install miniconda and set an environment in which we will install poetry with the following commands:

# install miniconda
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

# install poetry
conda create -n synto -c conda-forge "poetry=1.3.2" "python=3.10" -y
conda activate synto

# install Synto
git clone https://github.com/Laboratoire-de-Chemoinformatique/Synto.git

# navigate to the Synto folder and run the following command:
cd Synto/
poetry install --with cpu

If Poetry fails with error, a possible solution is to update the bashrc file with the following command:

echo 'export PYTHON_KEYRING_BACKEND=keyring.backends.null.Keyring' >> ~/.bashrc
exec "bash"

Optional

After installation, one can add the Synto environment in their Jupyter platform:

python -m ipykernel install --user --name synto --display-name "synto"

Usage

The usage is mostly optimized for the command line interface. Here are some implemented commands:

• synto_planning
• synto_training
• synto_extract_rules
• synto_policy_training
• synto_self_tuning

Each command has a description that can be called with command --help

Run retrosynthetic planning ^^^^^^^^^^^ .. code-block:: bash

synto_planning_data synto_planning --config="planning_config.yaml"

Run training from scratch ^^^^^^^^^^^ .. code-block:: bash

synto_training_data synto_training --config="training_config.yaml"

Documentation -----------

The detailed documentation can be found here

Tests

synto_training --config="configs/training_config.yaml"
synto_planning --config="configs/planning_config.yaml"