RTS-CESP

This is the repository for the manuscript “Chemoenzymatic Synthesis Planning Guided by Reaction Type Score”.

Reaction Type score (RTscore)

The Reaction Type score (RTscore) distinguishes synthesis reactions from decomposition reactions, and we trained two RTscore models using the USPTO and ECREACT databases separately for chemical and biological reactions.

Dependencies

• RDKit
• PyTorch
• NumPy

Model

The standalone model for chemical reactions is defined at:
RTscore/RTscore_chem/RTscore_chem_API.py

The standalone model for biological reactions is defined at:
RTscore/RTscore_bio/RTscore_bio_API.py

Training

To train with your own database, use the code available in the data_processing directory.

Chemoenzymatic synthesis planning using ML-based methods

Install the package:

pip install rxn4chemistry

Run your target molecules through the following script:

model_eval/10_examples/runmodel_IBM.py

For additional information about rxn4chemistry, see rxn4chemistry Github.

The 10 examples to evaluate RTscore are available at:
model_eval/10_examples/results

Chemoenzymatic synthesis planning using rule-based methods

1. Create a Conda environment and download Aizythfinder checkpoints according to Aizythfinder Github.

2. Clone Aizythfinder repository:

   git clone https://github.com/MolecularAI/aizynthfinder.git model_eval/1000_molecules

3. Run your target molecules using the following script:

   model_eval/1000_molecules/runmodel_aiz_retrobiocat.py

The validation results on 1000 molecules are available at:
model_eval/1000_molecules/1000molecules_results.json