MORetro

MORetro is a Multi-Objective Retrosynthesis planning tool.

Installation

Create the conda environment using the provided environment.yml file:

conda env create -f environment.yml
conda activate moretro

Usage

The main entry point for running the search is the moretro.moretro_star module.

python -m moretro.moretro_star --dataset <path_to_dataset> --output_dir <output_directory> --config_file <config_filename>

Arguments:

• --dataset: Path to the input file containing target SMILES (CSV format).
• --output_dir: Directory where results and logs will be saved. Default is output.
• --config_file: Name of the gin config file to use (located in moretro/configs/). Default is search_config.gin.

Configuration

The main configuration file is located at moretro/configs/search_config.gin. You can modify various parameters to customize the search behavior.

Single-Step Predictor

You can change the single-step retrosynthesis predictor by modifying the retro_model path and the OneStepModel.model_type.

In moretro/configs/search_config.gin:

# Checkpoint files
retro_model = "models/model_retro.pt"  # Path to your model checkpoint

# ...

# Retro model config
retro_prediction.OneStepModel.model_type = "st" # "st" for single-step, etc.
retro_prediction.OneStepModel.checkpoint_path = %retro_model

Make sure OneStepModel.model_type matches the type of model you are using (e.g., "st" for the standard model).

Sampling Strategy

You can choose between Bayesian Optimization (BO) or a simple Queue-based strategy for weight sampling during the multi-objective search.

Modify MOGraph.weight_update_strategy in moretro/configs/search_config.gin:

# Weight Selection Parameters
# ...
MOGraph.weight_update_strategy = "bo"  # Options: "bo" or "queue"

• "bo": Uses Bayesian Optimization to select weight vectors.
• "queue": Uses a standard queue for weight processing.