Syntheseus retro star benchmark

A wrapper for using the benchmark from retro* (Chen et al 2020) in syntheseus.

Usage:

from syntheseus_retro_star_benchmark import RetroStarReactionModel
model = RetroStarReactionModel()  # a syntheseus BackwardReactionModel object wrapping the pre-trained template classifier

from syntheseus_retro_star_benchmark import RetroStarInventory
inventory = RetroStarInventory()  # their inventory of ~23M purchasable molecules

from syntheseus_retro_star_benchmark import get_190_hard_test_smiles
test_smiles = get_190_hard_test_smiles()  # their recommended 190 test SMILES

from syntheseus_retro_star_benchmark import RetroStarValueMLP
value_fn = RetroStarValueMLP()  # their pre-trained search heuristic

Code was based on open source code from here. Some data was uploaded to figshare to ensure stable, consistent access.

Installation

Install either by cloning and using pip or running pip install syntheseus-retro-star-benchmark.

Development

Ensure to install all pre-commit hooks and run unit tests (provided by pytest).