README.md

SyntheMol-RL

Below are instruments for reproducing the results from the paper SyntheMol-RL: a flexible reinforcement learning framework for designing novel and synthesizable antibiotics, which uses the reinforcement learning ( RL) version of SyntheMol.

Data

The relevant data should be downloaded to SyntheMol/data. This can be done as follows:

DATA_PATH=$(python -c "import synthemol; from pathlib import Path; print(Path(synthemol.__path__[0]).parent)")/rl
mkdir $DATA_PATH
gdown "https://drive.google.com/drive/folders/1Yex2UBjjmNFMZjBXA-ToVYMkpDFOorr0?usp=drive_link" --folder -O $DATA_PATH

Note that here, we use the 2022 q1-2 version of the building blocks and the 2022 q1-2 version of the enumerated REAL Space molecules.

Instructions

real.md: Instructions for processing Enamine REAL building blocks, reactions, and molecules.

wuxi.md: Instructions for processing the WuXi building blocks, reactions, and molecules (May 2023 release).

antibiotics.md: Instructions for generating antibiotic candidates for Staphylococcus aureus using reinforcement learning (RL). Includes instructions for processing antibiotics data, training antibacterial activity prediction models, generating molecules with SyntheMol-RL, and selecting candidates.

ablations.md: Instructions for reproducing the ablation experiments in the paper to determine the importance of various components of SyntheMol-RL.

gflownet.md: Instructions for reproducing the GFlowNet experiments in the paper.