This repository provided implemented codes for the paper -- Genetic GFlowNets: Advancing in Practical Molecular Optimization Benchmark.
The codes are implemented our code based on the practical molecular optimization benchmark.
In addition, we implemented Mol GA and GEGL by adopting the original implementation; see the papers and codes below.
- Sample Efficiency Matters: A Benchmark for Practical Molecular Optimization (NeurIPS, 2022) (code: https://github.com/wenhao-gao/mol_opt)
- Genetic algorithms are strong baselines for molecule generation (code: https://github.com/AustinT/mol_ga)
- Guiding Deep Molecular Optimization with Genetic Exploration (NeurIPS, 2020) (code: https://github.com/sungsoo-ahn/genetic-expert-guided-learning)
Clone project and create environment with conda:
conda create -n genetic python==3.7
conda activate genetic
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch
conda install -c rdkit rdkit
pip install -r requirements.txt
Note: we highly recommend using Python 3.7, and PyTorch 1.12.1. Importantly, we use PyTDC 0.4.0 instead of 0.3.6.
For other baseline, please see README_PMO.md; we recommend using Python 3.8 for GP BO and requirement_gpbo.txt.
When you use Python 3.8, you can install rdkit with pip install rdkit-pypi.
CUDA_VISIBLE_DEVICES=0 python run.py genetic_gfn --task simple --wandb online --oracle qed --seed 0
To run genetic_GFN with SELFIES, use genetic_gfn_selfies instead of genetic_gfn.