[arXiv] Efficient Evolutionary Search Over Chemical Space with Large Language Models
[Website] MolLEO Project
MolLEO is an LLM-augmented evlotuionary algorithm for molecular discovery!
You need to get an OpenAI API key for GPT-4. BioT5 is an open-source language model which can work on either GPU or CPU. Currently, this code repo does not support MolLEO(MOLSTM), but we will update soon.
conda create -n molleo python=3.9
conda activate molleo
conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=12.1 -c pytorch -c nvidia
pip install PyTDC
pip install PyYAML
pip install rdkit
pip install transformers
pip install sentencepiece
pip install selfiesThen we can activate conda via following command.
conda activate molleo The experiments are conducted on the following categories: single obejective optimization and multi objective optimization.
To run experiments on single objective optimization task:
cd single_objective
# BioT5 on jnk3 task
python run.py molleo --mol_lm BioT5 --oracles jnk3 --seed 1 2 3
# GPT-4 on gsk3b task
python run.py molleo --mol_lm GPT-4 --oracles gsk3b --seed 1 2 3To run experiments on multi objective optimization task:
cd multi_objective
# objective summation on task 1
python run.py molleo_multi --mol_lm BioT5 --min_obj sa --max_obj jnk3 qed --seed 1 2 3
# pareto optimal set selection on task 2
python run.py molleo_multi_pareto --mol_lm GPT-4 --min_obj sa --max_obj gsk3b qed --seed 1 2 3
# pareto optimal set selection on task 3
python run.py molleo_multi_pareto --mol_lm GPT-4 --min_obj sa gsk3b drd2 --max_obj jnk3 qed --seed 1 2 3If you find our work helpful, please consider citing our paper:
@misc{wang2024efficientevolutionarysearchchemical,
title={Efficient Evolutionary Search Over Chemical Space with Large Language Models},
author={Haorui Wang and Marta Skreta and Cher-Tian Ser and Wenhao Gao and Lingkai Kong and Felix Streith-Kalthoff and Chenru Duan and Yuchen Zhuang and Yue Yu and Yanqiao Zhu and Yuanqi Du and Alán Aspuru-Guzik and Kirill Neklyudov and Chao Zhang},
year={2024},
eprint={2406.16976},
archivePrefix={arXiv},
primaryClass={cs.NE}
url={https://arxiv.org/abs/2406.16976},
}