This is the repository for ReBADD-SE, a multi-objective molecular optimization model that designs a molecular structures in the format of SELFIES. For more details, please refer to our paper.

• Latest update: 26 Jan 2024

conda env create -f environment.yml

• TASK1: ReBADD-SE for GSK3b, JNK3, QED, and SA (frag-level)
• TASK3: ReBADD-SE for BCL2, BCLXL, and BCLW (frag-level)
• TASK4: ReBADD-SE for BCL2, BCLXL, and BCLW (char-level)
• TASK7: SELFIES Collapse Analaysis between ReBADD-SE (frag-level) and GA+D

• (Important!) Before starting any TASK, please first run the scripts in the directory 'data/chembl' or 'data/zinc15'
• Read the training data
• Preprocess the data for model training

• Read the training data
• The generator learns the grammar rules of SELFIES

• (Important!) Please check first the 'ReBADD_config.py' in which a reward function have to be defined appropriately
• Load the pretrained generator

• Load the checkpoints stored during optimization
• Sample molecules for each checkpoint

• For each checkpoint, load the sampled molecules
• Evaluate their property scores

• Calculate metrics (e.g. success rate)
• Find the best checkpoint

If you have any further questions, please do not hesitate to let me know.

jonghwanc@hallym.ac.kr

@article{CHOI2023106721,
	title = {ReBADD-SE: Multi-objective molecular optimisation using SELFIES fragment and off-policy self-critical sequence training},
	journal = {Computers in Biology and Medicine},
	volume = {157},
	pages = {106721},
	year = {2023},
	issn = {0010-4825},
	doi = {https://doi.org/10.1016/j.compbiomed.2023.106721},
	url = {https://www.sciencedirect.com/science/article/pii/S0010482523001865},
	author = {Jonghwan Choi and Sangmin Seo and Seungyeon Choi and Shengmin Piao and Chihyun Park and Sung Jin Ryu and Byung Ju Kim and Sanghyun Park},
	keywords = {Drug discovery, De novo drug design, Multi-objective optimisation, SELFIES, Reinforcement learning}
}