This is the repository for our paper "Molecule Design by Latent Space Energy-based Modeling and Gradual Distribution Shifting" in UAI 2023. PDF

In this paper, we studied the following property optimization tasks:

• single-objective p-logP maximization
• single-objective QED maximization
• single-objective ESR1 binding affinity maximization
• single-objective ACAA1 binding affinity maximization
• multi-objective (ESR1, QED, SA) optmization
• multi-objective (ACAA1, QED, SA) optmization

We follow the previous work LIMO for setting up RDKit, Open Babel and AutoDock-GPU. We extend our gratitude to the authors for their significant contributions.

We use selfies representations of ZINC250k with corresponding property values. All the property values can be computed either by RDKit or AutoDock-GPU.

For model training given certain property (i.e. ESR1),

cd single_design_esr1
python main.py

For property optimizaton task,

python single_design.py or multi_design.py

@inproceedings{kong2023molecule,
  title={Molecule Design by Latent Space Energy-Based Modeling and Gradual Distribution Shifting},
  author={Kong, Deqian and Pang, Bo and Han, Tian and Wu, Ying Nian},
  booktitle={Uncertainty in Artificial Intelligence},
  pages={1109--1120},
  year={2023},
  organization={PMLR}
}