PrefixMol: Target- and Chemistry-aware Molecule Design via Prefix Embedding

📢 News

• Code: https://github.com/A4Bio/PrefixMol
• Paper: https://arxiv.org/abs/2302.07120

We propose PrefixMol, inserting learnable conditional feature vectors into the attention module to unify multi-conditional molecule generative models to support the modeling of customized requirements.

Installation

Dependency

The codes have been tested in the following environment.

Package	Version
Python	3.7.12
PyTorch	1.10.0
CUDA	11.3.1
PyTorch Geometric	2.0.3
RDKit	2021.09.4

Install via conda yaml file (cuda 11.3)

conda env create -f env.yml
conda activate PrefixMol

Install manually

conda create -n PrefixMol python=3.7.12
conda activate PrefixMol

# Install PyTorch (for cuda 11.3)
conda install pytorch==1.10.0 cudatoolkit=11.3 -c pytorch -c conda-forge
# Install PyTorch Geometric (>=2.0.0)
conda install pyg -c pyg

# Install other tools 
# conda install
conda install -c conda-forge rdkit
conda install -c conda-forge openbabel
conda install pyyaml easydict python-lmdb -c conda-forge

# pip install
pip install nni
pip install dill
pip install deepchem
pip install jax
pip install jaxlib
pip install tensorflow
pip install transformers
pip install partialsmiles
pip install pytorch-lightning

🔎 Tips for pytorch3d installation

Notice that we recommend using the following steps to install pytorch3d 👐

1. install the following necessary packages.

conda install -c fvcore -c iopath -c conda-forge fvcore iopath

2. Find the suitable version with your environment
3. Git clone the resporitory and then run the command as follows for example.

cd pytorch3d
python setup.py install

Datasets

Please refer to README.md in the data folder.

Training

We used DDP to accelerate the training process. Here are some command examples FYR.

# 4 GPUs
CUDA_VISIBLE_DEVICES="0,1,2,3" python -m torch.distributed.launch --nproc_per_node 4 train.py
# 8 GPUs
CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7" python -m torch.distributed.launch --nproc_per_node 8 train.py

Testing

When it comes to testing process, we loaded the checkpoint.pth and used 1 GPU to test the result.

CUDA_VISIBLE_DEVICES="0" python -m torch.distributed.launch --nproc_per_node 1 test.py

🔖 Tips

When running the codes, the path where the code appears is recommended to be changed to the path you need at the moment.

Citation

@article{gao2023prefixmol,
  title={PrefixMol: Target-and Chemistry-aware Molecule Design via Prefix Embedding},
  author={Gao, Zhangyang and Hu, Yuqi and Tan, Cheng and Li, Stan Z},
  journal={arXiv preprint arXiv:2302.07120},
  year={2023}
}

Contact

Zhangyang Gao ([email protected]) Yuqi Hu ([email protected])