Official implementation of MVMRL model in the paper: Self-Supervised Molecular Representation Learning With Topology and Geometry (IEEE Journal of Biomedical and Health Informatics 2024) [MVMRL]

conda env create -f environment.yaml

Download the pre-training dataset from [PCQM4Mv2] and place it in the "./Data/pcqm4mv2/pcqm4m-v2-train.sdf" directory.

wget http://ogb-data.stanford.edu/data/lsc/pcqm4m-v2-train.sdf.tar.gz
md5sum pcqm4m-v2-train.sdf.tar.gz # fd72bce606e7ddf36c2a832badeec6ab
tar -xf pcqm4m-v2-train.sdf.tar.gz # extracted pcqm4m-v2-train.sdf

You can pretrain the model by

python trainer_pretrain.py

Download the downstream dataset from [MoleculeNet]. Place the raw data in CSV format in the ./Data/**dataset_name directory and generate 3D coordinates using the "python get_lmdb_data.py" command, for example, generating a file named "./Data/bace/bace.pkl".

You can finetune the model for classification tasks by

python trainer_prediction.py

Please kindly cite this paper as follows. Thank you.

@article{zang2024self,
  title={Self-supervised molecular representation learning with topology and geometry},
  author={Zang, Xuan and Zhang, Junjie and Tang, Buzhou},
  journal={IEEE Journal of Biomedical and Health Informatics},
  volume={29},
  number={1},
  pages={700--710},
  year={2024},
  publisher={IEEE}
}