yuyangw / MolCLR

Implementation of MolCLR: "Molecular Contrastive Learning of Representations via Graph Neural Networks" in PyG.

Molecular Contrastive Learning of Representations via Graph Neural Networks

Nature Machine Intelligence [Paper] [arXiv] [PDF]

Yuyang Wang, Jianren Wang, Zhonglin Cao, Amir Barati Farimani
Carnegie Mellon University

This is the official implementation of MolCLR: "Molecular Contrastive Learning of Representations via Graph Neural Networks". In this work, we introduce a contrastive learning framework for molecular representation learning on large unlabelled dataset (~10M unique molecules). MolCLR pre-training greatly boosts the performance of GNN models on various downstream molecular property prediction benchmarks. If you find our work useful in your research, please cite:

@article{wang2022molclr,
  title={Molecular contrastive learning of representations via graph neural networks},
  author={Wang, Yuyang and Wang, Jianren and Cao, Zhonglin and Barati Farimani, Amir},
  journal={Nature Machine Intelligence},
  pages={1--9},
  year={2022},
  publisher={Nature Publishing Group},
  doi={10.1038/s42256-022-00447-x}
}

Getting Started

Installation

Set up conda environment and clone the github repo

# create a new environment
$ conda create --name molclr python=3.7
$ conda activate molclr

# install requirements
$ pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 -f https://download.pytorch.org/whl/torch_stable.html
$ pip install torch-geometric==1.6.3 torch-sparse==0.6.9 torch-scatter==2.0.6 -f https://pytorch-geometric.com/whl/torch-1.7.0+cu110.html
$ pip install PyYAML
$ conda install -c conda-forge rdkit=2020.09.1.0
$ conda install -c conda-forge tensorboard
$ conda install -c conda-forge nvidia-apex # optional

# clone the source code of MolCLR
$ git clone https://github.com/yuyangw/MolCLR.git
$ cd MolCLR

Dataset

You can download the pre-training data and benchmarks used in the paper here and extract the zip file under ./data folder. The data for pre-training can be found in pubchem-10m-clean.txt. All the databases for fine-tuning are saved in the folder under the benchmark name. You can also find the benchmarks from MoleculeNet.

Pre-training

To train the MolCLR, where the configurations and detailed explaination for each variable can be found in config.yaml

$ python molclr.py

To monitor the training via tensorboard, run tensorboard --logdir ckpt/{PATH} and click the URL http://127.0.0.1:6006/.

Fine-tuning

To fine-tune the MolCLR pre-trained model on downstream molecular benchmarks, where the configurations and detailed explaination for each variable can be found in config_finetune.yaml

$ python finetune.py

Pre-trained models

We also provide pre-trained GCN and GIN models, which can be found in ckpt/pretrained_gin and ckpt/pretrained_gcn respectively.

Acknowledgement

• PyTorch implementation of SimCLR: https://github.com/sthalles/SimCLR
• Strategies for Pre-training Graph Neural Networks: https://github.com/snap-stanford/pretrain-gnns