Molecular Property Prediction based on Bimodal Supervised Contrastive Learning

Install

To install the required packages, follow these instructions (tested on a linux terminal):

1- clone the repository

git clone https://github.com/syan1992/BSCL

2- cd into the cloned directory

cd BSCL

3- run the install script

pip install -r requirements.txt

We run the code on GPU.

Datasets

Please find the 'datasets' folder for the example of the data. The data should be split into train/validation/test subsets at first.

Usage example

We list all command lines in the shell script 'autorun.sh' for the seven datasets (freesolv, delaney, lipophilicity, bace, sider, tox21, clintox) we test in our experiments. Run 'autorun.sh' with the name of the dataset as a parameter.

./autorun.sh freesolv

We save the model with the best performance on the validation set and evaluate the best model with the test set. Both model and test results will be saved in the 'save' folder.

Hyperparameters

Some specific hyperparameters in this work,

Name	Description
wscl	The weight of the supervised contrastive loss in the loss function. Suggest to test values in [0.1 to 1]
wrecon	The weight of the reconstruction loss in the loss function. Suggest to test values in [0.1 to 1]
gamma1	The hyperparameter of the weighted supervised contrastive loss for the regression task. Suggest to test values in [2,3,4]
gamma2	The hyperparameter of the weighted supervised contrastive loss for the regression task. Suggest to test values in [1,2,3]

Acknowledgement

Supervised contrastive learning : https://github.com/HobbitLong/SupContrast
Deepgcn : https://github.com/lightaime/deep_gcns_torch

Reference

@inproceedings{sun2022molecular,
  title={Molecular Property Prediction based on Bimodal Supervised Contrastive Learning},
  author={Sun, Yan and Islam, Mohaiminul and Zahedi, Ehsan and Kuenemann, M{\'e}laine and Chouaib, Hassan and Hu, Pingzhao},
  booktitle={2022 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)},
  pages={394--397},
  year={2022},
  organization={IEEE}
}