Intro

This is a Pytorch implementation of the NeurIPS paper “Understanding the Limitations of Deep Models for Molecular Property Prediction: Insights and Solutions”

Installation

We used the following Python packages for core development. We tested on Python 3.7.4.

pytorch                   1.13.1+cu117
scikit-learn              0.20.1
XGBoost                   1.6.2
DGL                       0.4.2
RDKit                     2022.09.5
hyperopt                  0.1.2

Dataset download

All the necessary data files can be downloaded from the following link Google Drive, which include the molecules' SMILES strings, labels and various fingerprints. The SMILES vocabulary files can be found in Github

Create the Directory for the results

mkdir stat_res

How to run this code

To run this code, for xgboost you can use

python xgb.py --data_label tox21 --runseed 43 --patience 50 --opt_iters 50 --repetitions 50 --num_pools 5

or for rf

python rf.py --data_label tox21 --runseed 43 --patience 50 --opt_iters 50 --repetitions 50 --num_pools 5

or for svm

python svm.py --data_label tox21 --runseed 43 --patience 50 --opt_iters 50 --repetitions 50 --num_pools 28

or for gcn

python gnn.py --data_label tox21 --epochs 300 --runseed 43 --batch_size 128 --patience 50 --opt_iters 50 --repetitions 50 --model_name gcn

or for mlp

python ifm_mlp.py --data_label tox21 --embed None --epochs 300 --runseed 43 --batch_size 128 --patience 50 --opt_iters 50 --repetitions 50

or for IFM-MLP

python ifm_mlp.py --data_label tox21 --embed IFM --epochs 300 --runseed 43 --batch_size 128 --patience 50 --opt_iters 50 --repetitions 50

Citation

@inproceedings{
xia2023understanding,
title={Understanding the Limitations of Deep Models for Molecular property prediction: Insights and Solutions},
author={Jun Xia and Lecheng Zhang and Xiao Zhu and Yue Liu and Zhangyang Gao and Bozhen Hu and Cheng Tan and Jiangbin Zheng and Siyuan Li and Stan Z. Li},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023}
}

Acknowledgment

[1] SMILES Transformer: Pre-trained Molecular Fingerprint for Low Data Drug Discovery (Honda et al., Arixv 2019)
[2] Exposing the limitations of molecular machine learning with activity cliffs (Tilborg et al., JCIM 2022)