BayeshERG : A Bayesian Graph Neural Network for predicting hERG blockers

This repository is BayeshERG official repository. It contains the pytorch implementation of BayeshERG and trained model to predict arbitrary compounds. The implementation of BayehERG has referred to the official implementation of related studies [1-3].

The BayeshERG is developed with the python v3.6 and following packages:dgl, pytorch, and rdkit.

Prerequsites

• Anaconda

To avoid the package version issue, we open our code with Anaconda virtual environment. Therefore, the Anaconda should be installed in advance. https://www.anaconda.com/products/individual

Requirements

Input Format

Any .csv file with smiles column.

(Example)

ID	smiles

Usage

Create conda virtual environment

$ conda create -n BayeshERG -c conda-forge rdkit python=3.6

Activate the virtual environment

$ conda activate BayeshERG

Install dependencies

• If your system has GPU, check the CUDA version in advance (nvidia-smi).

Excute the installation shell script install.sh

$ sh install.sh

Then, type the cuda version to the shell and press enter.

$ sh install.sh
Input CUDA version of your GPU, ex. 10.2
: 10.2 (Enter)
DGL and Pytorch with CUDA v10.2 will be installed.
...
...

If your system has no GPU, excute the cpu-version shell script cpu_install.sh.

$ sh cpu_install.sh

Prediction

usage: $ python main.py [-i] input_csv_file_path 
                        [-o] output_file_name 
                        [-c] 'cpu' or 'gpu' (default 'cpu')
                        [-t] sampling time (integer, default 30)

• Example

// With GPU
$ python main.py -i data/External/EX1.csv -o EX1_pred -c gpu -t 30

// With CPU
$ python main.py -i data/External/EX1.csv -o EX1_pred -c cpu -t 30

Output Format

The prediction results (Prediction score, Uncertainties) are appened to the input .csv file and saved to prediction_results directory as output_file_name.csv.

ID	smiles	score	alea	epis

Also, the attention images(.svg) are also depicted and saved to attention_results/output_file_name directory.

Contact

Hyunho Kim, hyunhokim@gm.gist.ac.kr

Minsu Park, 15pms@gist.ac.kr

Hojung Nam (Corresponding Author), hjnam@gist.ac.kr

Reference

[1] Gal, Yarin, Jiri Hron, and Alex Kendall. "Concrete dropout." arXiv preprint arXiv:1705.07832 (2017).

[2] Scalia, Gabriele, et al. "Evaluating scalable uncertainty estimation methods for deep learning-based molecular property prediction." Journal of chemical information and modeling 60.6 (2020): 2697-2717.

[3] Yang, Kevin, et al. "Analyzing learned molecular representations for property prediction." Journal of chemical information and modeling 59.8 (2019): 3370-3388.

---

@ Last modified : 2022.02.07