metadata
language:
- en
metrics:
- accuracy
- AUC ROC
- precision
- recall
tags:
- biology
- chemistry
- therapeutic science
- drug design
- drug development
- therapeutics
library_name: tdc
license: bsd-2-clause
Dataset description
An integrated Ether-a-go-go-related gene (hERG) dataset consisting of molecular structures labeled as hERG (<10uM) and non-hERG (>=10uM) blockers in the form of SMILES strings was obtained from the DeepHIT, the BindingDB database, ChEMBL bioactivity database, and other literature.
Task description
Binary classification. Given a drug SMILES string, predict whether it blocks (1, <10uM) or not blocks (0, >=10uM).
Dataset statistics
Total: 13445; Train_val: 12620; Test: 825
Pre-requisites
Install the following packages
pip install PyTDC
pip install DeepPurpose
pip install git+https://github.com/bp-kelley/descriptastorus
pip install dgl torch torchvision
You can also reference the colab notebook here
Dataset split
Random split on 70% training, 10% validation, and 20% testing
To load the dataset in TDC, type
from tdc.single_pred import Tox
data = Tox(name = 'herg_karim')
Model description
Morgan chemical fingerprint with an MLP decoder. The model is tuned with 100 runs using the Ax platform.
To load the pre-trained model, type
from tdc import tdc_hf_interface
tdc_hf = tdc_hf_interface("hERG_Karim-Morgan")
# load deeppurpose model from this repo
dp_model = tdc_hf.load_deeppurpose('./data')
tdc_hf.predict_deeppurpose(dp_model, ['CC(=O)NC1=CC=C(O)C=C1'])
References
- Dataset entry in Therapeutics Data Commons, https://tdcommons.ai/single_pred_tasks/tox
- Karim, A., et al. CardioTox net: a robust predictor for hERG channel blockade based on deep learning meta-feature ensembles. J Cheminform 13, 60 (2021). https://doi.org/10.1186/s13321-021-00541-z