Accurate ADMET Prediction with XGBoost

If you find it useful, please cite: Tian, H., Ketkar, R. & Tao, P. ADMETboost: a web server for accurate ADMET prediction. J Mol Model 28, 408 (2022). https://doi.org/10.1007/s00894-022-05373-8

Installation

git clone https://github.com/smu-tao-group/ADMET_XGBoost
cd ADMET_XGBoost
conda env create -f environment.yml
conda activate tdc

Usage

1. Featurization: run python featurize.py TASK_NAME to convert SMILES to features. This step is time consuming and we provide the processed data that can be downloaded here.
2. Modeling: run python model.py TASK_NAME for model training and prediction.

Results

Tasks	Evaluation	Performance	Rank
Absorption
Caco2	MAE	0.288 ± 0.011	1st
HIA	AUROC	0.987 ± 0.002	1st
Pgp	AUROC	0.911 ± 0.002	5th
Bioav	AUROC	0.700 ± 0.010	2nd
Lipo	MAE	0.533 ± 0.005	1st
AqSol	MAE	0.727 ± 0.004	1st
Distribution
BBB	AUROC	0.905 ± 0.001	1st
PPBR	MAE	8.251 ± 0.115	1st
VDss	Spearman	0.612 ± 0.018	1st
Metabolism
CYP2C9 Inhibition	AUPRC	0.794 ± 0.004	3rd
CYP2D6 Inhibition	AUPRC	0.721 ± 0.003	2nd
CYP3A4 Inhibition	AUPRC	0.877 ± 0.002	3rd
CYP2C9 Substrate	AUPRC	0.387 ± 0.018	3rd
CYP2D6 Substrate	AUPRC	0.648 ± 0.023	5th
CYP3A4 Substrate	AUPRC	0.680 ± 0.005	1st
Excretion
Half Life	Spearman	0.396 ± 0.027	1st
CL-Hepa	Spearman	0.420 ± 0.011	2nd
CL-Micro	Spearman	0.587 ± 0.006	2nd
Toxicity
LD50	MAE	0.602 ± 0.006	2nd
hERG	AUROC	0.806 ± 0.005	4th
Ames	AUROC	0.859 ± 0.002	1st
DILI	AUROC	0.933 ± 0.011	1st