Accurate ADMET Prediction with XGBoost
git clone https://github.com/smu-tao-group/ADMET_XGBoost
cd ADMET_XGBoost
conda env create -f environment.yml
conda activate tdc
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 .
Modeling: run python model.py TASK_NAME
for model training and prediction.
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