A Spatial-temporal Gated Attention Module for Molecular Property Prediction Based on Molecular Geometry. This is the official code implementation of Drug3D-Net paper. But the algorithm has been optimized and improved, which is slightly different from the original version.
Requirements
Linux (We only tested on Ubuntu-16.04)
Keras (version == 2.3.1)
Python (version == 3.6.8)
tensorflow (version == 1.15.0)
matplotlib (version == 3.3.2)
numpy (version == 1.19.4)
scikit-learn (version == 0.23.2)
scipy (version == 1.5.4)
pandas (version ==1.1.5)
DataSets
esol_v2.csv
FreeSolv_SAMPL.csv
HIV.csv
tox21.csv
PreprocessData
grid3Dmols_delaney
grid3Dmols_freesolv
grid3Dmols_hiv
tox21grid3Dmols_tox21_NR_AhR
grid3Dmols_tox21_NR-AR
grid3Dmols_tox21_NR-AR-LBD
grid3Dmols_tox21_NR-Aromatase
grid3Dmols_tox21_NR-ER
grid3Dmols_tox21_NR-ER-LBD
grid3Dmols_tox21_NR-PPAR-gamma
grid3Dmols_tox21_SR-ARE
grid3Dmols_tox21_SR-ATAD5
grid3Dmols_tox21_SR-HSE
grid3Dmols_tox21_SR-MMP
grid3Dmols_tox21_SR-p53
SaveModel (Take the HIV dataset as an example)
6-0.979.hdf5
best_weight.hdf5
events.out.tfevents.1608729525.lab406
hyper.csv
raw_results.csv