hassanmohsin / deep-bs

deep binding site

DEEP BINDING SITE

KDEEP Model

To train:

python train.py --dataroot /home/sunhwan/work/pdbbind/2018/refined-set \
                --csvfile /home/sunhwan/work/pdbbind/deep/data/train.csv \
                --gpu_ids 0 --batch_size 64 --nThreads 16 --init_type kaiming \
                --lr 0.0001 --niter 50 --niter_decay 25 --save_epoch_freq 5 \
                --model kdeep --grid_method kdeep --grid_size 24 --grid_spacing 1.0 \
                --channels kdeep --rvdw 2

add --continue-train to resume training from the latest weight.

To test:

python test.py --dataroot /home/sunhwan/work/pdbbind/2018/refined-set \
               --csvfile /home/sunhwan/work/pdbbind/deep/data/test.csv \
               --gpu_ids 0 --batch_size 64 --nThreads 16 \
               --model kdeep --grid_method kdeep --grid_size 24 --grid_spacing 1.0 \
               --channels kdeep --rvdw 2

GNINA Model

To train:

python train.py --dataroot /home/sunhwan/work/pdbbind/2018/refined-set \
                --csvfile /home/sunhwan/work/pdbbind/deep/data/train.csv \
                --gpu_ids 0 --batch_size 32 --nThreads 16 --init_type kaiming \
                --lr 0.0001 --niter 50 --niter_decay 25 --save_epoch_freq 5 \
                --model gnina --grid_method gnina --grid_size 48 --grid_spacing 0.5 \
                --channels gnina

add --continue-train to resume training from the latest weight.

To test:

python test.py --dataroot /home/sunhwan/work/pdbbind/2018/refined-set \
               --csvfile /home/sunhwan/work/pdbbind/deep/data/test.csv \
               --gpu_ids 0 --batch_size 32 --nThreads 16 \
               --model gnina --grid_method gnina --grid_size 48 --grid_spacing 0.5 \
               --channels gnina

GNINA with docked pose Model

To train:

python train.py --dataroot /home/sunhwan/work/pdbbind/2018/refined-set \
                --csvfile /home/sunhwan/work/pdbbind/deep/data/train.csv \
                --gpu_ids 0 --batch_size 32 --nThreads 16 --init_type kaiming \
                --lr 0.0001 --niter 50 --niter_decay 25 --save_epoch_freq 5 \
                --model gnina_docked --grid_method gnina_docked --grid_size 48 \
                --grid_spacing 0.5 --channels gnina --dataset_mode pdbbind_docked

add --continue-train to resume training from the latest weight.

To test:

python test.py --dataroot /home/sunhwan/work/pdbbind/2018/refined-set \
               --csvfile /home/sunhwan/work/pdbbind/deep/data/test.csv \
               --gpu_ids 0 --batch_size 32 --nThreads 16 \
               --model gnina --grid_method gnina --grid_size 48 --grid_spacing 0.5 \
               --channels gnina

Preprocess PdbBind dataset

To make reading data faster, use the following command to preprocess PDB/Mol2 files and determine Smina atom types prior to training.

python data.py --dataroot /home/sunhwan/work/pdbbind/2018/refined-set
python data.py --dataroot /home/sunhwan/work/pdbbind/2018/other-set