PInet

This repo is under major reconstruction, new version is coming soon!

Setup

run

pip install -e .

Train

run

python utils/train_richdbd2_fixed6mmgk.py -dataset [your dataset] \

for example here \

python utils/train_richdbd2_fixed6mmgk.py -dataset dbd

Load pretrained model

In python with torch imported

classifier.load_state_dict(torch.load('model/seg_model_protein_15.pth'))

Sample dataset folder

Dbd 5 dataset folder should follow dbd folder structure input data should be a n-by-5 matrix, with columns' order [x,y,z,electrostatic,hydrophobicity]
dbd data are now avalable at https://www.dropbox.com/sh/qqi9op061mfxbmo/AADibYuDdMF4n2bDS3uqiEVha?dl=0

EpiPred dataset folder should follow dbd folder structure input data should be a n-by-5 matrix, with columns' order [x,y,z,electrostatic,hydrophobicity]
EpiPred data are now avalable at https://www.dropbox.com/sh/wnrn65y4vfs2m84/AABSFL8IeWh_U7gQO2__ZFEHa?dl=0

Preprocess Helper

utils/pdb2wrlpymol2_pub.py
change pdb files folder in the script and for a given complex ABCD.pdb, split the ligand and receptor as ABCD_l.pdb and ABCD_r.pdb.

utils/pdb2pqrall_pub.py
compute apbs for all pdb files. Need to change path in script for your pdb2pqr and apbs binary exe file.

utils/transdata2.m
set your path to data folder which will be your train script --dataset input. It will create coordinate data and interface label file.

utils/RichFeatureApbsCon_pub.py
modify matlab output 3 dims coordinate feature data file to 5 dims coordinate+electrostatic+hydrophobicity feature.