lennylv/GraphPPepIS

Simultaneous prediction of interaction sites on both protein and peptide sides of complexes through multi-layer graph convolutional networks

GraphPPepIS

############################################################################

The source code, training and test datasets of paper 'Simultaneous prediction of interaction sites on both protein and peptide sides of complexes through multi-layer graph convolutional networks'

############################################################################

Install dependency:

• conda create -n ppis python==3.7.10 -y

• conda activate ppis

• pip install Bio

Prerequisites:

• python: 3.7.10

• CUDA: 10.1

• pytorch: 1.2.0

All relevant inputs：

Due to the limitation of Github, some inputs larger than 25MB are not uploaded. Please contact me directly at 20204227031@stu.suda.edu.cn.

############################################################################

Reproduce experimental results:

(settrain: Train4094, settest: Test169, settest2: Test53)

Test our model GraphPPepIS

Step 1: cd code

Step 2: python predict_graph.py

Test our model SeqPPepIS

Step 1: cd code

Step 2: python predict_seq.py

Train your own model GraphPPepIS

Step 1: cd code

Step 2: python train_graph.py --layers 8 --units 512 --epochs 300 --pw 0.8 --lw 0.1

Train your own model SeqPPepIS

Step 1: cd code

Step 2: python train_seq.py --layers 8 --units 512 --epochs 300 --pw 0.8 --lw 1.0